Ten reasons why we prefer conventional code over AI agents
We are an AI-native company. That is exactly why we know where the line is. The question was never agent or code. It is whether the path was knowable before you started.
Your AI agent is a literal genie
AI is brilliant at how to build something and still struggles with what to build. An appreciation for the engineers, architects, and designers who decide the what, and hold the line on everything the model ignores by default.
The DSL moment
In the late 1990s I couldn't convince anyone that DSL was worth paying for. Then they used it. Agentic AI is sitting in exactly that spot right now, and the shift is experiential, not logical.
Just Claude it
A friend said it without thinking. It's the whole mindset shift in three words: stop asking AI how to do the work, and hand it the work.
How good are coding agents today? What do the benchmarks tell us?
A 93% on SWE-Bench Verified does not mean an agent does 93% of an engineer's job. It means something much more specific, and reading the specifics correctly is the whole game.
Stop asking AI questions. Start having it do things for you.
Most people who use AI are stuck in transactional mode: type a question, read the answer, close the tab. The leverage is somewhere else, and almost nobody is there yet.
Control of your codebase is an illusion
Every engineering leader has said it: 'if we had more time to test, review, and define requirements, we wouldn't have these production issues.' AI doesn't change that problem. It changes the economics of solving it.
AI should never be the reason you're laying off people
Over 100,000 employees were impacted by AI-attributed layoffs in 2025. In most cases, AI is either the excuse, the misunderstanding, or the tell that imagination ran out.
Working with global teams: top 10 rules
18 years of managing offshore teams distilled into ten rules. They apply to any team, but they matter most when distance amplifies every mistake.