Arthur Brito

forever enthusiast. builder by design.

[Arthur Brito @arthurbritom]

@LukeberryPi @acgfbr Luke, acho que vais gostar muito disso aqui, eu ainda uso o frontend-design quando quero ser mais direto, mas quando é pra algo maior, tô gostando demais do impeccable, é uma stack bem mais completa de steering de UI prós modelos
https://t.co/fbDq9sSf4X

[Arthur Brito @arthurbritom]

@acgfbr Isso aqui ajuda demais, tem inclusive direcionamentos específicos pro Codex
https://t.co/fbDq9sSf4X

[Arthur Brito @arthurbritom]

@mattpocockuk That was my experience too. The model only invoked the grilling skill, not the domain-modeling, even though the instructions explicitly say to read it

I ended up with a very good Wayfinder containing 18 resolved sub-issues, but only one ADR and very little shared language

[Arthur Brito @arthurbritom]

The main thing was the grill-with-docs skill, which offers a clear method for better, cleaner collaboration with AI. It is cost-effective, leads to greater alignment, and purposefully encourages people to have a better understanding of what they are doing, directly opposing the idea of using AI to generate things you don't actually understand.

The other key aspect was the documentation part, specifically regarding ADRs and CONTEXT dot md

[Arthur Brito @arthurbritom]

@benln Hey Ben, would love to get in!

[Arthur Brito @arthurbritom]

@LukeberryPi Estás usando o /to-prd no fim? Vem fazendo muita diferença pra mim. As vezes uso o /to-prd e /to-issues inclusive, pra fazer com que os agentes implementando as issues se mantenham na "smart zone" que o Matt chama, até 120k tokens. Implementando por issue fica massa

[Arthur Brito @arthurbritom]

@jullerino Codething 👀

[Arthur Brito @arthurbritom]

@LukeberryPi O faz muuuita diferenca pra mim é que muitas vezes eu concordo em partes, ou realmente discordo, e o entendimento compartilhado vai sendo construído de uma forma muito massa

[Arthur Brito @arthurbritom]

@joshwoodward @VictorTaelin

[Arthur Brito @arthurbritom]

@mattpocockuk @facmartoni It's almost always with Codex. I feel like you have to steer it more to group similar topic questions to get better behavior for the skill. I know that you primarily use them with Claude Code, but how often do you use them with Codex, Matt? Just out of curiosity

[Arthur Brito @arthurbritom]

@ClaudeDevs Interesting that it's that now it's very similar to @mattpocockuk /handoff skill

[Arthur Brito @arthurbritom]

Everyone like: WOW

[Andrej Karpathy @karpathy]

Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.

[Arthur Brito @arthurbritom]

@arnestrickmann One of the best, https://t.co/Kw6pMBTuek , from @theo

[Arthur Brito @arthurbritom]

Using @mattpocockuk skills, I feel so in the loop, having more ownership of the solution, and at the same gaining prpoductivity with. Grilling, PRD, issues, handoffs, etc

As a bonus, I feel a sense of elegance and beauty of software engineering what is being done. Really happy

[Arthur Brito @arthurbritom]

@LukeberryPi Interessante que é trazer a capacidade de multi agente que todas as GUIs muiti agentes, que a do próprio Claude desktop tem, pro terminal. Pra vc ver como o ideal mesmo no mundo atual são essas interfaces, mas conhecem tanto o Claude Code pela CLI, que faz sentido trazer

[Arthur Brito @arthurbritom]

Sim, mas pelo que oercebi o Codex não injeta diretamente no prompt, como o Claude Code faz por exemplo, e sim referencia e logo depois lê, por isso as vezes ele se confunde como se a referência fosse parte do contexto ao invés de uma instrução específica. Em geral funciona, mas já tive esse problema mais de uma vez

[Arthur Brito @arthurbritom]

@aye_aye_kaplan Do the thing

[Arthur Brito @arthurbritom]

I'm loving Matt's crusade on making the skills we all need for software engineering, based on literature principles, and know also focusing more on the productivity side

Every day he's releasing some nice

[Matt Pocock @mattpocockuk]

Fuck I think I've figured out good creative writing with AI

It's really fun

[Arthur Brito @arthurbritom]

@w0rddriven @VicVijayakumar @mattpocockuk How has your experience been using it with and without plan mode?

I found that without plan mode, it tends not to use the tool to ask user questions and, counterintuitively, asks more complete questions. I just still haven’t tried plan mode while telling it not to use the tool