AI-assisted development is here to stay. Engineering leaders are scrambling to adapt. Despite the hype, many things remain the same. We’ve been through much of this before!

TL;DR

  • Remember lessons learned building global engineering organizations
  • Know the limits of what the technology can do
  • Focus on core engineering practices. They are more important than ever

Everything has changed for ICs.

The day-to-day workflow for developers has changed. Even if devs aren’t using AI to generate code. It can create documention, explain new technologies, and aid in exploring new code.

The fundamental laws of software development haven’t

The automobile completely changed how we transport ourselves. But it didn’t change the laws of physics. We still need roads and fuel. Cars still break down. Software is in a similar state.

A lot of snake oil salesmen are promising the world to keep their supply of VC (and sovereign wealth fund) money coming in. Use their predictions with a grain of salt.

Writing code has never been the hard part

We’ve been able to generate low cost code for a generation. There are armies of devs available (offshore and onshore) who are happy to copy and paste their way to something close to what you asked for. This wasn’t an easy win then, and still isn’t now.

AI generated code is similar. It shows lots of promise, but ultimately can’t produce something that can be released.

Context and communication are still hard

AI generated projects will fail in the same way offshore projects did.

Context is king

We discovered this the hard way. Offshore projects looked great for the first 80% of the work. Then progress stalled. Many projects needed to be scrapped altogether. Lack of context was a primary reason for these failures: the offshore team didn’t understand enough of the product to make smart decisions. It took a lot of late-night meeting and documentation effort to give the offshore teams the context they needed to build shippable software.

Today, there are lots of skilled and productive offshore teams. But they require ongoing investment to keep everyone on the same page.

Context is still hard. Your $20/month subscription is a toy. Only good for simple tasks. Even plugging in your API key to unlock complex prompts has it’s limits. Your LLM can only handle so much context before it starts spitting out nonsense. Someone has to make sure the LLMs has the right amount of context. Context Engineering is a crucial skill.

Long term thinking matters

LLMs have no idea where your company is heading. And have no stake in your success.

Offshore consulting companies only cared about delivering the project. We learned to build relationships and even open remote offices. This gave the folks doing the work same long term incentives that our local employees had.

Don’t expect long term thinking from ai-generated code. Humans still need to own the roadmap and architecture.

Slow feedback kills projects

Slow feedback loops caused delays that drained the expected benefit of lower cost development.

This is a clear win for AI-assisted development. Feedback time is reduced to seconds.

AI generated code is mid

LLMs do not know how to reason. They predict. They guess at what we want them to do based on the code used to train them. Does the training code match your expected quality and security? Probably not. Beware the codebase filled with average, unremarkable code.

Engineering practices are more important than ever

Tech debt will bite you sooner. A lot of teams could slide for months or years before seeing the problems. AI generated code can show the cracks in your design in days.

Test coverage is critical for making sure your software remains in a working state. Finding breaking changes quickly is a super-power for your team. Even if the code is being written by AI.

Linters and static analyers become even more effective. AI can check it’s own work to deliver code to your desired spec. Without humans even reminding it to.

Takeaways

You have the tools to navigate this world

The principles that worked in the past are still valid.

You still need strong engineers (and leaders)

Humans need to be in the loop. At every stage.

Automate everything

Automations have always been force multiplyers. Now even more so.

Double down on Fast Feedback

Build that prototype. Try things out!

What do you think?

The above seemed controversial when I started jotting them down in fall 2025. Deep in the initial hype cycle. But I’m seeing more and more devs coming to the same conclusions. 2026 will be a fun ride.

I’ll dig deeper into these topics in future posts. Stay tuned!