Staff AI Engineer @brexhq. I like to build cool shit and ride bikes 🏍️ Leading our new AI Assistant to help customers automate their work
Sao Paulo, Brazil
Joined March 2016
- Tweets2.9K
- Following1.3K
- Followers2.9K
- Likes6.8K
Maybe I’m just stupid but I’m having a hard time to adapt to Grok Bot’s UX. I have a few bots created by Dr Egg, like “shipper”, “prod-watcher”, “reviewer”, etc. However, this always-on experience doesn’t seem to click for me - when I’m working on something, I often have a handful of side quests I want to work on and start new sessions for.
Sending them to the agent who’s working on the task feels wrong? Ok fine, you can have a side quest agent. But what if I have 3 side quests? Do I just queue the asks? Or should I have a few idle agents to spread work across?
I don’t know, I must be doing something wrong. The mental model is pretty different than regular Claude/Codex single-session-per-task approach.
What am I missing?
Since everyone is complaining that iPhone Duo lacks great apps, here's a skill to migrate your app today: github.com/MarceloPrado/skil…
It learns what the app is for, audits your screens, proposes a layout following Apple's HIGs and builds it, proving it works with before and after screenshots.
Enjoy! Feel free to contribute
My new workout app MCP is sooo good. It fetches your latest workout from iPhone/Watch, read your avg/max BPM + notes and help you adjust it automatically.
It's like having a superstar personal trainer. Getting close to launch, waiting for Apple, Anthropic and OpenAI reviews
One of the thing I love most about @poteto's pstack is the fact it names the principles shaping its response.
It gives me a higher confidence certain task is trending on the right trajectory.
I vibed an explainer video about the key @Cloudflare primitives. I knew very little about the company beyond their CDN offering and the video helped me realize all the cool things I was missing.
I hope it's helpful to you too, let me know!
I'm trying the "Be more ambitious" approach from
@felixrieseberg and asked Opus + Fable to act like a professional game designer.
It rigged the crap out of my models, and built many verification tools (deterministic and judge-like). But my simulated world is not fully there yet.
After LOTS of turns, it finally got the dumbbell curl in a good spot. But it doesn't generalize to other exercises like I expected - 2nd video shows what it outputs during the first few passes.
It's also really token hungry. My end-goal is a skill where I can ask it "record a muscle-up in 3 angles, female model" and produce production-quality artifacts ready for my app.
Long way ahead!
Building a new workout tracker app with Opus. Current milestone: rigging my 3D character models to record the exercise poses.
I have no idea if this will work lol
What I love about Opus 5.5: I can finally build the wedding website of my dreams.
Inspired by Shopify Renaissance's release - to me, the most beautiful website ever published.
I think my fiancée will 🤯
Thanks @AnthropicAI
I think I've never been more frustrated with a vendor than Twilio. @zenorocha / @vcapretz you guys need to dominate this market seriously
Thanks Peter! no drops in coverage either! amazing.
OpenClaw deleted around 400k LOC of its own tests without much change in code coverage. Modern models just love writing tests for every tiny change, even if they aren't useful. This skill helped. github.com/openclaw/openclaw…
nat.org is full of gems. Here’s one that resonated with how my team has been operating our product:
•Smaller teams are better
â—¦Faster decisions, fewer meetings, more fun
â—¦No need to chop up work for political reasons
â—¦No room for mediocre people (can pay more, too!)
â—¦Large-scale engineering projects are more soluble in IQ than they appear
â—¦Many tech companies are 2-10x overstaffed
I’ll die on this hill: there’s nothing better than a small team with large ownership
Luna ❤️
Please welcome GPT-6 Sol and GPT-6 Luna to the GPT-6 universe.
GPT-6 Sol and Luna build on the advances behind GPT-6 Astra, bringing much of its strengths into faster and more affordable models to support work at scale.
We’ve also made caching and inference more efficient, and we’re passing the savings directly to you: 50% lower API prices for Sol and Luna compared with GPT‑5.6 promotional pricing.
One of the things I’m most excited about Jev is using it in CI. You can now enforce all your team and agent’s conventions as prose and feed into Jev checkers
Will make it impossible to commit code out-of-spec. Style guides will become style linters.
Somewhat of a different approach than @VictorTaelin Bend2, which tries to make bad/faulty code impossible at the compiler level.
Excited about both directions honestly