Into browsers and agentic systems. Building https://xitter.cf/t.co/4HvWdzHsxi Into film photography and web performance nerd. 📸 dev.aperture (insta)

Stockholm, Sweden
Joined October 2008
I have a 'neuve' side project and I'll start writting more about it here. This time the writeup brings some of my experience when developing the browsing aspect of, I stopped telling AI agents what to build. Instead, I carefully created an experiment on the problem. No shared context, no instructions and no specific method in mind and let them figure out how to drive a browsing session based on a goal independently. They converged on similar patterns I've seen other browsing agents use under the hood. But some of the findings and the process I took taught me quite a lot.
1
2
2
7
1,482
Vinicius Dallacqua retweeted
We have presents for you! 🎁 - A new WebMCP Skill with best practices for security, development, & consequential flows - WebMCP Types: TypeScript definitions, with type inference from InputSchema Give em to your agent, it helps scaffold out WebMCP so fresh and so clean clean
39
126
16
871
71,106
Vinicius Dallacqua retweeted
Preparing for Launch Day in SF tea houses. Ship all the features you want. People still need to understand why they should install the thing before you can consider it fully shipped!
1
1
6
541
This is the “Drake vs Kendrick” for devs 🔥🔥
Replying to @thsottiaux
yeah i can see why shipping is such a big event for you
1
67
Time to go (tell your agent to) learn Effect!
Effect v4 is here. One ecosystem. Zero dependencies. The next chapter of Effect and the foundation for building reliable software and AI agents in TypeScript.
1
1
156
Vinicius Dallacqua retweeted
I feel similarly when I see y'all generating slides and "artwork." So cringe. Everyone outside the bubble is not impressed. You're impressed because you never had the skills nor appreciation. And you come in like everyone should be impressed?
(dude who’s not watched a movie/read a book/listened to an album in 6 years) “the movie/music/literature industry is so screwed, you can now generate all of it with a single prompt!!” that’s what a lot of you sound like to me. it’s annoying as hell. you are but vessels for LLMs.
6
2
35
3,636
Building and programing is about understanding a problem and its principles, the risk of using AI without that guiding your work is that you don't even know what is being done and will get easily fooled by shining lights and baubles. Get your principles right before delegating the building parts to agents.
I work in CRISPR discovery research. This is one of the most exaggerated nothing-burgers ever and would be laughed out of the room if a human scientist attempted to publish something like this. (cont.)
42
One of the pieces of advice I used to give before AI as a staff engineer still holds true in a way. "The best thing as a dev you can practice is detachment, the code you are most proud of today is tomorrows legacy to be removed." In a way this shielded me of being too precious and exploring early.
Best part of AI is shattering dev's egos, the worst part of this industry. You are not special. Your code actually sucked like everybody else's. You were never the smartest person. The things my sales guy BIL and designer brother are doing is beyond anything I have done.
75
Vinicius Dallacqua retweeted
I always thought it was clear that browsers would be deeply integrated in to a chat/co-work/agent and that the companies building them were making a mistake to build a dedicated browser. The traditional way of using the web has ossified to the point where browsers eek out points per year in market share from each other while not changing much. At the same time, its amazing to see WebMCP being adopted by these Agent Harnesses - we have a web of linked functionality now - but make no mistake, this is the start of a full assault on the browser as the entry point to your work-day. As much as I love web browsers, I think this actually a very good thing for the web and will strengthen it further as one of humanities greatest inventions. People will be able to publish and share their ideas and work, and have everyone discover and work with them at rates that we've not seen in years. Browsers need to get better. The web platform has a chance to evolve too and I hope the agent harness companies can get more involved in shaping web standards. You won't find this in Google search (I don't know why my site's still not visible), but you can also read more on my blog: aifoc.us Also ht to @rachelnabors who has been saying similar thing for a long time too.
1
4
1
20
2,777
Vinicius Dallacqua retweeted
The WebMCP Challenge is here. We’ve teamed up with @ChromiumDev, @CloudflareDev, @ShopifyDevs, @vercel, @render, and @Netlify for a 10-day hackathon. Up for grabs: $35,000 in cash prizes, Codex Micros, ChatGPT Pro subscriptions, and more prizes from our supporters.
146
308
151
3,737
1,916,313
Vinicius Dallacqua retweeted
Where are my creative design-gineer sisters at? NOW IS YOUR TIME
Replying to @OpenAIDevs
Tune in to the kickoff livestream today at 3pm PT for product demos, example builds, prizes, and more details on how to enter. xitter.cf/i/broadcasts/1qGoNYbWd…
2
2
7
2,348
Vinicius Dallacqua retweeted
Tune in to the kickoff livestream today at 3pm PT for product demos, example builds, prizes, and more details on how to enter. xitter.cf/i/broadcasts/1qGoNYbWd…
6
16
1
211
37,194
Vinicius Dallacqua retweeted
Huuuuggeeee
Replying to @OpenAIDevs
We’re adding support for WebMCP in the ChatGPT desktop app’s built-in browser and ChatGPT Sites. When you visit a compatible website, ChatGPT or Codex can automatically use it to complete your task. Update to the latest version of the ChatGPT desktop app, then just ask Codex to create a WebMCP-enabled app and deploy it to Sites.
4
4
52
21,424
Vinicius Dallacqua retweeted
introducing anydoc now your agents get 100x faster local parsing for pdf, docx, pptx & 10 more formats - sub-5ms md conversion - 500 docx files in 1.7s - top quality across all 13 formats - rust-based - open source already powering @firecrawl /parse github.com/firecrawl/anydoc
293
722
133
9,036
2,549,530
Just as a reminder for you to read the code, GPT 5.6 Terra just tried to BS me with a .patch for better-auth instead of doing the correct thing, and a 5.6 SOL reviewer shrugged it off. Always. Read. The. Code.
1
190
GPT 5.5 is so BAD at UI/UX and design software it's appalling. I generally prefer 5.5 for hard engineering problems and reviewing its code is generally not a bad experience. But give it ANY interface task and you are in for a world of pain if you minimally care.
2
158
Vinicius Dallacqua retweeted
Did you know that the .png format was created out of spite? Or that .jpeg just discards like half of the colours in your image because you won't notice? To celebrate last week's launch, I published a chapter on image compression. makingsoftware.com/chapters/…
47
297
20
3,563
156,044
I'm building a context engine for reviews and used @OpenAI Codex to its limits to stress test the graph and responsiveness of the commands on a large changeset whilst keeping the outcome relevant. The goal feature is super well though out, and with the context engine I'm building for the reviews and sourcing the relevant information for validation and correctness it managed to keep things on track. Now I need to wait a bit to reset my usage limits 😅
1
169
Maybe @sama can help me out with a quick reset so I can get it to have another pass 🫡. Stress testing the checkpoints so the graph is built progressively uppon git changes.
49