This is a #paidpartnership with Commonwealth Bank.

The hardest part of enterprise AI was never making the model smarter.

Most teams are still trying to win a race that’s already been decided. They wait for the next model, whether that’s a higher benchmark, a longer context window, a cheaper token, as if the distance between a promising demo and a system customers depend on were a capability gap. It isn’t. It’s a trust gap. And there’s a number that proves it.

On our latest episode of What The Tech, we sat down with Beibei Guo, Distinguished Engineer at CommBank, the bank that just ranked #4 in the world and #1 in Asia-Pacific in the 2025 Evident AI Index. Twenty-plus years across systems engineering and data platforms. She’s watched the internet, virtualisation, and cloud all arrive with the same promise and the same trap. Her read on AI is unsentimental: the model getting smarter was never the bottleneck.

We’re not the only ones finding this. MIT’s NANDA initiative, in The GenAI Divide: State of AI in Business 2025, reported that 95% of enterprise generative-AI pilots fail to move the P&L and the cause isn’t model quality. It’s what MIT calls the “learning gap”: the failure to wire AI into the workflows, structures, and habits of the organisation. The number that should frighten you sits next to it: the share of companies abandoning most of their AI initiatives jumped from 17% to 42% in a single year.

The conventional wisdom

The default mental model says AI value is a function of model intelligence. Pick the smartest model, point it at the problem, wait for it to be “good enough.” Under that logic, every failed pilot is a model problem and every model problem gets solved by the next release.

It’s a comfortable story, because it asks nothing of the organisation. You just wait.

Why it’s incomplete

Beibei has a sharper frame. AI without trust, she says, is like a talking dog: a delightful novelty you don’t quite know what to do with. Impressive, briefly. Useless in production. What turns the talking dog into something you’d deploy, her words is trust: “with trust, it’s like Mr. Data in Star Trek.” Still a little strange, but dependable enough to bring along on the journey.

That is exactly what MIT found, stated in the language of P&L. The 95% don’t fail because the model isn’t clever enough. They fail because the cleverness never became something anyone could rely on. Intelligence was abundant. Trust was missing.

What trust actually looks like

Trust isn’t a feeling you wait to develop. It’s a property you engineer into the workflow. Three moves from the conversation:

Put the gate in the workflow, not the prompt. CommBank is an early adopter of the AWS DevOps Agent, announced at re:Invent 2025. When an incident fires, it correlates logs, telemetry, code, and deployment data and hands the on-call engineer a root-cause hypothesis, work that used to mean 40 minutes of dashboard-spelunking before anyone could even think. What it does not do is fix anything. As Beibei put it, she’s glad it does what she told it to do and not go ahead and fix it. The human approves the action. “If it gives you a step that drops your production database,” she said, “it relies on you to tell it not to.” Accountability never leaves the person.

Shape what you can’t define. You can’t specify exactly what a model will output. You can shape it and you have to understand the shape well enough to know where it’s safe to point. That understanding is the work. It doesn’t arrive in a release note.

Keep a human on the decision that carries weight. Beibei described a small production change like adjusting a system parameter with real downstream consequences. She used Claude to draft the test cases and run them; she made the call and stayed accountable for it. The AI took the repeatable work. The judgment stayed with her, on purpose. “I’d like to be accountable in that way,” she said. “Selectively.”

What doesn’t get easier

Here’s the hard edge. When a model can do more, the few things it can’t do matter more, not less. Accountability doesn’t compress. Judgment doesn’t compress.

The MIT data has a tell buried in it: pilots that paired internal specialists with outside expertise hit a 67% success rate; IT-only builds managed 22%. The differentiator wasn’t a smarter model. It was people who owned the outcome.

Beibei’s version is human-scale. Her graduate engineers shipped production-grade data-comparison systems across hundreds of terabytes within months, not because she handed them a prompt, but because she handed them “a few grads, a few models, and a principal engineer” who met them weekly and owned the result.

The delivery

So here’s the whole thing, paid off: the smartest model on the leaderboard will not move you from the 95% to the 5%. Trust will. The teams crossing the divide aren’t the ones who waited for a better model, they’re the ones who built the gate into the workflow, kept a human on the consequential call, and treated accountability as the part that never gets automated.

Your AI doesn’t have a smartness problem. It has a trust problem and unlike the model, that one is yours to fix.

Listen to the full episode

In our conversation with Beibei Guo, we get into:

• Why “demo is easy, and nothing is easy inside the corporate world”

• The talking-dog-vs-Mr-Data test for any AI system

• How CommBank lets agents do the heavy lifting while accountability stays human

• Why the engineers who win pair models with mentorship, not prompts

• The trade-off every leader is now navigating: the risk of adopting vs. the risk of not adopting

🎧 Listen on Spotify · Apple Podcasts · Amazon Music · YouTube

📬 Subscribe on Substack for episode breakdowns and behind-the-scenes thinking.

Question for readers

If you pulled the smartest model out of your stack tomorrow and dropped in the second-smartest, what would actually break? If the honest answer is “not much,” your bottleneck was never intelligence.

Sources:

• MIT NANDA, “The GenAI Divide: State of AI in Business 2025”;

• AWS, “From AI agent prototype to product: Lessons from building AWS DevOps Agent” (re:Invent 2025);

• Evident AI Index 2025.

Episode quotes from What The Tech (AU), our conversation with Beibei Guo.

This is a #paidpartnership with CommBank.

When people picture AI, they picture a chatbot, a text box you converse with. But the AI that has touched the most Australian lives has no text box, no personality, and no launch event. It runs underneath things you already do, and the better it works, the less you notice it. On our latest episode, Beibei Guo, Distinguished Engineer at CommBank, the bank that ranks #1 in Asia-Pacific for AI maturity by the Evident AI Index - kept returning to that point. We took her examples and checked each one against the public record. They hold up.

The unexpected finding: the most valuable AI in this country may be the kind you don’t notice at all.

The scams you never got

AI is now an important tool in the fight against scams and fraud with CommBank’s market-first safety tools NameCheck, CallerCheck, and CustomerCheck, contributing to significant reductions in financial crime and mistaken payments. By the first half of FY26, fraud losses were down more than 20% year-on-year. You don’t feel a scam that didn’t happen. That’s the point.

Then the part that sounds like fiction. Macquarie University spin-out and cyber-intelligence firm Apate.ai is collaborating with CommBank, deploying thousands of AI “honeypot” bots, engineered to engage and disrupt scammers. Complete with colloquial Australian phrasing, the bots answer scammers over phone and text, waste their time, and gather live intelligence. Every minute a scammer spends talking to a bot is a minute they’re not targeting a real person. Most Australians will never know a bot fought on their behalf.

$1 billion, four questions

CommBank’s Benefits Finder launched in September 2019 in the CommBank App. It’s a machine-learning tool that asks a handful of questions and matches you to government grants, rebates, and concessions you’re entitled to but probably never claimed. Quiet. Unglamorous. It has connected customers to over $1 billion. No chatbot. Just an AI-powered solution that found money people were eligible for and never chased.

Why invisible is the goal, not an accident

Notice what none of these looks like. None is a chatbot. None announced itself. Beibei named the ambition directly: the best AI is invisible - woven into the fabric so completely that you experience the outcome, not the technology.

The bank doesn’t want you to admire its data quality. It wants you to feel understood: to never get asked why you’re topping up a home loan when your own business account, at the same bank, already explains it. Trust accumulates from consistency you never have to think about, the bank that quietly knows the difference between your personal context and your professional one, and treats you with care, as the same individual.

The delivery: the AI that matters most in Australia right now isn’t the one with the cleverest demo. It’s the one you’ve already used this week without realising, the billion-dollar tool that doesn’t look like AI at all. And in 2026, that isn’t a sign the technology is behind. It’s the clearest sign it’s really working.

Listen to the full episode

In our conversation with Beibei Guo, we get into:

• The everyday Australian AI you’ve used without recognising it

• Why the best AI is invisible and why that’s ambitious

• How good data foundations quietly turn into customer trust

• The machine-learning era everyone forgot when generative AI arrived

• What “delivering real-world impact” actually looks like at national scale

🎧 Listen on Spotify · Apple Podcasts · Amazon Music · YouTube

📬 Subscribe on Substack for episode breakdowns and behind-the-scenes thinking.

Question for readers

Which invisible AI touched your week and would you have recognised it as AI if no one had told you?

Sources:

• CommBank Newsroom (Benefits Finder, anti-scam program, Apate.ai partnership, 2026 AI report);

• Microsoft Customer Stories (Telstra One Sentence Summary; Benefits Finder figure);

• Premier of South Australia / SAPOL (mobile phone detection cameras);

• Evident AI Index 2025.

Episode references from What The Tech (AU), our conversation with Beibei Guo.

This is a #paidpartnership with Commonwealth Bank.

Inside Commonwealth Bank, engineers can access a sanctioned AI key through a single internal page.

Beibei Guo will tell you that was the easy part.

The demo is never the hard part. The hard part is everything after: the testing, the data, the judgment calls and the unglamorous engineering that turns a hackathon win into something millions of customers can safely touch.

This episode is the clearest look we've had at what AI adoption actually looks like at scale in Australia, not the pitch, the practice.

About the Guest

Beibei Guo is a Distinguished Engineer at CommBank where she has spent more than two decades building large-scale systems and the data platforms that now sit underneath the bank’s AI work.

She has lived through several technology waves, the internet, virtualisation, cloud and treats AI as the next one in that lineage: fundamental but not magic. Her framing is refreshingly unromantic.

“Elapsed time alone is not an asset it’s a data point.”

This isn’t a conversation about distant AI futures. It’s about what AI already looks like inside a major organisation on a normal weekday.

Everyone Got an API Key. That Was the Easy Part.

Beibei describes a single internal page: this is where you get your key, these are the models you can talk to and a banner setting the data classification you’re cleared to use.

The amount of engineering behind that one page was not small.

That’s the gap most AI conversations skip. Getting a demo running is trivial now, give anyone a model and 24 hours and they’ll show you something. Getting that same capability to thousands of engineers, safely, inside a regulated bank, is a different category of problem.

The cycles all look the same, she says: first chatbot, first lab model, everyone vibe-coding their own little tools, then the discovery that vibe-coding alone doesn’t ship. What separates organisations isn’t who runs the hackathon. It’s how fast that energy turns into something enduring that actually creates business value.

She credits the engineers who built the foundations underneath it all:

“the brightest minds and the steadiest hands.”

When Shipping Late Is the Right Call

One story stuck with us.

A rollout was ready. Beibei’s team delayed it by roughly three months to test against a much larger dataset first. More time, more cost, a missed deadline.

Why? Because shipping early on the smaller dataset made the odds of a bad outcome reaching a customer unfiltered an order of magnitude higher.

That’s the kind of call that doesn’t show up in a demo. Someone has to carry the weight of it, with no formula that tells you you’re right.

It’s also why she reframes the usual debate. Not hype versus reality. Hype versus progress. Reality can be bad; progress is the part worth chasing, and you can only tell the difference once you’ve lived through a few landings.

“I became less of a romantic about innovation but more serious about it.”

AI Needs Living Data

One of Beibei’s strongest points is that AI does not become useful just because a model is powerful.

It needs data that is alive.

By that, she does not just mean “clean data.” She means data that carries meaning. Data that is fresh. Data that is trusted. Data that remains consistent as it moves across systems, teams, domains and use cases.

Because in a large organisation, the data you need is often not created by you. It was captured somewhere else, for a different reason, by another system, under another team’s ownership.

If every team has to rediscover what a customer is, what an account is, what a transaction means, or where the freshest version of that data lives, AI does not become intelligence inside the organisation.

It becomes a confident layer on top of confusion.

Good data foundations do not always show up as a feature. They show up as fewer silly questions, less friction, better context, and better customer trust.

An Amplifier and a Mirror

Beibei’s sharpest line about what AI does to a person and a team:

“It’s an amplifier and a mirror. It doesn’t change who you are, it amplifies your impact, in both good and bad ways.”

The good: a colleague who manages IT assets, not a software engineer, needed to understand how thousands of assets connected. Three years ago, that’s a static spreadsheet. Now he built a live, switchable view: Venn, Gantt, list. He built it with AI, deployed it himself, and used it to make a sharper argument. Same job, amplified.

Her graduate engineers shipped production-grade data-comparison systems across hundreds of terabytes within months. But she’s precise about why it worked:

“We didn’t leave them with just a grad and a prompt.”

It was a few grads, a few models, a principal engineer checking in weekly, and clear ownership: this is yours. The amplification came from the structure around the tool, not the tool alone.

The mirror has a darker side too. Lean on it wrong and you get a sycophantic echo chamber, or a team so dependent that pulling the prompt away is an outage. Her warning to early-career engineers is one every team should hear:

“You answer questions way too quickly, before you learn to ask meaningful questions.”

The Wins You Never Notice

Ask where AI is already delivering for ordinary Australians, and Beibei doesn’t reach for anything flashy. She reaches for the invisible.

• Cost Benefit Finder (running since around 2019): machine learning that surfaces government entitlements customers are owed. More than $1 billion in benefits found for customers over its life.

• Small-business loans: turnaround on smaller loans moved from overnight to roughly two hours.

• Scam protection: name-check and account-check tools plus newer AI that proactively engages scammers directly with voice and messaging.

None of it announces itself as “AI”.

A customer doesn’t think “good data quality.” They just feel understood.

The accountant with a business account and a home loan is not asked a silly question, because the context is already obvious. The customer calling support does not think about machine learning; they just experience a shorter wait time or a better-informed operator.

That is the through-line: the best AI often disappears into the service.

Why This Matters Now

The old divide was between organisations experimenting with AI and those that weren’t. That line is gone!

Everyone is experimenting.

The new divide, on Beibei’s account, is judgment infrastructure: the habits, processes and informed people who can make a good call when the rules do not quite fit.

That is not just a governance problem. It is a capability problem. A leadership problem. A workforce problem.

It is the difference between using AI to make people better and using AI in a way that quietly reduces the person doing the work.

Beibei’s read on Australia is optimistic, but not boosterish. The talent is here. The rooms are here. The opportunity is here to build something the rest of the world may eventually look back on.

But only if we approach it with judgment and care rather than panic.

Listen to the Full Episode

In this episode of What The Tech (AU), we get into:

• Why the demo is the easy part and what real AI adoption costs

• The judgment call behind shipping late on purpose

• Why AI needs living data, not just clean data

• AI as an amplifier and a mirror

• How to keep graduates learning in an AI-native workplace

• The invisible, mundane wins that actually build customer trust

• Why long-term AI success depends on judgment infrastructure

🎧 Listen now on:

• Spotify

• Apple Podcasts

• Amazon Music

• YouTube

Question for readers

If everyone in your organisation already has the AI keys: what’s your version of the three-month delay and who’s carrying the weight of that call?

AI is moving from hype and experimentation into real-world execution but what does it actually take to make AI work safely and responsibly at scale? 

In this episode of What The Tech Podcast (AU), Rene and Vivienne sit down with Beibei Guo from CommBank to explore what AI adoption looks like inside large organisations. 

This conversation goes beyond demos and tools. We unpack the deeper foundations required for AI to create real value: trust, governance, data quality, engineering discipline, workforce capability and human judgment. 

Beibei shares a grounded perspective on why AI is different from previous technology waves, why strong data foundations matter, how AI is changing the way people learn and work, and why leaders need to think carefully about what should be automated, augmented or left firmly in human hands. 

We also explore real-world examples of AI in financial services, fraud prevention, customer experience and workforce enablement and what these examples reveal about the future of responsible AI adoption in Australia. 

In this episode, we cover: 

• Why AI experimentation does not automatically become execution 

• What makes AI different from cloud, internet and previous technology waves 

• Why trust and governance must be built into AI workflows 

• How data foundations shape AI reliability and customer experience 

• Why AI is both an amplifier and a mirror 

• How AI is changing learning, leadership and capability building 

• The trade-offs leaders must navigate between speed, risk and rigour 

• What responsible AI adoption looks like inside large enterprises 

If you are interested in AI adoption, responsible AI, enterprise technology, data foundations, AI governance or the future of work, this episode is for you. 

Subscribe to What The Tech Podcast (AU) for grounded conversations on AI, technology and the real-world trade-offs shaping Australia’s future. 

Software development is not just getting faster — It is changing shape.

That’s the unfashionable claim worth defending right now. Most engineering teams in Australia are running their AI work the way they’ve always run software — discovery, design, build, ship, maintain — with a faster autocomplete bolted on. They’re measuring the wrong things, hiring for the wrong skills, and spending money on the wrong models in the wrong phases.

On S02E04 of What The Tech, we sat down with Gerardo Estaba, Solutions Architect at Supabase, to dig into what is actually happening underneath the AI boom. The conversation kept circling back to the same uncomfortable observation: the Software Development Lifecycle as we knew it is being compressed into something with a fundamentally different geometry. And the teams who don’t see the shift are quietly losing.

We’re not the only ones saying it. EPAM has formalised it as the Agentic Development Lifecycle (ADLC) — “not SDLC augmented with AI coding assistants, but a lifecycle for building systems where LLMs sit at the core of product behaviour.”

Microsoft, IBM, and McKinsey have all published versions of the same observation since the start of 2026. The shift is real, named, and underway.

The conventional wisdom

For two decades, the SDLC has been the default mental model for shipping software. Plan, design, architect, code, review, test, integrate, deploy, monitor, maintain — somewhere between 6 and 12 stages depending on whose diagram you’re looking at.

Each stage is a handover. Each handover is a meeting. Each meeting is a new feature. The lifecycle wasn’t really about the software — it was about coordinating the people building it.

When AI arrived, most teams reached for the obvious framing: (1) code completion in the build phase, (2) AI-assisted review in the review phase, (3) synthesised tests in the test phase.

McKinsey puts the productivity, code-quality, and delivery-timeline gain at 20-45% when AI is integrated this way. Real numbers but they describe a faster version of the old shape.

That framing isn’t wrong, it’s just incomplete.

Why it’s incomplete

The handovers were never overhead. They were the lifecycle. The work that happens between phases — translating designs into specs, specs into code, code into tests, tests into deployments — is what most of the SDLC actually represents.

When the handovers collapse, the stages collapse with them.

On the episode, Rene put it like this:

“If you look at the software development lifecycle, you have so many bubbles, so many circles you have to go through. The AI development lifecycle is like two circles.”

Gerardo’s response was the part that matters:

“In the past, you always needed a team of engineers — 14 engineers — in order to do a thing. Now you have one person and all these tools.”

If you don’t have 14 engineers, you don’t need 14 handovers. And if you don’t need 14 handovers, most of the stages those handovers existed to connect simply stop earning their keep.

The work didn’t disappear. It got absorbed into the loop:

• Review still happens: inside the editor, before the PR is raised

• Documentation still happens: generated alongside the code, not after

• Test scaffolding still happens: written in the same prompt as the feature

• Security checks still happen: flagged at prompt time, not at audit time

What looks like two circles from the outside is doing the same work as twelve. Concurrently, rather than sequentially. This is what EPAM means when they say ADLC isn’t SDLC-plus-AI — it’s a different lifecycle for systems whose behaviour is non-deterministic and evolves continuously after deployment.

The new frame

The interesting question is no longer which stages of the SDLC survive in the AI era. The interesting question is which model you pick for what’s happening inside each steps.

This is where most teams are getting it wrong — and it’s where Gerardo was sharpest on the episode. Three phases, three different optimisation problems.

Plan with the smartest model you can afford

Most teams underspend on planning. They reach for whatever model is cheapest, run a quick scoping pass, and start coding.

“When you’re coding, you probably want to use the most advanced model during your planning session. You want that plan, the strategy, and those choices that you make early in the beginning — before you write a single line of code — to be rock solid. So in planning mode, yes, go for the most expensive one.”

Every wrong architectural choice in planning compounds. The smartest models — currently Claude Opus 4.7, GPT-5.4, or whatever’s leading the reasoning charts the week you’re reading this — have the highest chance of catching the wrong call before a single line of code gets written. This is exactly where saving $20 in tokens costs $20,000 in rework.

Plan once. Plan well. Let the cheaper models handle the rest.

Build with the cheapest model that still works

Most teams overspend on the build phase. Long coding sessions on top-tier models burn budget and add latency for output a mid-tier model would have produced cleanly.

“The top models for both OpenAI and Anthropic cost five times what the second best does. And the differences in terms of what they achieve is small.”

A useful fact-check: Anthropic’s current lineup has narrowed that gap considerably since older Opus pricing. Opus 4.7 sits at $5/$25 per million tokens, Sonnet 4.6 at $3/$15, and Haiku 4.5 at $1/$5 — so the flagship-to-mid-tier ratio is now closer to 1.7x. The 5x ratio still applies, but it’s now flagship-to-budget (Opus to Haiku). The principle Gerardo is pointing at holds: most build-phase work doesn’t need flagship reasoning, and the savings from routing down a tier are real and measurable.

How real? A March 2026 routing study by Augment Code found that three-tier Claude routing saves 51% versus uniform deployment of the flagship model. CloudZero’s range across customer deployments is 40-60% spend reduction with no measurable quality loss on routed tasks. This is one of the highest-leverage cost decisions in any AI deployment, and most teams aren’t making it deliberately.

Gerardo flagged Anthropic’s Claude Code as a common default for the teams he works with — “Claude Code is becoming the standard coding agent” — partly because the tiered model selection inside it (Opus / Sonnet / Haiku) maps cleanly to this kind of phase-aware workflow. Plan with the top tier. Build with the cheap tier. Whatever the equivalents are in the model family you’ve standardised on.

Ship with the model your users can afford

The step that matters — production — runs on a completely different optimisation function.

“When you deploy and break into production, having the best model out there is not going to help you deliver more value to customers necessarily. Those bigger models might be smarter, but they take longer and they cost significantly more.”

“The most popular models in production are almost never the smartest one.”

In production you’re not solving for smartest. You’re solving for cost, latency, availability, and scalability. A model that nails the answer 100% of the time in 30 seconds is unusable. A model that nails it 95% of the time in 800 milliseconds is a product.

The teams shipping AI well in Australia are quietly running three or four model tiers in production, routed by the actual nature of each request — benchmark for the use case, not the leaderboard; build fallbacks for capacity ceilings; design for cost-per-interaction, not cost-per-token.

What doesn’t compress

The reframe has a hard edge. Some things in the lifecycle didn’t get easier — they got more important.

“The foundations haven’t gone anywhere. The foundations are even more important than ever.”

When one person and an AI agent can ship in days what used to take a team an entire quarter, the foundations carry an outsized share of the risk. Security cannot be prompted into existence. It has to live at the data layer — access controls, row-level security, audit logs, infrastructure-grade guardrails. Not bolted on later as policy.

The pattern is: speed up everything you can; slow down deliberately on the few things that have to survive contact with real users.

What this means for the tooling

The most visible casualty of the lifecycle change is the entire category of collaboration tools built for the old shape.

“This is why Atlassian — one of the biggest, probably the first unicorn out of Australia — they’re not doing so well. They have all these development tools that are built for humans.”

At the time of recording, the market agreed: Atlassian’s stock had fallen 87% from its 2021 peak, hitting an all-time low near $57 in April 2026. Since then, the picture has gotten more nuanced — the company is leaning hard into the shift Gerardo describes. Rovo, Atlassian’s agentic AI layer, has crossed 5 million monthly active users, Q3 FY26 revenue grew 32% year-over-year, and the stock has rebounded roughly 30% off its lows.

But that rebound proves the underlying point. Even one of the most successful incumbent software companies in Australia is being forced to fundamentally retool its developer tools for a world where teams are smaller and lifecycles are shorter. The companies that don’t pivot this hard are the ones genuinely in trouble.

For Australian product builders this is both a warning and an eye-opening. The incumbents built around the twelve-circle lifecycle are being forced to redesign for less steps, and the white space for tools designed natively for AI-DLC has barely been touched.

The delivery

The AI development lifecycle is two circles, not twelve.

Inside the first circle: plan with the best model you can afford, build with the cheapest one that still works. Inside the second: ship with the model your users can afford, and hold the foundations that the speed depends on.

That’s the whole game. The teams in Australia running their AI work this way are quietly extracting 40-60% cost savings without giving up quality, shipping in weeks what used to take quarters, and building on foundations that survive contact with real users. The ones still running the old SDLC with AI bolted on top are paying flagship prices for build-phase work, leaving routing savings on the table, and discovering at scale what their foundations should have done in week one.

Listen to the full episode

In S02E04 of What The Tech, we explore:

• Why the AI development lifecycle compresses into two circles, not twelve

• How to choose the right model for each phase — plan, build, ship

• What doesn’t compress, and why foundations matter more than ever

• Why traditional collaboration tools are struggling to keep up

• What software development might look like five years from now

🎧 Listen on:

• Spotify

• Apple Podcasts

• Amazon Music

• YouTube

📬 Subscribe on Substack for episode breakdowns, reflections, and behind-the-scenes thinking.

Question for readers

Inside the two circles — Which one is quietly burning your budget?

“Anyone can build software now.”

That sentence would have sounded unrealistic five years ago.

Back then, shipping an application usually meant assembling engineers, architects, DevOps support, cloud infrastructure, authentication, databases, APIs, queues, deployment pipelines, and enough technical knowledge to know what could go wrong before it did.

Today, it is completely different story. A single person with a clear idea, an AI coding assistant, and a modern backend platform can build and launch something real over a weekend.

That is exciting and also slightly terrifying.

Because if AI-assisted development makes it easier for anyone to build, it also raises a much harder question: Can everyone build safely?

That is the centre of our latest episode of What The Tech Podcast (AU).

In this conversation, we sat down with Gerardo Estaba, Solutions Architect at Supabase, to explore how AI is reshaping the software development lifecycle, why the modern stack is becoming simpler, and why the boring parts of engineering — databases, security, architecture, and governance — suddenly matter more than ever.

About the Guest

Gerardo Estaba is a Solutions Architect at Supabase, one of the most widely adopted developer platforms among modern builders today.

He works closely with startups, product teams, and engineering organisations across APAC, helping them design, build, and ship modern applications faster.

Before joining Supabase, Gerardo spent almost nine years at Amazon Web Service, working with engineering teams through the cloud adoption era. That experience gave him a useful perspective on what has changed, what has stayed the same, and what many organisations still get wrong when they think about scale.

This episode is not just about Supabase.

It is about the bigger shift happening across software development:

The stack is compressing. The barrier to building is collapsing. But the foundations still matter.

The Software Stack Is Getting Simpler Again

For years, software teams borrowed architectural ideas from companies like Google, Netflix, and Facebook.

Microservices. Distributed systems. Complex cloud environments. Multiple databases. Layers of orchestration.

For companies operating at internet scale, those decisions often made sense.

But for many teams, especially in markets like Australia, the complexity arrived long before the scale did.

Gerardo puts it bluntly in the episode:

“Engineering teams largely do not have a scaling problem… 99% of the time, engineering teams have a complexity problem.”

That idea sits at the heart of this conversation.

In a world where AI can help people build faster, the cost of unnecessary complexity becomes even higher. Every extra service, every unnecessary abstraction, and every poorly understood integration becomes another place where security, reliability, and maintainability can break.

This is why platforms like Supabase are interesting in the current AI moment.

They do not just help developers move faster. They represent a broader return to simpler foundations: Postgres, authentication, storage, APIs, and security controls sitting closer to the data layer.

The lesson is not that every system should be simple forever; it is that teams should earn complexity.

AI Has Raised the Floor for Builders

One of the most exciting parts of the episode is the discussion around the rise of the non-technical builder.

Tools like Lovable, Claude Code, Cursor, and modern backend platforms have made it possible for founders, operators, designers, domain experts, and internal teams to build working software without starting from a traditional engineering background.

That matters.

Because some of the best product ideas do not come from people who know Kubernetes.

They come from people who deeply understand a problem.

Gerardo describes this as one of the most empowering shifts in AI: giving more people the ability to turn ideas into working products quickly.

This is the optimistic side of the AI development lifecycle.

More people can build.

More ideas can be tested.

More problems can be solved by the people closest to them.

But there is another side to the story.

AI Has Also Raised the Risk

If more people can build software, more people can also accidentally build insecure software.

That is where the episode becomes especially important.

AI can generate code. It can scaffold applications. It can create interfaces, database schemas, and workflows. But it does not automatically understand your compliance obligations, your internal security policies, your data access rules, or the long-term consequences of a poor architectural decision.

Gerardo makes this point clearly:

“The foundations haven’t gone anywhere. The foundations are even more important than ever.”

That line captures the real catch behind AI-assisted development.

The faster we build, the more important it becomes to design systems where security is enforced at the right layer.

Prompting an AI model to “be secure” is not a serious enterprise control. If permissions, access, and data boundaries only live in application code generated by AI, the system is fragile from the beginning.

But if those controls are enforced closer to the data layer, through proven infrastructure patterns such as Postgres row-level security, teams can allow more experimentation while still protecting the organisation.

That is the balance every company is now trying to find:

How do we let people move faster without letting risk move faster too?

Why Postgres Is Having a Moment

One of the most interesting parts of the conversation is Gerardo’s explanation of why Postgres has become so central to modern application development.

For many people, Postgres sounds like an old, boring relational database.

But that is exactly why it matters.

It is mature. It is widely understood. It is open source. It has a massive community. And because of its extension ecosystem, it can now support far more than traditional database workloads.

In the episode, Gerardo talks about extensions like:

• PostgREST, which helps expose Postgres through APIs.

• pg_cron, which supports job scheduling.

• pgvector, which allows teams to store and search AI embeddings directly inside Postgres.

That last point is especially relevant for AI products.

Instead of introducing a separate vector database too early, some teams can keep their application data and AI embeddings closer together, reduce orchestration complexity, and simplify the system.

Again, the theme is not “never use specialised tools.”

The theme is: Do not add complexity before the product has earned it.

What Still Requires Deep Engineering?

This episode also pushes back against the simplistic idea that AI will remove the need for engineers.

Yes, AI is changing the work.

Yes, more code will be written by AI.

Yes, smaller teams will be able to ship more.

But the need for engineering judgement is not disappearing.

In fact, it may become more valuable.

When systems scale, when security matters, when costs need to be controlled, when customer data is involved, and when an AI-generated product needs to become a production-grade business, deep engineering still matters.

Someone still needs to understand the trade-offs.

Someone still needs to know when Postgres is enough, when it needs optimisation, when read replicas make sense, when sharding becomes necessary, and when a simple product has become a serious system.

Gerardo makes this practical near the end of the episode when he mentions that Supabase is hiring people with deep operational Postgres expertise — because running millions of databases still requires serious engineering knowledge.

That might be the clearest answer to the “will AI replace engineers?” question.

AI will change the job.

But the deeper the system becomes, the more valuable real engineering judgement becomes.

The Bigger Shift: From Software Development Lifecycle to AI Development Lifecycle

One of the strongest ideas in the episode is that the traditional software development lifecycle is being compressed.

Planning, designing, building, testing, deploying, and iterating used to happen across teams, tools, meetings, and long delivery cycles.

Now, AI collapses many of those steps into a much faster loop.

Prompt. Generate. Review. Deploy. Iterate.

That does not mean the old lifecycle disappears completely.

But it does mean the shape of software development is changing.

The teams that succeed will not simply be the ones that use AI the most.

They will be the ones that understand where AI accelerates the process, where human judgement is still required, and where strong infrastructure must sit underneath everything.

That is the real message of this episode.

AI is changing software development. But the catch is that faster building requires stronger foundations, not weaker ones.

Listen to the Full Episode

In this episode of What The Tech Podcast (AU), we explore:

How AI-assisted development is changing the way software is built.

Why the traditional software development lifecycle is being compressed.

How platforms like Supabase are helping builders move faster.

Why Postgres is becoming central to modern AI-enabled applications.

What organisations should think about when non-technical users start building production tools.

Why security, compliance, and data access controls need to sit closer to the infrastructure layer.

And what deep engineering still means in the age of AI.

🎧 Listen now on:

• Spotify

• Apple Podcasts

• Amazon Music

• YouTube

📬 You can also subscribe here on Substack for episode breakdowns, reflections, and behind-the-scenes thinking from What The Tech Podcast (AU).

Question for readers

If AI has made it easier than ever to build software, what matters more now:

the speed of building, or the strength of the foundation underneath it?

In this episode of What The Tech Podcast (AU), Rene and Vivienne sit down with Gerardo Estaba from Supabase to explore the rise of the AI development lifecycle and what it means for developers, founders, product teams and modern organisations.

We unpack how AI-assisted coding, vibe coding, Postgres, open-source platforms and tools like Supabase are changing the way software gets built.

Gerardo also explains why many teams do not really have a scaling problem — they have a complexity problem — and why faster software development can create bigger risks if security, data access and engineering fundamentals are ignored.

In this episode, we cover:

• How AI is compressing the traditional software development lifecycle
• Why vibe coding is helping more people become software builders
• How Supabase simplifies the backend for modern applications
• Why Postgres still matters in the AI era
• The difference between building a prototype and shipping production software
• Why prompt-based security is not enough
• How row-level security helps protect data access
• Why technical debt may increase as more non-technical users build apps
• Why deep engineering judgement still matters when AI writes more code
• How open-source platforms reduce vendor lock-in
• What software development could look like in the next few years

This episode is for developers, founders, product leaders, AI builders and anyone trying to understand where software development is heading next.

Subscribe to What The Tech Podcast (AU) for grounded conversations on AI, software, data and the future of technology.

Newsletter: www.whatthetech.com.au

Hi everyone,

Quick check-in this week — we’ve got something new to share.

Over the past year, across two seasons of What The Tech Podcast (AU), we’ve sat down with some incredible people: founders building AI in Australia from the ground up; leaders shaping governance and regulation; designers working at the scale of hundreds of millions of users; engineers, researchers, and operators all trying to answer the same question:

What does AI actually look like in the real world?

But when we stepped back and looked at all these conversations together, something interesting stood out.

They weren’t isolated - They were connected.

The conversation about leadership linked directly to product. Product linked to governance. Governance linked to security. And across nearly every episode, the same themes kept showing up:

• Future of Work

• AI Adoption

• Governance & Trust

• Data Ownership

• Capability Building in Australia

So we wanted a way to see that.

Introducing: The Constellation 🌟

We’ve built something new.

An interactive map of every episode we’ve released.

Each episode is a star.
Each connection represents a shared idea, theme, or tension.

Instead of scrolling through a list, you can now explore the podcast like a system:

• Follow themes across episodes

• Discover connections between guests

• See how ideas evolve over time

Because the real value isn’t just in individual conversations.

It’s in how they connect.

Why this matters

One of the biggest takeaways from our latest episode with Dr Sue Keay is this:

“People feel like they’re being rushed to make decisions… and don’t actually have the tools to assess what AI means for their business.”

That’s exactly the problem we’re trying to solve with the podcast — and now, with the Constellation.

Not more noise.

But clearer patterns.

Built with AI (but led by humans)

We built this with Claude (Anthropic’s AI).

Not by handing everything over — but by collaborating:

• We described the system

• AI helped structure it

• We iterated until it felt right

It’s a small but real example of something we talk about a lot on the show:

AI works best when it augments thinking — not replaces it.

What to explore first

When you open the Constellation, try this:

• Filter Future of Work → see how the theme runs across the entire season

• Jump between episodes → follow how ideas evolve

• Click any episode → get key insights instantly

What’s next

This isn’t static.

Every new episode adds another “star”.

Which means over time, this becomes more than a podcast.

It becomes a map of how AI is actually evolving in Australia.

👉 Open the Constellation

Until next time,
Vivienne & Rene

Most organisations today would say they take diversity seriously.

They publish targets, report on representation, and speak confidently about inclusion in leadership forums. At a glance, it looks like progress is being made. Yet despite this growing attention, the gender gap in technology remains stubbornly persistent, both globally and in Australia. Women make up only around 28–30% of the technology workforce locally, and their representation drops even further in highly technical and senior roles.

The industry often describes this as a pipeline problem, as though the issue lies somewhere upstream in education or interest. But that framing, while convenient, obscures something far more immediate and far more actionable.

Because pipelines do not just leak. They are shaped.

And as Dr Sue Keay explains, one of the most decisive points in that system is not where most leaders are looking.

It is the very first job.

“…the first job that they get is crucial.”

That statement sounds simple. It is not.

The Gap Doesn’t Start Where We Measure It

Most diversity conversations focus on visible outcomes. Leaders talk about representation at the top of organisations, promotion rates, and executive pipelines. These are the metrics that get tracked, reported, and debated publicly. However, they are also the least useful place to intervene.

By the time a gap appears at leadership level, it has already been compounding for years. In fact, research consistently shows that one of the most significant drop-off points for women in STEM is not mid-career, but the transition from education into the workforce.

This is the moment where potential becomes trajectory.

And trajectory, once set, is difficult to change.

The First Job Is a System, Not an Event

The first role someone receives in Tech is not just a job. It is the beginning of a system that determines how their career unfolds.

• It is where they gain their first real experience.

• It is where they build relationships and networks.

• It is where they are seen, evaluated, and remembered.

These factors do not operate independently. They reinforce each other. Early experience leads to stronger CVs. Stronger CVs lead to better opportunities. Better opportunities lead to faster progression.

This is what makes the first job so powerful. It is not about immediate output. It is about long-term positioning.

The Quiet Decisions That Shape the Future

In the conversation, Sue describes a pattern that exists in many organisations but is rarely examined closely. Formal hiring processes tend to include structured checks, such as balanced candidate pools and defined evaluation criteria. However, not all hiring decisions follow the same rules.

Internships and short-term opportunities are often treated differently. They are seen as low-risk, informal, and easier to fill quickly. In practice, this means they are frequently allocated based on familiarity rather than process.

“We know this person… we just want to put this person on.”

And over time, those informal decisions follow a predictable pattern.

“Invariably, that person will be a man.”

No one intends for this to happen. It does not feel like bias. It feels efficient. It feels practical. But it is precisely this kind of decision-making that creates structural imbalance.

How the System Reinforces Itself

The mechanism is subtle, but once understood, difficult to ignore.

When internships are disproportionately given to men, those individuals gain early exposure, practical experience, and internal visibility. When full-time roles later become available, hiring panels evaluate candidates based on experience and familiarity. Naturally, those who were given appear more qualified.

At that point, the decision appears fair.

However, the outcome was already shaped months earlier.

“You’ve weighted your pool of applicants towards men.”

This is not bias in a single decision. It is bias embedded across a sequence of decisions.

The Pay Gap Starts Before It’s Measured

The same pattern applies to compensation.

Sue highlights that when men and women are hired at the same time, men are often offered higher starting salaries. The justification is rarely explicit and usually framed in terms of experience or negotiation. Each decision appears reasonable in isolation.

“The men will be on more money right from the get-go.”

However, salary is not static. It compounds.

Annual increases are typically percentage-based. Promotion thresholds are influenced by current pay levels. Opportunities often follow perceived seniority. Over time, a small initial difference becomes a significant structural gap.

“That is a gap that you may never close.”

This is why the gender pay gap in technology remains persistent, with estimates suggesting it sits around 20% in the sector.

It is not simply created later in careers. It is embedded from the start.

The Myth of “No Suitable Candidates”

One of the most common explanations for the lack of diversity in hiring is that there are not enough qualified women applying for roles. Sue encountered this argument directly and chose not to challenge it rhetorically. Instead, she changed the incentive structure.

She proposed a simple condition: If a team wanted a role filled, they would receive an additional role if they successfully hired a woman into the first position.

The result was immediate.

“It is amazing how quickly people were able to find a woman.”

This reveals something uncomfortable but important.

The issue is rarely the absence of talent. It is the level of effort the system is designed to exert in finding it.

When incentives change, behaviour follows.

This Is a Systems Problem, Not a Pipeline Problem

What makes this insight powerful is that it extends beyond gender.

It reflects a broader truth about how organisations operate. Outcomes are not primarily driven by intention. They are driven by systems.

This same pattern appears in another part of the conversation, when discussing AI adoption in Australia. Many organisations express strong intent to adopt artificial intelligence, yet struggle to translate that intent into measurable outcomes.

“Organisations are still trying to work out what AI means for their business.”

The barrier is not technology — it is structure.

Incentives, decision-making frameworks, and internal capability determine outcomes far more than strategy statements.

The gender gap in tech follows the same logic.

What Leaders Need to Reframe

If this is true, then the question for leaders changes.

It is no longer about fixing the pipeline in abstract terms. It is about examining the specific decisions that shape it in practice.

It is about asking who gets access to opportunity, who gains early experience, and who becomes visible within the organisation. These are not large strategic initiatives. They are everyday operational choices.

• Internship selection.

• Informal hiring.

• Starting salary decisions.

Individually, they appear minor. Collectively, they define careers.

The Work That Actually Changes Outcomes

The implication of Sue’s argument is not that organisations need more policies. It is that they need better-designed systems.

This means recognising that internships are not low-stakes decisions but high-leverage ones. It means analysing starting salaries at the point of entry, rather than focusing only on aggregated pay gap data. It means designing incentives that encourage equitable hiring behaviour, rather than relying on guidelines that can be bypassed.

Most importantly, it requires shifting attention from outcomes to origins. Because the system does not fail at the top. It performs exactly as it was designed at the beginning.

The Bigger Picture

The gender gap in technology is often framed as complex and long-term, and in many ways it is. But complexity should not be confused with abstraction.

The gap is not created by a single decision. It is created by many small ones.

• Decisions that feel efficient.

• Decisions that feel harmless.

• Decisions that feel temporary.

But compound over time.

“We have to make place at the table.”

The mistake is assuming that table is the boardroom. In reality, it is the very first opportunity someone is given or not given.

Final Reflection

If you look at your organisation today, the question is not whether you have a diversity strategy.

The question is much simpler:

• Who got the last internship?

• Who was given the first opportunity?

• Who started just one step ahead?

Because that is where the next generation of leaders is being shaped right now.

Listen to the Full Episode

This article builds on our conversation with Dr Sue Keay on the What The Tech (AU) podcast. In the full episode we cover:

• What AI adoption really looks like inside Australian organisations

• Why physical AI has accountability that software AI doesn’t

• The structural and cultural barriers facing women in AI leadership

• What real AI capability looks like at the national level

🎧 Listen now on:

• Spotify

• Apple Podcasts

• Amazon Music

• YouTube

📬 Subscribe to What The Tech (AU) on Substack for episode breakdowns and deeper explorations of how AI is reshaping work, technology and policy in Australia.

Question for readers

If you look at the internship hires your team made in the last twelve months — who got those opportunities, and who didn’t?

References & Further Reading

This article draws on insights from our conversation with Dr Sue Keay, Director of the UNSW AI Institute, on What The Tech (AU), alongside supporting research and industry data on gender diversity in technology and STEM.

• Australian Government — State of STEM Gender Equity 2024
  https://www.industry.gov.au/news/state-stem-gender-equity-2024
  (Women represent around 15% of the STEM workforce in Australia)

• Tech Council of Australia — Next Wave: Women in Tech
  https://techcouncil.com.au/next-wave-women-in-tech/
  (Women make up only 20% of the highly technical workforce, with significant drop-off over time)

• ACS / Industry Reports — Women in Technology Workforce Data
  (Women account for roughly 28–30% of the tech workforce in Australia)

• Workplace Gender Equality Agency (WGEA) — Gender Pay Gap Data
  https://www.wgea.gov.au/pay-and-gender/gender-pay-gap-data
  (Gender pay gaps persist across industries, with higher gaps in male-dominated sectors)

• Diversity Council Australia — Lifecycle Approach to Gender Equity in STEM
  https://www.dca.org.au/news/blog/more-than-getting-girls-into-science-the-lifecycle-approach-to-gender-equity-in-stem

  (Women remain underrepresented in STEM and experience persistent pay disparities)

• UNSW — The Fix for STEM Workplace Inequity? Change the System
  https://www.unsw.edu.au/newsroom
  (Highlights structural causes of inequity, including participation and pay gaps)

Artificial intelligence is no longer a future concept.

It is already embedded in the tools we use every day — from Generative AI writing assistants to systems that automate decisions across organisations.

On the surface, it feels like AI adoption is accelerating rapidly.

But when you look closer at what’s actually happening inside organisations across Australia, the reality is far less mature.

“I think it’s still pretty nascent… many sectors are really trying to work out what it’s going to mean for their business.”

Despite the noise, most organisations are still trying to answer a basic question:

What does AI actually mean for us?

About the Guest

In this episode, we are joined by Dr Sue Keay, Director of the UNSW AI Institute and Founder of the Robotics Australia Group, who has spent her career operating at the intersection of AI, robotics, and real-world deployment.

Her perspective is not shaped by theory alone, but by direct experience working across research, industry, and policy, ensuring that artificial intelligence moves beyond conceptual discussions and translates into tangible outcomes within organisations and across the broader economy.

This is not a conversation about what AI could become.

It is a conversation about what AI looks like today on the ground in Australia.

The Gap Between AI Hype and Reality

There is no shortage of AI conversation right now. But inside organisations, that conversation often turns into pressure — pressure to act quickly, without fully understanding the implications.

“People feel that they’re being rushed… and don’t actually have the tools to assess what AI is going to mean for their business.”

This creates a dangerous dynamic:

• High urgency

• Low clarity

• Unclear ROI

And in that environment, activity is often mistaken for progress.

Why Most AI Projects Stall

One of the most important insights from our conversation is that successful AI adoption is not primarily a technical problem.

It is an organisational one.

“If the CEO and the board are not behind AI, then just forget it.”

The organisations that are able to move beyond pilots and deliver meaningful outcomes are not necessarily those with the most advanced tools, but those with:

• Clear leadership alignment

• Strong decision-making structures

• A deep understanding of their internal processes

• And a willingness to experiment while accepting failure as part of the process

Because ultimately, AI does not operate in isolation.

It interacts directly with how an organisation already functions.

“If you don’t understand how you are making those underlying decisions… you’re just working in the dark.”

In that sense, artificial intelligence acts less as a solution and more as a mirror, exposing inefficiencies, gaps in accountability, and weaknesses in data and process maturity.

The Rise of AI Enablement

As organisations begin to recognise that AI is not simply a tool to deploy but a capability to build, a new type of role is emerging: AI enablement.

This is not about adding more engineers or relying solely on external consultants.

It is about embedding expertise within the organisation — individuals who can translate AI capabilities into real business impact.

“Having someone in-house… who can advise people… what these AI tools are useful for… what are the pitfalls… is really important.”

These roles serve as a bridge between technology and operations, helping teams understand not only how to use AI, but when to use it, where it adds value, and where it introduces risk.

Because adoption does not happen at the strategy level.

It happens in the day-to-day decisions made across teams.

The Real Asset: Data

If AI is the engine, then data is the fuel.

And yet, one of the most consistent challenges across industries is that organisations do not fully understand the value of the data they possess — or worse, they have already given it away.

“We’ve been conditioned to accept that we should give up our data for convenience.”

Across sectors such as mining, agriculture, and enterprise software, organisations have historically outsourced data collection and analysis, often through contracts that transfer ownership or control to third parties.

What seemed efficient at the time has now created a structural disadvantage.

Because in the age of AI, data is not just an asset — it is the foundation of competitive advantage.

And losing control of that data means losing control of future value.

The Problem with “AI Washing”

At the same time, the rapid rise of Generative AI has created a new challenge: the tendency to equate all AI with a single category of tools.

“When people think of artificial intelligence, they think exclusively about generative AI… and it’s possibly not the one that’s going to give you the most productivity benefits.”

This has led to what can only be described as AI washing, where products and strategies are rebranded to appear AI-driven without fundamentally changing how value is created.

The risk here is not just misplaced investment. It is a misunderstanding of what AI actually is.

Because while Generative AI is highly visible and accessible, the deeper impact of AI often lies in less visible systems:

• Decision optimisation

• Process automation

• Predictive modelling

• Robotics and physical AI

Focusing only on the surface layer limits the potential of what AI can deliver.

Why This Matters Now

We are moving into a phase where AI is no longer optional.

It is becoming part of the underlying infrastructure of how organisations operate and how economies compete.

The real divide is no longer between those who are experimenting with AI and those who are not.

It is between those who:

• Understand their systems and data

• Invest in long-term capability

• Build internal expertise

And those who:

• Rely on external solutions

• Chase trends without strategy

• And underestimate the complexity of implementation

Listen to the Full Episode

In this episode of What The Tech (AU), we explore:

• What AI adoption actually looks like in Australia

• Why organisations struggle to move beyond pilots

• The rise of AI enablement roles

• The importance of data ownership

🎧 Listen now on:

• Spotify

• Apple Podcasts

• Amazon Music

• YouTube

📬 You can also subscribe here on Substack for episode breakdowns, reflections, and behind-the-scenes thinking.

Question for readers

If AI is exposing how your organisation really works —

Are you ready for what it reveals?

Who owns the data behind AI — and who really benefits from it?

As artificial intelligence moves from hype to real-world adoption, the conversation is no longer just about what AI can do. It’s about data ownership, decision-making, and how organisations build capability in an increasingly complex landscape.

We unpack what AI adoption actually looks like on the ground in Australia today, why many organisations are still stuck in experimentation, and what separates those who deliver real outcomes from those who fall behind. We also explore the growing importance of AI enablement inside organisations, the role of leadership and culture, and why data is becoming a true source of competitive advantage.

This is a grounded, practical conversation about AI in the real world — from boardrooms to industry, not theory or hype.

If you work in tech, leadership, or policy or are trying to understand where AI is heading in Australia, this episode will challenge how you think about data, ownership, and innovation.

🎧 New episodes monthly
You can also follow us on Substack

For most of the history of software, product design meant designing visuals.

Designers focused on layouts, typography, colours, navigation patterns and interaction flows. The job was to translate product ideas into interfaces that felt intuitive and usable for real people.

But the rise of AI-powered products is quietly changing that definition.

When artificial intelligence becomes part of the product itself, the design challenge expands far beyond the screen. The interface is no longer the entire experience, it is just the visible layer sitting on top of a much more complex system.

In our recent conversation with Christina (CJ) Jones, Head of Design at Canva AI, we explored how this shift is reshaping the role of designers inside modern product teams.

Design Still Starts With the Problem

One of the first points CJ made during the episode is that AI doesn’t change the fundamental purpose of design.

At its core, design is still about solving a problem.

“Design is getting from the problem that’s at hand to the solution that you need.”

That solution might be an interface, a feature or even something more abstract like communication through visuals or storytelling.

But when AI tools can generate polished outputs quickly, the design process itself can appear less necessary.

Christina described how this perception is increasingly common, particularly among junior designers.

“It looks good enough… so people think they can get it from there.”

The problem is that looking good and solving the right problem are not the same thing.

And that difference becomes more obvious once you start digging into how AI systems actually work.

AI Changes What “The Product” Actually Is

Traditional software behaves predictably.

If a user presses a button, the software executes a predefined action written by engineers.

AI-powered products behave differently.

Behind a simple interface element might sit:

• A large language model

• prompt engineering

• evaluation frameworks

• guardrails and safety layers

• latency and infrastructure constraints

This means that the experience users interact with is no longer purely visual.

It is the result of a complex system working behind the scenes.

At Canva, CJ explained that designing AI features requires close collaboration across multiple disciplines.

“My team focuses on the AI features… and we actually need to consider the technical limitations of everything we’re building.”

In practice, this means design decisions can’t be made in isolation.

A feature idea might start with designers observing user behaviour, or with engineers experimenting with what a model can do.

“Either the designers have an idea, or the engineers have experimented with something… and then we ask, is it worth putting into the product or not?”

Design becomes a collaborative exploration of what the technology can realistically support.

Designing for Uncertainty

Another important shift with AI products is uncertainty.

Unlike traditional software systems, AI models don’t always produce the same output twice. They may generate incorrect results, hallucinate information or behave inconsistently across languages and contexts.

This means product teams must design not just the feature itself, but how the system behaves when things go wrong.

At Canva, this requires rapid experimentation cycles.

CJ described how teams often work in short development loops where ideas are quickly tested with users.

“We work in a six-week cycle… trying to design the product or build an MVP, getting that in front of users to see if it’s working.”

And sometimes the result of that process is deciding not to ship the feature at all.

“At the end we ask: is it worth putting into production, or is it too slow, costs too much, or the results aren’t great?”

These decisions involve trade-offs between user experience, performance and cost, all of which now influence the design of AI products.

Designers Are Becoming System Thinkers

As AI capabilities expand, designers are increasingly required to understand the broader systems behind the product.

Instead of focusing only on interface details, they must think about:

• model behaviour

• evaluation frameworks

• technical constraints

• user trust

In other words, design is evolving from screen design to system design.

This doesn’t mean designers need to become machine learning engineers.

But it does mean understanding how intelligent systems behave, and designing experiences that help users interact with them effectively.

A New Kind of Product Team

Companies building AI-native products are already adapting to this new reality.

Designers, engineers and product managers are working more closely than ever before, because the experience users see is shaped by decisions across the entire stack.

Christina described this dynamic as a collaborative process between technology and creativity.

AI is not replacing designers; it is becoming another creative tool.

“The way I look at it is that it’s a collaborator… it’s there to help and enhance your creativity.”

But like any collaborator, it still requires direction, judgment and oversight.

And that is where human designers continue to play a critical role.

Why This Matters

The shift from screen design to system design has profound implications for the technology industry.

The most effective designers of the next decade will likely be those who can move fluidly between:

• product thinking

• systems thinking

• human behaviour

They will not just ask: “What should this interface look like?”

They will ask: “How should this system behave?”

Listen to the Full Conversation

This article builds on our conversation with CJ, on the What The Tech (AU) podcast.

In the full episode we discuss:

• How generative AI is reshaping product design

• Why AI outputs can appear polished but still miss the problem

• How design teams experiment with AI features at scale

• Why AI should be treated as a creative collaborator

🎧 How AI Is Reshaping Design is now live on:

• Spotify

• Apple Podcasts

• Amazon Music

• YouTube

📬 If you enjoy these breakdowns, subscribe to What The Tech (AU) on Substack for deeper explorations of how AI is reshaping technology, creativity and the future of work.

Question for readers

If AI products are increasingly systems rather than interfaces, what new skills will designers need to develop?

Artificial intelligence is transforming creativity at a speed that would have seemed impossible just a few years ago.

Today, anyone can generate a website layout, a brand identity, a product interface, or even a marketing campaign within minutes. Tools powered by generative AI can now produce work that appears polished, coherent and ready to ship with very little effort.

For many people, this moment feels revolutionary. The barriers to creating digital products and visual work have never been lower.

But during our recent conversation with Christina Jones, Head of Design at Canva, a deeper tension began to emerge.

Because when everything looks good, something else becomes much harder.

Recognising what is truly great.

The Arrival of the “Good Enough” Era

Generative AI has dramatically lowered the barrier to entry for creative work.

Where designers once had to begin with a blank canvas and slowly iterate their way Where designers once had to begin with a blank canvas and slowly iterate their way toward a refined concept, AI tools can now generate dozens of plausible design directions almost instantly. A prompt can produce layouts, typography, imagery and interaction ideas in a matter of seconds.

This acceleration removes one of the most frustrating aspects of creative work: getting started.

But it also introduces a subtle risk.

As Christina explained during the episode:

“A lot of those outputs are just the average of everything.”

That observation captures something essential about generative AI systems.

AI models are trained on enormous datasets of existing designs, images and interfaces. When they generate something new, they are not inventing from first principles; they are synthesising patterns that already exist.

The result is often competent.

It is balanced, aesthetically pleasing and familiar.

But competence is not originality, and familiarity rarely leads to breakthrough ideas.

When the Floor Rises

In many ways, AI is raising the floor of creativity.

More people can participate in creative work. Ideas can be explored faster. Prototypes that once required days of effort can now be assembled in minutes.

For startups, product teams and independent creators, this shift is extraordinarily powerful. The ability to rapidly test ideas and visualise concepts removes a significant amount of friction from the innovation process.

However, raising the floor does not automatically raise the ceiling.

In fact, it can make the ceiling more difficult to reach.

When everyone can generate something that looks polished, the difference between average work and exceptional work becomes harder to see at first glance. The visual baseline improves, but the signal of true originality becomes more subtle.

Execution, which was once a scarce skill, begins to lose its scarcity.

And when execution becomes abundant, something else becomes the new constraint.

Judgment.

The Question of Craft

One of the most interesting parts of the conversation with CJ touched on how creators traditionally develop their craft.

She reflected on her time studying art:

“When I was in art school, we had to draw classical statues… that’s how it looked.”

Exercises like this may appear old-fashioned, but they served a purpose. Imitation allowed students to learn proportion, composition and technique. Repetition helped develop the intuition that later becomes taste.

In other words, mastery emerged through practice.

But generative AI now performs much of that imitation instantly. As we discussed during the episode, AI effectively becomes the imitator for you, producing variations and interpretations of existing styles without requiring the same years of repetition.

That shortcut raises an important question for the future of creative professions.

If the early stages of learning are automated, how do creators develop the deeper intuition and taste that separates great work from merely competent work?

From Creation to Curation

The answer may lie in a shift that is already beginning to take place.

The role of designers is not disappearing, but it is evolving.

When AI can generate dozens or even hundreds of possible solutions, the value of the human creator moves from producing the work to selecting and shaping it.

Someone still needs to decide:

• Which idea actually solves the problem

• Which direction aligns with the brand and product strategy

• Which interaction will feel intuitive for real users

This is where human judgment becomes essential.

As Christina explained:

“You need that craft and curation to actually say, this is good or this is bad.”

Generative AI can expand the creative search space dramatically, but it cannot evaluate meaning, context and intent in the same way humans do.

That responsibility still belongs to people.

Creativity in the Age of AI

This is the paradox of generative technology.

AI democratizes creation, enabling far more people to produce digital content, products and designs than ever before.

But at the same time, it raises expectations.

When the baseline level of work becomes “good enough,” excellence becomes harder to achieve.

The creators who succeed in this new environment will not simply be the ones who produce the most output. Instead, they will be the ones who develop a deeper ability to recognise quality.

They will:

• Define better problems

• Recognise stronger ideas

• Understand context and nuance

• Exercise taste and restraint

Execution is no longer the bottleneck.

Judgment is.

Watch the Full Episode

This article builds on our recent podcast episode with CJ, Head of Design at Canva, where we explored how generative AI is reshaping product design inside one of the world’s most widely used creative platforms.

In the full episode of What The Tech (AU), we discuss:

• How generative AI is changing the role of product designers

• Why “good enough” outputs can be misleading

• The tension between imitation and mastery

• How large design teams integrate AI responsibly

🎧 Listen now on Spotify, Apple Podcasts, Amazon Music, YouTube

And if you’d like deeper breakdowns like this, subscribe to What The Tech (AU) on Substack as we continue exploring how artificial intelligence is reshaping creativity, product development and the future of work.

Question for readers

If AI raises the floor of creativity, what skills will define the creators who still reach the ceiling?

Artificial intelligence is no longer a future feature. It is already embedded in the tools millions of people use every day.

• Design software generates layouts.

• Copy tools rewrite content instantly.

• Image models create visuals in seconds.

The surface layer of product design is accelerating at unprecedented speed.

The question is no longer whether AI will change design. It’s whether designers and product teams understand where the real leverage has moved.

Because when execution becomes instant, judgment becomes the differentiator.

About the guest

Christina (CJ) Jones is Head of Design at Canva, one of the world’s largest design platforms, used by hundreds of millions of people globally.

She leads Generative AI Design, working at the intersection of product, engineering and AI to shape how artificial intelligence is integrated responsibly at scale.

Her work focuses not just on what AI can generate, but how it should be evaluated, governed and deployed inside a real product ecosystem.

This isn’t theoretical AI.

It’s operational AI inside a global platform.

And at the core of her thinking is a deceptively simple definition of design:

Design is… can it get from the problem that’s at hand to the solution that you need.

That framing matters.

Because AI changes how we get to solutions — but it doesn’t remove the responsibility to define the problem.

The “Good Enough” Illusion

One of the most honest parts of the conversation came when CJ described what she’s seeing with junior designers:

I see this with a lot of junior designers where they feel like, what’s the point of my role now? … I could just put something into a model, and it can… it looks good enough.

That phrase — “it looks good enough” — captures the moment we’re in.

AI can produce polished outputs almost instantly. But CJ cautioned against mistaking polish for precision:

Once you start diving deep into those things that are generated… some things don’t work… a lot of those outputs are just the average of everything.

Average doesn’t differentiate.

Average doesn’t build enduring products.

Average gets you close — but not necessarily right.

From Craft to Curation

AI changes the entry point to creativity.

As we discussed in the episode, AI can act as what we called “the imitator for you” — aggregating patterns that previously required years of manual repetition.

But that shortcut raises a deeper concern about mastery. CJ reflected on her own training:

When I was in art school, we had to draw classical statues… that’s how it looked.

• Imitation builds proportion.

• Repetition builds taste.

• Constraint builds judgment.

If AI removes that early friction, what happens to the development of craft?

CJ doesn’t dismiss AI, instead she uses it deliberately. But she emphasises something critical:

If AI is generating all of these average outputs, you kind of need that craft and curation to actually say, this is good or this is bad.

That is the new bar.

• Not just generating — Curating.

• Not just prompting — Judging.

Finding Where AI Helps — and Where It Hinders

Perhaps the most practical insight from the episode was this:

It’s about finding in your process where AI actually really helps you and where does it hinder you.

This is not a binary debate. It’s not “AI replaces designers” versus “designers ignore AI.”

It’s about integration.

CJ even noted that some designers on her team deliberately avoid certain AI tools because:

It doesn’t help me think through the way I would normally do things.

That’s a powerful reminder that AI is a tool, but thinking remains human

Why This Conversation Matters Now

Creative industries are among the first to experience deep AI integration.

What happens in design will ripple into product management, marketing, engineering and beyond.

If AI can generate execution-level output in seconds, then the competitive advantage shifts upward:

• Toward taste

• Toward judgment

• Toward systems thinking

• Toward defining quality

This is not about replacement. It’s about evolution and that evolution is already underway.

Listen to the Full Episode

In this episode of What The Tech (AU), we explore:

• How Canva integrates generative AI at scale

• Whether AI is replacing or elevating designers

• How teams evaluate AI outputs before shipping

• The ethical tension around copyright and originality

• Why systems thinking is becoming the defining skill

🎧 How AI Is Reshaping Design is now live on:

• Spotify

• Apple Podcasts

• Amazon Music

• YouTube

📬 You can also subscribe here on Substack for episode breakdowns, reflections, and behind-the-scenes thinking.

Question for readers:

If AI makes execution instant, what becomes the true measure of a great designer?

Let us know what you think.

How does AI change the way we design products used by millions of people?

As generative AI becomes embedded into everyday tools, design is no longer just about layout and aesthetics. It’s about systems thinking, experimentation, evaluation frameworks, and making responsible choices at scale.

We unpack what “good design” means in the age of AI, how teams evaluate AI-generated images and content before they reach users, and why AI literacy is becoming a core skill for designers and product leaders alike. Christina also shares insights on ethics, originality, cultural context, and the tension between automation and human craft.

This is a grounded, practical conversation about building AI responsibly inside real products — not theory, not hype.

If you work in tech, product, design, innovation, or policy, this episode will challenge how you think about creativity in the AI era.

🎧 New episodes monthly

You can also follow us on Substack

An API call, sending prompts, receiving responses, and hoping the system behaves well enough to be trusted. This is a common approach taken today where organisations don’t build AI: They consume it. For early experimentation, this works. For real enterprise systems, that is when the wheel falls off.

In our Season 2 opener of What The Tech Podcast (AU), Dave Lemphers, CEO and co-founder of Maincode, explained why this gap exists and what sits underneath organisations that successfully move beyond demos.

That foundation is what he calls a model factory.

What a “model factory” actually is

A model factory isn’t just model training. It’s the entire system required to deliberately design, operate, and evolve AI models.

As Dave put it:

“A model factory is the environment and infrastructure… racks of compute and storage, allocations of GPUs for training, serving models, massive storage for data, and the software layer that makes it all work.”

This includes far more than inference:

• Data ingestion and curation

• Training and evaluation pipelines

• Model versioning and rollback

• Deployment and serving layers

• Monitoring, benchmarking, and governance

Most organisations never see this layer because SaaS APIs abstract it away.

That abstraction is convenient. It also hides trade-offs.

SaaS-style AI vs a model factory

SaaS-style consumption looks like this:

• Send prompt

• Get response

• Pay per token

• Trust the provider’s decisions

Dave summed up the real limitation succinctly:

“They don’t really care about GPUs and hard drives. They care about tokens.”

That’s fine until organisations need:

• Determinism

• Explainability

• Cost predictability

• Regulatory clarity

• Control over where data moves and who sees it

At that point, abstraction becomes friction.

A model factory flips the relationship:

• You choose which model runs which task

• You decide how data is used and retained

• You benchmark models against your requirements

• You optimise cost by routing work intelligently

Dave described how this works in practice:

“Running top-tier models for any possible question is insane… so you categorise the prompt and route it to the most appropriate model.”

This isn’t theoretical. It’s how teams survive at scale.

Prompt routing, distillation, and efficiency

One of the most misunderstood ideas in enterprise AI is that every task needs the best model.

Dave was blunt about this assumption:

“If you have a straightforward question… do you really need a top-tier model to answer that?”

Model factories rely on:

• Prompt routing to classify intent and complexity

• Smaller models for simple tasks

• Specialised models for reasoning-heavy work

They also leverage distillation and quantisation to reduce cost and compute:

“You can teach a smaller model to learn trends from a larger model… and then lower the precision to reduce the computation needed.”

This is how organisations operate within real constraints:

“We don’t have enough compute to serve the entire planet.”

Efficiency isn’t a nice-to-have. It’s survival.

Why benchmarking and ownership matter

Most enterprises rely on public benchmarks to choose models. Dave cautioned against that:

“Create your own benchmarks… questions that are specific to what you actually need.”

A model factory enables this because:

• You control the evaluation data

• You track improvement over time

• You avoid optimising for irrelevant capabilities

This also ties directly to governance and risk:

“If you do not have ownership and transparency over what the AI model is doing… you are running a risk.”

Ownership isn’t ideological. It’s operational.

Why this matters for Australian enterprises

Australia doesn’t have unlimited compute. It doesn’t have hyperscaler-level infrastructure density. And it can’t rely on cross-region ambiguity when compliance matters.

Dave highlighted a hard truth:

“There is not enough compute around… especially here in Australia.”

Model factories allow organisations to:

• Work within local constraints

• Reduce dependency on opaque platforms

• Build capability incrementally

• Prioritise high-impact use cases first

This isn’t about rejecting global platforms. It’s about knowing when abstraction helps and when it hurts.

The takeaway

A model factory is not about building the biggest model.

It’s about building:

• The right model

• For the right task

• At the right cost

• With the right controls

Or as Dave put it more simply:

“It’s much easier to roll your sleeves up and start building.”

Watch to the full episode

🎧 Australia vs Big Tech is now live on: Spotify, Apple Podcasts, Amazon Music & YouTube

If you’re responsible for AI outcomes; technical, financial, or regulatory; this episode is worth your time.

References

• What The Tech Podcast (AU) – Season 2, Episode 1: Australia vs Big Tech
  Primary source discussion with Dave Lemphers, CEO & Co-founder of Maincode, covering model factories, enterprise AI trade-offs, compute constraints, and governance.
  👉 https://www.whatthetech.com.au/p/australia-vs-big-tech-why-build-ai

• F5 Ecosystem - What is an AI Factory?
  👉https://www.f5.com/company/blog/defining-an-ai-factory

• US Federal Government – Understanding and managing the AI lifecycle
  👉 https://coe.gsa.gov/coe/ai-guide-for-government/understanding-managing-ai-lifecycle/

They experiment with summarisation, text generation, and simple assistants. The results are often impressive at first. Then teams try to apply the same systems to real business processes, and momentum slows.

This topic came up repeatedly in our Season 2 opener of What The Tech Podcast (AU), where we discussed why early success often gives way to frustration.

Context for this discussion:
https://www.whatthetech.com.au/p/australia-vs-big-tech-why-build-ai

As Dave Lemphers explains, the problems begin when businesses try to move from demos to real work:

They start with summarisation and text generation… then they look at their business processes and realise the model can’t actually do the task.

The issue isn’t effort or intent. It’s architectural mismatch.

Large language models were not designed to deeply understand rules, classifications, or domain-specific decision logic. When organisations try to force them into those roles, they compensate by adding more context, longer prompts, and increasingly fragile instructions.

You’re cramming the context window with data… trying to come up with an incantation that’s going to make the magic pop out.

Even then, accuracy rarely reaches a level that businesses can trust.

This is why many AI initiatives stall. The model works, but not in the way the business actually needs. At that point, organisations face a hard choice: redesign the system, or accept permanent limitations.

Watch to the full episode

🎧 Australia vs Big Tech is now live on: Spotify, Apple Podcasts, Amazon Music & YouTube

Australia is adopting artificial intelligence at pace. But when it comes to building AI locally, the gap between ambition and reality is becoming harder to ignore.

Most of the AI systems shaping Australian businesses, governments, and institutions today are designed offshore, governed by foreign frameworks, and optimised for global platforms rather than local needs.

The question is no longer whether AI will shape Australia’s future.
It’s whether Australia will remain dependent on Big Tech, or develop real AI capability of its own.

About the guest

Dave Lemphers is the CEO and co-founder of Maincode, a Melbourne-based artificial intelligence company working across public and private sectors.

Dave specialises in AI engineering, open-source models, and enterprise adoption, with a focus on governance, data sovereignty, and deploying AI systems that organisations can operate responsibly at scale.

Australian-made vs “Sovereign” AI

A key part of the conversation focused on language, particularly the difference between “Sovereign AI” and what Maincode calls Australian-made AI.

Dave explained that after starting with the idea of sovereignty, the team’s thinking evolved once they began building:

“What we quickly realised is sovereign AI as a concept is very narrow, and it’s very difficult. It’s more rhetoric.”

Instead, they focused on something more concrete:

“Australian-made AI is literally what’s written on the can. We’re building in Australia, with Australians… and we want to sell it overseas.”

For Dave, this isn’t about status or symbolism.

“It’s not about publishing research or deploying models for vanity. It’s actually building the foundation for grassroots AI locally for people to work in this industry and contribute globally.”

Introducing AI Model Factory

Dave also spoke candidly about why many organisations hit a wall after early experimentation.

“They start with summarisation and text generation… then they look at their business processes and realise the model can’t actually do the task.”

Fine-tuning alone often isn’t enough.

“An LLM from the very basics cannot do that task because it wasn’t built for it.”

This is where Maincode’s idea of a model factory comes in, not just infrastructure, but the ability to design, train, and operate models end-to-end.

“They don’t really care about GPUs and hard drives. They care about tokens.”

Human-Centred Design

The episode also addressed concerns about jobs and displacement.

Dave pushed back strongly on fear-based narratives:

“Every major technological evolution has had people feel afraid they’re going to be displaced.”

He argued that many of these fears are amplified by large incumbents:

“The doom rhetoric comes from massive technology companies with agendas who need people to feel terrified.”

Maincode’s stated principle is explicit:

“The very first leadership principle we have is ‘for humans’. We think about humans and what’s good for humans.”

The goal is not replacement, but augmentation.

“We don’t want to replace them. We just want to make them better.”

Why this conversation matters now

Australia is at a crossroads.

We can continue consuming AI built elsewhere and adapt our systems around it.
Or we can invest in local capability, even if that path is slower, harder, and more constrained.

Neither option is simple. But avoiding the trade-offs is no longer viable.

Listen to the full episode

🎧 Australia vs Big Tech is now live on:

• Spotify

• Apple Podcasts

• Amazon Music

• YouTube

📬 You can also subscribe here on Substack for episode breakdowns, reflections, and behind-the-scenes thinking.

Question for readers:
Should Australia prioritise building its own AI capability, even if it’s slower, or is relying on Big Tech the pragmatic choice?

Let us know what you think.

Should Australia build its own AI or stay dependent on Big Tech?

Recorded ahead of Maincode’s public showcase of Matilda, an Australian-developed AI model, this episode cuts through hype to explore what Australian-made AI really means in practice.

We unpacked open-source vs closed models, data sovereignty, AI agents vs chatbots, enterprise readiness, and the real trade-offs between speed, control, and capability. Dave brings a grounded, engineering-first perspective on why many organisations aren’t ready for full autonomy yet and what Australia must do to move from AI consumer to AI builder.

If you care about Australia’s role in the global AI landscape, this episode is for you.

🎧 New episodes monthly
📬 Also available as a newsletter on Substack

The systems shaping our work, data, and decisions are increasingly designed offshore, governed by foreign priorities, and optimised for global platforms rather than local needs.

So the real question isn’t whether AI will transform Australia.
It’s whether Australia will remain a customer or become a builder.

For the opening episode of Season 2, we sat down with Dave Lemphers, CEO of Maincode, to unpack what it actually takes to build AI capability in Australia beyond slogans like “sovereign AI” and beyond Big Tech dependency.

Recorded just ahead of Maincode’s public showcase at SXSW Sydney 2025, this conversation cuts through the noise and focuses on real trade-offs, constraints, and hard realities.

🎬 Watch the trailer below
The full episode drops next week.