underlying society is changing.
That was one of the ideas from Max Buckley’s talk at AI Engineer Singapore, and it has stuck with me ever since.
For decades, software engineering was organised around scarcity. Code was expensive to write, engineers were scarce and features took time. This assumption shaped how teams worked. We prioritised carefully because every feature had a huge opportunity cost.
But AI has broken that assumption.
As coding agents become more capable, the cost of implementation is falling dramatically. What used to take weeks can now be prototyped in days or even hours. Max, who is Head of Knowledge Research at Exa after spending 12.5 years at Google, framed this as a change in game theory: it is not just about asking what you should do, but what you should do when everyone else is also behaving this way and trying to win.
And you cannot opt out of these changes. Whether or not we’re ready for it, the fundamental ways of working are changing.
But cheaper implementation does not mean better software.
No amount of AI can save us from building the wrong thing. In fact, AI may make that problem worse. When building becomes easier, it becomes much easier to create things that are technically impressive but strategically irrelevant: More dashboards, more workflows, more internal tools, more apps that work but do not deserve to exist.
That is why I think engineering judgment is becoming more valuable.
One of Max’s examples stayed with me. In the old software economy, teams must narrow 30 ideas down to 3 before building anything. With coding agents today, the decision process changes. You can build more, evaluate more, benchmark more, and discard what does not work with less emotional attachment.
The cost of trying is lower and experimentation becomes more attractive.
That sounds liberating but it also creates a new bottleneck.
If anyone can prototype an idea, attention is now a scarce resource. That also means I should say a thank you to each person reading this. Your attention is not free, and I hope I made this piece worth your time.
I recently attended the inaugural AI Engineer conference in Singapore, held from 15 to 17 May 2026. It brought together speakers from companies like Google DeepMind, Vercel, OpenAI, Exa, NanoClaw, and others. This article will detail 3 points from 3 speakers that stood out to me.
AI is not removing the need for engineering discipline. It is moving that discipline to a different part of the system.
Technical knowledge is changing shape too.
Models have jagged intelligence: They can be extremely good at some tasks, but surprisingly bad at nearby tasks that seem equally easy to humans.
Models often know the answers to complicated things but will not surface it unless you know what to ask.
So the question is no longer just whether we can build something. It is whether it should exist.
Jimmy Lai, Director of Next.js at Vercel, shared a similar sentiment from a different angle. His point was that AI has made creation cheap, but ownership more expensive.
When building becomes easier, the number of things we can create goes up. But every prototype that survives becomes something someone has to maintain, debug, document, secure, and explain. The cost of writing the first version may fall, but the cost of owning the system does not disappear.
Jimmy made three predictions that stood out to me.
First, we are now building for agents. Agents are becoming a new kind of software user. A stale README is no longer just annoying for a human. It is a hallucination waiting to happen.
Second, we are now building with agents. Ironically, though it now gets easier to be able to build something you don’t quite understand, the truth is that the fundamentals have not changed and have in fact become more important than ever before. If you become excellent at building with agents while also being strong in the fundamentals, you become unstoppable.
Third, we have to learn what not to own. Just because you can build something does not mean you should. The ease of creation has become a maintenance burden.
This does not mean we should ship less. It means we should be more intentional about what we allow to survive. The advantage goes to teams that know what makes their product different, what deserves their attention, and what they should deliberately not build.
In a world where software is cheap to create, focus becomes an engineering asset.
Finally, my last key takeaway came from a design talk.
Phil Hedayatnia from Airfoil gave a talk on how to create design agents that actually have taste in a sea of very normal AI slop. I am not a designer, so I usually think of design in terms of what a good design should or should not contain. His talk reframed that for me.
Design isn’t trying to teach someone about what to do and not what to do. That is training on outcomes.
Good design is about understanding how people think, how they act, and why certain flows, visuals, and narratives resonate with them. Phil alluded it to human psychology.
It’s less about looking at what people make, but spending more time trying to understanding why they made it that way and the thought process behind it.
In other words, taste is not a checklist. It is judgment applied to context.
Phil gave the example of the Shinkansen bullet train and the kingfisher’s bill. The train had a problem: when it exited tunnels, it created a loud “tunnel boom” caused by compressed air. Engineers reduced the noise by modelling the train’s nose after a kingfisher’s bill. A kingfisher can dive from air into water with very little splash because its long, narrow, tapered bill reduces sudden pressure changes. The engineers applied the same principle to the train, using a longer and more tapered nose to compress air more gradually.
What I liked about this example is that it was not just about copying nature. It was about understanding why something worked, then applying that principle in a different context.
And as AI makes it easier to produce outputs, the valuable skill is not about simply knowing what a good output looks like. It is understanding the why behind it.
To wrap up
Across many of the talks, there were many recurring themes, such as building personal assistants, trying new tools, and learning how to work more efficiently with agents. But underneath all of that, the same idea kept surfacing: Code is becoming cheaper, but judgement and taste are not.
To summarise my 3 key takeaways:
- Implementation is no longer the main bottleneck. AI lets you try more ideas and lower the costs of being wrong. But that makes engineering judgement more critical. We have to decide what deserves to exist.
- Cheap creation creates a maintenance burden. Decide what not to own.
- In a world of abundant output, create products with better taste. Understand the context behind why something works.
AI has changed the way we build software, but it has not remove the responsibility and ownership behind it.
That’s it from me. I hope this was worth your time. The full talks are on the AI Engineer Youtube here. See you in the next article!
