Blog
Thoughts on software engineering, tech reviews, and insights from the ever-evolving world of development.
I Automated the Work. The Work Moved.
Every.to studied its own operations and found that deeper automation doesn't shrink the human workload — it relocates it. That matches what my own pipeline keeps teaching me.
Transcribing Isn't Building
A writer keeps finding vibe-coded apps that leak user data — without even looking for them. The builders aren't careless. They don't know the question exists.
Fluent Code, No Taste
Teams are generating thousands of lines a day with nobody left to read them, and the agents writing the code don't know what good looks like. That gap is the real bottleneck now.
Where Do Senior Engineers Come From Now?
AI is best at exactly the tasks juniors used to learn on. If the bottom rungs of the ladder are gone, growing judgment has to become deliberate work — it won't happen by osmosis anymore.
OpenAI Bought the Tools You Type Every Day
uv and Ruff live in millions of Python workflows, and now they belong to OpenAI. When a model company owns your toolchain, the interesting questions start.
The Market Read the Release Notes
Software stocks shed $285 billion after a model release. Panic aside, investors reached a conclusion engineers should sit with: agentic AI will eat categories of software work.
The Boring Kind of Important
Unified LLM APIs, middleware, telemetry, structured outputs — .NET AI Essentials is plumbing. Plumbing is what decides whether your AI feature survives contact with production.
Check Model-Market Fit First
Before you interview a single user, test whether any model can actually do the task reliably. A simple checkpoint that would have saved a lot of dead prototypes.
Both Headlines Are True
Anthropic says AI built Claude Cowork in under two weeks. DHH says AI can't match a junior programmer. The paradox resolves once you notice they describe different kinds of work.
Forty-Eight Hours or It Doesn't Ship
Boston Dynamics says Atlas must learn a new factory task within 48 hours to be worth deploying. That's the most concrete definition of enterprise-ready AI I've heard yet.
Types All the Way Down
End-to-end static type checking from PostgreSQL to TypeScript means a schema change can fail a frontend build before it reaches production. Quiet news, big deal.