There was a moment in this conversation where the penny properly dropped.
A year ago we were excited about AI coding assistants. They scaffolded functions. They generated tests. They helped engineers move faster. It felt meaningful, but incremental. A productivity layer over an existing way of building software.
Now we are somewhere else entirely.
As we discuss in the episode, the shift is no longer about assistance. It is about delegation.
The Agent is not helping to write the code. IT IS WRITING THE CODE!
That sounds dramatic. It is. But it is also true.
When we first explored AI tooling inside engineering teams, the gains were obvious. Faster prototyping. Quicker scaffolding. Better test coverage. Engineers remained firmly in control. The AI was a tool, albeit a powerful one.
Today, we are orchestrating agentic platforms that can reason over requirements, generate architecture, write services, refactor modules, and wire systems together at a pace no human team can match.
That does not mean humans are redundant. It means their role has changed.
The hard part of software has never been typing. It has been thinking.
It is deciding whether to use this queue or that queue. Whether to adopt PubSub or polling. Whether a change will triple your database bill. Whether your architecture will survive scale, regulation, or a security audit.
Agents can generate code. They do not carry accountability.
That still sits with us.
There is a temptation to believe this becomes a pure speed story. That we can let agents run wild and ship enterprise systems in days.
That is naïve.
Complex systems still require architectural judgement. They require cost awareness. They require a deep understanding of cloud services, trade offs, and long term operational implications.
If you let agents loose without guardrails, they will build. They will optimise for local success. They will not optimise for your margin, your risk profile, or your regulatory exposure.
The real opportunity is not replacing engineering discipline. It is amplifying it.
Agentic development works when experienced engineers move up the stack. When they become orchestrators. When they shape the system and let agents execute within boundaries.
That is the shift.
There is another truth buried in this transition.
The entry point has dropped.
Almost anyone can now spin up a functional application with the right prompts and tools. The barrier to experimentation has collapsed. That is extraordinary for founders, product teams, and internal innovation squads.
But enterprise software is not a weekend project.
Significantly complex systems still demand experience. They demand systems thinking. They demand someone who understands how cloud services are strung together and why you would choose one pattern over another.
In other words, the floor has lowered. The ceiling has not.
This is not happening in a vacuum.
Boards are asking harder questions. What is our AI strategy? Where are we applying agents? Why are competitors shipping faster?
At the same time, many organisations are still wrestling with legacy estates, brittle integrations, and governance models built for a slower era.
The result is tension.
CIOs are being pulled from enablers of strategy to authors of it. They are being asked to reduce cost, increase velocity, and transform operating models simultaneously.
Agentic software development sits right in the middle of that pressure.
If agents can genuinely build and maintain significant portions of enterprise systems, then delivery capacity is no longer linearly tied to headcount. That challenges decades of thinking in both consulting and in house IT.
The most important insight from our real world experimentation is not that agents can build software.
It is that they need to be controlled.
Context windows are finite. Reasoning is powerful but imperfect. Agents do not understand commercial nuance. They do not instinctively care about cost or governance.
You need someone in the loop who does.
The future engineering leader is not a typing machine. They are a systems thinker. A constraint manager. An orchestrator of agents and humans.
This is not about replacing engineers. It is about redefining them.
We are at the beginning of a structural shift.
Just as cloud rewrote infrastructure economics, agentic development is rewriting delivery economics. The speed of change will not be uniform. Mature organisations will move more slowly. Startups will move aggressively. Consultancies will have to rethink how value is measured.
But the direction of travel is clear.
Agents are building software now.
The question is not whether that will continue. It is whether you will shape how it happens inside your organisation, or whether it will happen around you.
This is not hype. It is lived experience from production environments.
And it is only getting started.