Kapekost engineering blog
Working notes on AI coding agents, AI code review, and shipping real software, from a remote-first consultancy.
A Project Template That Fixes Itself
Every new project starts the same way: GitHub scaffolding, a Claude Code baseline, and instructions that go stale within a week. I built a Copier template that sets all of that up once and keeps itself updated in every project it creates, including the day it almost shipped without the one file that made that possible.
Read →An AI Agent I Did Not Have to Trust
Most AI-agent safety comes down to a system prompt saying "please do not." I wanted my photo library culled by an agent that structurally could not delete anything, right or wrong. Here is the guardrail pattern that made that true, and the day it caught itself shipping something I had asked for, then had to undo.
Read →Automating the Agent Handoff
Long tasks outlive context windows. Instead of compacting and hoping, have the agent write its own handoff to disk and trigger it automatically. Here is the pattern, what other tools already do, and what I would change.
Read →Claude Code vs Cursor vs Copilot in 2026: An Opinionated Take
Three tools, three different bets on how AI fits into engineering work. We use all of them on real client codebases. Here is where each one earns its place, and where it gets in the way.
Read →What AI Code Review Actually Costs
The sticker price is the smallest number. The real cost of AI code review is setup, tuning, and the model bill as your PR volume grows. Here is the honest breakdown for a small team.
Read →Fractional Senior Engineer (UK): When It Actually Makes Sense
A fractional senior engineer is not a cheap full-timer or a body-shop contractor. It is a specific answer to a specific problem. Here is when it fits a UK team, and when it does not.
Read →Why Your AI Agent Config Falls Apart (and Nobody Notices)
Most developers configure their AI coding agent once and never revisit it. The config drifts, corrections compound into rules that no longer fit, and the output quietly gets worse. Here is how to keep it accurate.
Read →Why Engineering Teams Are Switching to AI Code Review
Most engineering teams have the same review problem. Not enough senior engineers, PRs sitting in a queue, feedback quality depending on who picks it up. AI review handles the first pass so the people who matter can focus on the parts that actually need their judgment.
Read →