My @limitlessai pendant was delivered yesterday & it already paid for itself while brainstorming UI copy. Team: "What did you just say? It was perfect." Me: "No idea, I was just throwing out ideas as they came to me, lemme check Limitless...here we go." * mic drop *

Running 2 robot vacuums, a dishwasher & a washer all at the same time feels close to a desired future of devices doing my bidding - All I really need is a humanoid robot to help out & I think I could be satisfied.

There’s nothing quite like getting a surprise email from OpenAI that your account has been funded, only to realize that at some point Cursor changed the model in your code to a more expensive one and just cost you another $50 for no reason

Recorded my week of conversations with my Limitless pendant & created a workflow to summarize everything into an automated weekly post—with a 4o hero image of course. Not bad for v1, the tone still needs some polish & less cheese though. andrewdraper.com/lifelog/2025-0… /cc Dan Siroker

Protip (is that still a thing?): When AI can’t quite get you the result you want, sometimes switching tactics & asking it to write code to accomplish what you want is a faster, better way of getting what you want.

I've been a fan of Jake Knapp's Design Sprints for awhile now, & recently realized that AI could be an amazing force multiplier for them. What used to take 5 days I can now do in less than 12 hours 🎉. This is the basic flow of what I've been doing: andrewdraper.com/projects/ai-de…

Feels like ChattyG’s flattery goes to 11 on this one…

Nothing quite like shaming the AI into submission (I swear I usually use please and thank you with it but some times...)

If you’re in San Francisco and don’t tweet about it are you really in San Francisco? I’m here all week (try the roast beef?)

I wish all my morning emails, especially the newsletters, would just go to NotebookLM and generate a podcast I can listen to each morning rather than having then all stack up unread for days (let’s be honest, weeks)

Who’s vibecoding hasAnewLLMmodelBeenReleasedToday[dot]com?

Working on something yesterday & was very picky that it match my designs. Tried 3 models in Cursor: - Sonnet 3.5/3.7: ignored the design, but functional - Gemini 2.5: better & functional but lots of work still required - GPT 4.1: majority of layout correct, functionally correct

It’s only been 90 minutes but so far Codex truly does live up to the CodeBro™ hype.

One of the things that keeps me using ChatGPT/Codex more than other tools is how seamlessly I can jump between my phone and computer and do the same thing.

It used to be an exercise in futility, but I’ve finally found “somebody” who can make sense of local government websites and find the actual information I want, they’re name is Perplexity

Clode not clod

Shared memory across LLMs needs to be solved (and embraced) asap. It’s exhausting starting from scratch with each siloed chat instance in every app.

Does AI dream of zero-shot sheep?