A client hired me to build a landing page with a booking integration. Built the whole site with Claude Code in under a day. Then they put me in touch with the company providing the booking API. First thing the guy asks, unprompted: “Did you build your site with AI?” Then he tells me he tried building a website with AI once and it was so bad he’ll never do it again. I say nothing. I just send him the site I made for the client. He sends me his back. One look and I know exactly what it is, pure AI slop. the kind of thing you get from typing “make me a website” and shipping the first output. Made with Claude, ironically. I didn’t say a word. Not my job to judge. We talked through the integration and I kept my doubts to myself: if this is how they build a simple site, what does their actual backend look like? A few days later the client sets up an in-person meeting we all live nearby. I finally meet the guy behind the API... I ask him about his architecture. How he handles X, how he structures Y. He barely answers. Turns out he builds “most of it” with Composer in Cursor. Pure vibe coding. So I go home. I now have their app url, so out of curiosity, I run a few tests... 15 minutes later I’m looking at their entire financial records and every client booking in their system. No auth. Nothing. Wide open. The guy who’ll “never use AI again” because it’s unreliable shipped a production system leaking every customer’s data. What would you do?

Agent Spooner ?

Is it just me, but I find Opus 5.5 in the desktop app look like more opus 6. In the cli, it feel less more powerful. Just me?

What's going on with gpt-4o cult?

I can smell new GPT model dropping soon, when Astra behave like a junior Pakistanese developer from fiverr

I'm blown away that Google put such a useless Cloud Assistant... Not only GCP require some nerves and patience to navigate. But the AI assistant does not even use their most up to date model. @googlecloud

Most of my apps have been on the OpenAI API because Claude’s API was usually too expensive to justify. Anthropic just rolled out monthly Claude Platform credits for Max and Team plans , so I moved two of them over. Let's see how they hold up in production. Who else is making the shift?

One client app used to take 42 minutes to fully deploy. After some work with Astra, it’s down to 16 minutes. Image build and publish: 7 min 15 s → 2 min 52 s. Total cut by about 63%.

Rule I always use with AI APIs: load $50–100, turn auto top-up off, one project one key. Today an agent update went into a retry loop and ate $50. Fine. If auto top-up was on that same loop could’ve been hundreds. Small prepaid balance is the cheapest insurance you can buy.

Spent the last 3 hours finishing a prod server setup for a client workflow: Codex → Cmd+C Claude Code on the server → Cmd+V repeat I should’ve just given Codex SSH access and let it cook instead I became the world’s most expensive clipboard between two AIs client still thinks this is devops