ai
Generation got a hundred times faster and reading did not. What that actually breaks in code review, why spec-driven development collapses into a compiler, why clean code matters more when agents read it, and what to build instead: an ontology of your own development process.
ai
The future of programming is not more code. It is a knowledge graph, an event log and agents you route in code, not in prompts. My opinion on which skills pay off this year, the mistakes that sink agent projects, and why a small team with legible data beats a big one.
ai
Jev, TypeSafe's System One model, returns typed decisions with calibrated probabilities in 70-500ms instead of generating text. Here is what the "cannot hallucinate" claim actually guarantees, where it breaks in production, a forty-line Ruby client, and how to test it on your data.
AI writes confident Rails code that quietly breaks production standards. Build a quality workflow that catches it: Lefthook gates, Strong Migrations as a hard line, N+1 detection with Prosopite, database checks in layers, and CI that mirrors local hooks.
AI agent architecture comes down to three strategies: single agent, sequential, and parallel. When one big prompt loses the plot, how sequential chains buy reliability, where parallel agents crush latency, and the pitfalls of each before you commit.