My thoughts on AI

It's been 7 months since my first post in Jan 2026 - and a lot has changed. I'm writing this post to share my thoughts on AI and to keep a record for future reflections.

TL;DR

• Can AI replace you - a software engineer? No - unless you refuse to use it.
• Should you still write code by hand? Rarely. AI-generated code is good now.
• The hard part is balance. You delegate the typing, not the thinking.

Code is cheap - software isn't

I'm not saying understanding your code stopped mattering. I'm saying writing it by hand has become the exception. A large majority of what ships today - company codebases and side projects - is AI-generated.

Why? Generation got fast and accurate, and the tools now read a codebase more thoroughly than most of us ever bother to.

However, software is still expensive. The real cost of software isn't writing the code, it's everything before and after it.

• Before writing code, people have to sit through endless meetings, brainstorming sessions, UI/UX reviews, and back-and-forth communication.
• After writing code, people have to handle maintenance, marketing, and customer supports.

This blog post is worth reading: Code Is Cheap Now. Software Isn’t.

Engineers who don't use AI will be replaced

We don't only use AI to write code. It's in everything now: planning, investigating, reviewing, even replying to messages.

Not because we're lazy - because velocity is the point. Leadership expects work in parallel, not in sequence. That cuts both ways. More output. Also more pressure and less focus.

If you're holding out, I'd push you to change your mind. Start small: pick one part of your daily workflow and hand it over.

Engineer hiring has changed too

The interview loop that screened for "can you write this function on a whiteboard" is measuring a skill that no longer differentiates anyone. What's left to test is judgement: can you scope a problem, review generated code critically, and notice when the machine is wrong?

Coinbase wrote a year of rebuilding their process around exactly this, and it's worth reading.

Cognitive debt

Cognitive debt in the age of AI refers to the loss of independent critical thinking, memory retention and problem solving skills. That happens when you constantly let AI tools do your mental heavy lifting.

Technical debt is at least visible. You can see it in the codebase and track it in Jira. Cognitive debt compounds silently - and you only notice it the day nobody on the team can debug what you shipped, and nobody is confident enough to change it.

So where is the balance

Use AI for everything you'd be happy to explain to someone else afterwards. If you couldn't defend the change in a review, you didn't delegate the work - just skip it.