Been using GPT-6.1 Sol for a few hours now. No hype, just my experience so far: I’m not impressed. It’s slow, burns through ChatGPT usage, doesn’t consistently follow instructions, and the efficiency just isn’t there. Even if it’s cheaper to run, I’d still take Opus 5.5 or Sonnet 5.5 over it right now. Maybe my opinion changes with more testing, but first impressions? Pretty disappointing.

Dev Day wasn’t that interesting as it was hyped

4 PRs merged. Qostly is live on Vercel, and every commerce gate is still closed. No fake listings. No partnership claims. No data shown before the feeds are ready. An empty marketplace is awkward. A fake one kills trust before launch.

Neon branching is useful for testing schema changes without touching production. Create a branch, run the migration there, test the app against that database, then delete the branch when you're done. Much safer than finding out your migration was wrong after it reaches the database that matters.

Comparing AI coding models is easy until you realise you've spent more time arguing with them than actually writing code. And somehow I'm still choosing which one to open next.

Okay, this is impressive. Just looking back at how far we have come from voice models in just 3 years.

AI: “All tests are passing.” Me: Great. Also me, opening the diff to find out what crime it committed to make that happen:

The biggest lie in AI coding: “I’ve reviewed the codebase.” Five minutes later it’s creating a second auth system. Brother, WHICH codebase did you review? 😭

AI: “I’ve fixed the issue and made a few improvements.” Me reading “a few improvements”: 😐

Nothing humbles you faster than an AI saying you've hit your usage limit when the code it just wrote doesn't even run. YOU'RE LEAVING ME WITH THIS???

Vercel’s AI Gateway saw Fable 5 lose two-thirds of its spend share in August while the cheaper Opus 5 tripled. Turns out the most persuasive benchmark is still the invoice 😭

The scariest sentence in AI coding: "I've also taken the liberty of..." BRO WHAT DID YOU TOUCH 😭

GPT-6 Sol’s API input price doubles once a prompt passes 272K tokens: $2 → $4 per million. Output moves from $10 → $15. So yeah, “just give the agent the whole repo” has a price.

Microsoft just announced Copilot Code and Autopilot. We're going from "AI helps me write code" to "AI builds an entire internal app for my company." Now imagine every department building its own software 😭 Huge opportunity for developers or an absolute maintenance nightmare?

The orbs are boarding. Let’s see what OpenAI actually shows this year.

AI coding hot take: Shipping 10 AI-generated features you don't understand isn't a flex. The scary part isn't AI writing your entire codebase. It's nobody knowing what to do when production breaks at 2am 💀 Are we becoming better engineers or just faster at shipping?

Opus 5.5 is reportedly 40% cheaper per typical workload. GPT-6 Sol has also dropped its API pricing. And somehow I still feel like the real competition is which subscription lets me finish building my app before hitting a limit 💀 Anyone actually seeing longer sessions?