An agent can be done. The proof can still fail.

Finally got Claude Max 20x. And of course, I still have Cursor credits expiring today. So now I'm trying to use both :))

Working on my AI product listing tool again. It can write a great description of something the product doesn't actually have. Very convincing. Completely useless :))

I had one AI review some work. Then another AI review the first one's findings. Guess who ended up reviewing both of them :))

AI can write a product listing in seconds. Getting it to write one that actually helps sell the product, without making stuff up? That's a different challenge :)) Working on that now.

My laptop has basically become a desktop at this point. WSL, two coding windows, tests running all day. Unplugging it is mostly a formality :))

GPT-6 rocks. Period :))

Curious what people are actually using GPT-6.1 Sol for now. Coding? Research? Agents? Something else?

I really need to stop improving things that are already good enough to ship :))

Computer use is in the Agents API now. Anyone already using it for something real?

I keep ending up with Codex, Claude and Grok open at the same time. This is getting stupid :))

OpenAI just simplified API usage tiers from five to three: Build, Launch, Grow. Small update, but thank you. I already spend enough time checking limits :))

Too many projects open at once lately. Trying to actually finish them now :))

Turns out building a naming engine was easier than finding a name I actually want to keep :))

Been buried in builds lately. What are you working on right now?

One of my products is basically sitting there waiting for a name. So I built a naming engine to solve it. Now I'm improving the naming engine instead :))

I switch between different agents all the time while building. The annoying part isn't switching agents. It's making sure the project doesn't lose the plot every time you do. I got tired of dealing with it, so I built something for it. More soon.

340 followers now. Still small, but it's starting to feel like there are actual people here instead of me talking into the void :)) Most of the growth came from just talking to people building real things, not trying to manufacture viral posts. Now I just need to give everyone a reason to stick around :))

I'm at that annoying stage where the product is mostly built, but somehow the launch still has a whole second project hiding behind it :)) Still left: final name license, privacy and security docs and demo install, update and uninstall rollback support one clean release rehearsal Building the thing really is only half the job.

Made a change this week that looked better in the internal numbers and worse when I actually used it. Reverted it. Easy trap with AI products: getting better at what you can measure instead of what a person actually wants.

I'm changing how I think about one product I'm building. “AI writes your product listing” isn't that interesting anymore. There are already plenty of ways to generate copy. The useful part is catching what's unsupported, missing or weak before it goes live. Writing is the easy part. Checking it properly is harder.