How product teams are restructuring for AI

AI changed what your teams can build but not how they're organised to build it

AI agents can triage bugs, generate code, and ship fixes to production. In 2024, these capabilities felt experimental. By 2026, every team has them. The capability gap between augmented teams and everyone else has evaporated faster than most people expected.

But look at how companies have responded. You've grafted AI onto the same org charts, the same handoff chains, the same role definitions you had three years ago. Your engineers got coding copilots. Your product managers got summarisation tools. The org chart stayed identical. The reporting lines didn't move.

The companies pulling ahead have rethought how work flows through the organisation now that the old constraints have dissolved.

You've given your team a jet engine and kept the horse

There's a concept in design called skeuomorphism. It's when a new technology mimics the old one it replaced. When the iPhone launched in 2007, the first wave of apps did this. The notes app looked like a yellow legal pad. The calculator looked like a desk calculator. It took years for designers to realise the phone was its own medium, with its own rules.

In an episode of The Airwallex Podcast, Shannon Scott spoke with Dhruv Amin, CEO of Anything, a platform that lets non-technical people build production software from a prompt. Dhruv sees the same pattern in how companies have organised around AI. You've been given powerful new agents and your first instinct is to slot them into the old workflow. Open twelve browser tabs, babysit each agent to completion, follow the same PM-to-designer-to-engineer handoff chain you've always followed.

Your traditional model runs in a straight line. A product manager writes a spec. A designer translates it into mockups. An engineer builds it. QA tests it. Each handoff introduces delay, and each delay opens a gap where context bleeds out.

At Anything, a 20-person company, that handoff has started to vanish. Their customer support person used to flag a bug in Slack, wait for an engineer to investigate logs, diagnose the issue, and ship a fix. Now that same support person pings an agent. The agent triages the bug, investigates the codebase, and ships a first-pass fix, often before an engineer would have even opened the ticket. The support person hasn't become an engineer. The boundary between "reporting a problem" and "fixing a problem" has simply collapsed.

The bottleneck has moved and most teams haven't noticed

For years, the biggest constraint in building software was engineering bandwidth. Every company had more ideas than capacity. The backlog was always long. The answer was always "hire more engineers" or "prioritise harder."

AI has blown that constraint apart. Your backlog is no longer gated by how many people can write code. And that's created a new, less obvious problem.

When more people can contribute to production, coordination gets harder. You can deploy unlimited agents, and the beautiful thing about them is they're 24/7, always running. But getting humans aligned on what to build and why is still the grind. As Dhruv puts it, you can have tons of agents, but coordinating humans is still the real constraint. Getting people on the same page around a roadmap hasn't gotten any easier.

Your customers' expectations compound the challenge. When software gets cheaper to build, the bar for what "good" looks like rises across the board. A solo founder can now ship in a weekend what a startup couldn't ship in a quarter five years ago. That recalibrates the competitive maths for everyone.

The companies thinking with precision about this are asking a different question, and it's a good one. Now that building is fast, where do you invest the human attention you've freed up? In most cases, the answer points to product differentiation, customer understanding, and distribution, all areas where your judgment is the irreplaceable ingredient.

And there's a question coming down the track that most teams haven't grappled with yet. Dhruv frames it well. How do you start designing products and experiences where your user is going to be an agent itself? When your customer's AI agent is the one interacting with your software, the entire UX conversation changes.

The person who knows the problem can now build the fix

The value shift happening underneath all of this is worth pausing on. For decades, the ability to write code was the bottleneck skill in technology. If you could code, you could build. If you couldn't, you needed to hire someone who could.

That equation has flipped. The scarce input is domain knowledge, the kind you build over years of operating in a specific industry or workflow. Vibe coding has become a consumer phenomenon. The tools are getting to the point where, wrapped in the right systems and with the right tooling, they can accomplish very real work.

Think about SMB owners running recruiting agencies, finance operators managing multi-entity cash flows, logistics coordinators who've spent a decade learning where freight shipments break down. These people have always had ideas for better software. Now they can turn those ideas into working applications without hiring a single developer.

The moat for these businesses is the knowledge itself. Someone can copy the code. They can't copy ten years of understanding how a recruiting agency operates, which candidates ghost at which stage, which fee structures create the wrong incentives, which compliance requirements trip up new entrants in different states.

Dhruv sees this playing out across Anything's two million users. The most successful are people with deep expertise in a problem, not necessarily technical founders, from firefighting to real estate to finance, who can now build software shaped by that expertise rather than by what an outsourced developer understood from a brief.

Your org chart is the thing holding you back

You've already adopted the tools. The question is whether your organisation is still structured as if engineering is the bottleneck.

In the podcast episode, Shannon and Dhruv explore how AI is reshaping the way product teams are built, why most companies are still layering AI onto outdated workflows, and what changes when domain experts – not just engineers – can ship directly to production.

In many companies today, finance teams have ideas for internal tools but are forced through a six-month IT backlog. Customer support spots product issues but can only surface them through tickets that disappear into sprint cycles. The tooling to close these gaps already exists, but most org structures haven’t evolved to use it.

At Airwallex, we’re already seeing a shift: non-engineering teams are building internal tools shaped by their own context and workflows, replacing third-party software with solutions designed by the people closest to the problem.

The early adopter tax is real, but it’s shrinking. Dhruv compares the current moment to the chip era, where rapid gains in processing power repeatedly reset what computers could do. AI is following a similar curve – every few months, capability jumps expand what’s feasible to build. Waiting for the “right time” is increasingly a false assumption.

The companies that adapt their structures to this shift will move faster than those that don’t. The tools are here. The constraint is no longer technology – it’s whether your org chart has caught up.