Derniers Articles

Fifteen Claude Code projects, thirteen complete matchups: runtime, API-equivalent cost, tests and limits.

Dots, models, Agents API, Codex and ChatGPT: fifteen DevDay announcements and the limits of their availability.

Codex is now my default, while Claude Code still leads on specific missions. Here is why the balance shifted.

I gave Claude Opus 5.5 ten real projects in isolated environments. Here are the protocol, results, cost and limits of the experiment.

Learn how Jev turns text or JSON state into typed answers with probabilities, where it fits in an AI architecture, and which limitations matter.

A look inside VillaSlot and ShipSaaS: multitenant architecture, clear application layers, data isolation, and the quality loop that keeps AI agent output maintainable.

The workflow behind VillaSlot: product framing, 20 user stories, mockups, a 9-hour-20-minute autonomous run, and proof for every feature.

The full workflow behind VillaSlot: product framing, 20 user stories, mockups, a 9-hour autonomous run, and strict QA.

How I organize Claude Code and Codex with Herdr on a VPS, then use Hermes to check remote work without confusing availability with autonomy.

A practical method for isolating branches, worktrees, ports, databases, servers, migrations, test data, and E2E browsers when Claude Code and Codex work in parallel.

GPT-6 Astra combines long context, tools, computer use, and strong cyber capabilities. Here are the evidence, costs, and limits behind the scores.

Inside AgentsMail: controlled templates, batch sending, deliverability, per-contact history, and the engineering method behind it.