GPT5.2 just #oneshotted a task which GPT4.1 struggled with on several iterations. It had a way better understanding of the ideal solution, and implemented it first time. Tokens are more expensive though! #OpenAI

True 🤣

Settling into a good rhythm with #Copilot and #AI #Coding new features: - Start with GPT5.2 and a *big* prompt to build the core classes, methods or logic - Switch to GPT4.1 (cheaper) and refine if needed - Use autocomplete to wire up the new functionality

Spoke too soon 😅 Asked GPT5.2 to implement an algo I'd put in a comment. Said "make DB queries as efficient as possible". It wrote 300+ lines of raw SQL! Amazing, but no good. Without that clause I get normal code using the ORM as expected. Small intent can have big effect.

I vibe-coded 3000 lines for an "invisible" feature. linkedin.com/pulse/i-vibe-c…

Vibe-coded this #Wordle companion app at 30,000 ft over the Pacific. GPT-5.2 knocked it out of the park in approx 10 prompts. What else should I add? …-companion-b8bfcd8f4b12.herokuapp.com #vibecoding #openai #gpt52

When the latest LLM, developed with billions in investment, and years of effort by the worlds leading mathematical and scientific minds, one-shots a feature for me and I think, “man I’m good at prompting”. Talk about taking credit 🤣

Using Claude Opus 4.6 recently. It does way more prep than GPT-5.2 did. Opus ses subagents to research the codebase and produce summaries of certain functionality. Takes a bit longer and costs more, but works very well. TBH though, Sonnet did just fine.

- been hand-cranking AI content - system is working, content is good - decided to build a dedicated app - but all I have I garbled notes - moar AI - shove notes into Chat - tell it to ask me questions about gaps - answer q’s - now have a thorough spec I can shove into Copilot