234 Photos and videos
In light of what happened, I'm doubling down on skills like /improve.
A frontier model got pulled. If it happened once, it's gonna happen again. Fable today. 4.9 tomorrow or maybe gpt 6 one day.
So, treat intelligence as borrowed. Drain intelligence when it's available. Build a catalog of plans today. Then implement later with a cheaper, open source, or a model you control.
Build the backlog now.
github.com/shadcn/improve
137
407
6,969
302,567
Tute Costa retweeted
Jun 9
The interesting part is not the bug.
It is watching Codex:
- read the right product spec
- follow the repo startup checklist
- choose a focused test
- make a narrow change
- run targeted verification
- explain why the broader suite was not needed
1
1
62
Jun 9
The path from human-gated to autonomous isn’t a single switch. It’s a gradual transition backed by data. The agent earns trust through a track record of correct decisions, verified by the approval log.
damiangalarza.com/posts/2026…
27
Jun 5
My story on five AI agents on my iMac. One Telegram contact routes everything. $60/mo in API usage.
Calendar, email, invoicing, smart lights, even drafting blog posts. And I stopped fighting Siri not getting Argentinian Spanish.
withatwist.dev/one-telegram-…
1
1
59
Jun 5
Preparing a blog post on Mastra agents, I ask Claude for a review:
> Lines 93–94 ("I was ready to add some more responsibilities..."). Throat-clearing before you get to the actual invoicing problem.
Gotta love it.
33
Tute Costa retweeted
May 5
AI-run business experiments are interesting and fun up to the point where they waste the time of humans who haven't opted into the experiments - I think they need to keep their own human operators in the loop for outbound actions that affect other people simonwillison.net/2026/May/5…
36
11
136
24,674
Tute Costa retweeted
Apr 30
Front-load context on your first turn.
Opus 4.7 stopped reading between the lines. 4.6 would interpret a vague prompt. 4.7 won't.
Name the files. State the constraints. Define "done." Stop there.
1
1
49
Apr 27
Let's enable MFA for all Ruby gems
thoughtbot.com/blog/lets-ena…
Thank you @matheusrich for improving the Ruby ecosystem and community!
1
30
Apr 23
I just open sourced my AI toolkit, mostly for Ruby on Rails development:
2
50
Tute Costa retweeted
Apr 21
thank you cursor 3
this morning i woke up to 5 PRs that i kicked off with cloud agents yesterday night.
bugbot had reviewed all of them, suggested fixes, and auto pushed it while i was sleeping. then, another agent tested the changes to make sure they were functionally correct
can also just spin up a bunch of agents on the same problem, see whicih one fixed it the best way
i can never go back to not using cloud agents anymore, its changed the way i approach problems
37
11
585
37,356
Apr 17
Fond memories of my college project reviving "old" desktops with Xubuntu. Configuring them without Claude was rough. And gotta love Omarchy, see blog post below.
1
68
Tute Costa retweeted
[ENG] Now on stage! @matheusrich — "Production Data Doesn't Have to Be Scary"
[PT-BR] No palco! @matheusrich — "Production Data Doesn't Have to Be Scary"
#TropicalOnRails
2
9
348
Tute Costa retweeted
[ENG] Day 1 of Tropical on Rails 2026 — the energy, the talks, the community, all in one video.
[PT-BR] Dia 1 do Tropical on Rails 2026 — a energia, as talks, a comunidade, tudo num vídeo.
#TropicalOnRails #RubyOnRails #SamedayEdit
3
13
416
Tute Costa retweeted
Apr 7
Your AI agent ran without errors. The output looked reasonable. So you shipped it.
That is not testing.
New video: How to Know If Your AI Agent Actually Works
youtu.be/Kleu3ROhpvY
1
2
2
137
Apr 7
Best piece I've read on building real software with AI. A full SQLite toolchain, with honest and objective breakdowns of where agents helped and where they made things worse. See within:
1
1
57
Apr 7
36