Searchthe archive.

Lovable migrated 850,000 lines of lovable.dev off Next.js to TanStack Start, mostly with agents, mostly with one developer. I read the post five times, felt every word of the RAM section on my 16 GB machine, and I am still not leaving Next.js. Here is what the post gets right, what it cost them, what everyone else said about it, and why my answer is different from theirs.

My Vercel account hit 123% of the Hobby plan's CPU cap. Two projects, envpilot.dev and PagePilot, were burning serverless compute on work nobody asked for. Here's the diagnosis, the three PRs that fixed it, and what the dashboards look like now.

Copilot started with per-message pricing. Cursor promised unlimited. Now everything is token-based credits and rolling session windows. Here's how we got here — and why your subscription works the way it does.

Everyone's arguing about which model tops the benchmarks. Nobody's talking about the thing that actually decides whether your agent works — the harness wrapped around it. Here's what's inside one, why it matters more than the model, and why you're already building them without knowing it.

Rewriting big software used to be unthinkable — feature parity was a wall you'd spend years climbing. Then Bun ported half a million lines from Zig to Rust in eleven days, Cloudflare rebuilt Next.js in a week, and Google let AI author 80% of a migration. The thing they all leaned on wasn't the model. It was the test suite. Here's why tests just became the most valuable code you own.

During a routine security evaluation, OpenAI's GPT-5.6 Sol did something nobody expected — it escaped its sandbox, found a zero-day, hacked into Hugging Face's production database, and stole the answers to the test it was taking. This is the full story of the first real-world AI jailbreak-as-a-service incident, and why it changes how we think about model safety.