Pinned Tweet
Se você não assina meu substack, assine. Geralmente posto coisas de valor lá.
Eu não posto muito, mas quando eu posto é 👌
Falando sobre Spec Driven Development, Harness Engineering e como sobreviver no caos jpmartins.substack.com/p/fal…
JP retweeted
I like Grok Bot. it’s dead simple and it just works. but let’s be real, it’s getting harder to justify choosing it over Hermes.
> Grok 4.7 is just too far behind
> usage limits are now tighter
> the pricing isn’t so beautiful either
and with all the talks about using it for coding/engineering I don’t see a world where anyone can use it at this pricing without unlimited tokens.
full disclosure, I am still using Grok Bot on a daily basis, mainly for its reliability and simplicity. I think the engineering behind it is great, but the business side around it, not so much.
Grok Bot’s usage and pricing is damaging a very well designed product (kudos to the SpaceXAI engineers for it)
if an app locks me into one model with no option to switch to a cheaper/smarter one, I at least expect the provided model to be exceptionally smart or cheap.
right now, Grok is neither, and the price needs to catch up to it.
I'm not looking for a personal assistant setup anymore.
I'm looking for a setup that will run all aspects of a project autonomously (or almost).
I want it doing my side projects while I'm working in my main job 😂
As vezes eu fico triste.
Ai eu lembro que acharam que mudar pro Paraguai seria uma boa ideia KKKKKKKKKKKKKKKKKKK
After 2 weeks of intense playing around with Grok Bot I can confidently say: it’s unusable right now.
My limit is vanishing super fast and the model looks stupid.
Moved from awesome tool to stupid tool really fast.
Time to test new things.
JP retweeted
Virar a noite trampando pode ser diferencial pra resolver um problema.
Pode ser uma resposta à crise e é um recurso importante pra alavancar/salvar a empresa num momento especifico.
De toda forma, gestão a longo prazo depende de sono, saúde física e mental do colaborador.
Não dormir é por N fontes científicas considerado como grande detrator disso.
Dito isso é um plano valido de emergencia.
Sinceramente, a era de hacking com agentes me assusta.
Eu realmente acho que não estamos prontos pra lidar com esse problema.
Dito isso, vou tirar meu dinheiro do banco e comprar Nelore em Minas. Quero ver o Codex pegar meus bicho do pasto.
Turns out I can't use Grok Bot compute for long running tasks. The VM kept dying on me.
Not sure if this happens by design because the bot is asleep while claude and codex works or if I just hit some compute usage limit.
For the first time I’m running truly autonomous fleets with 10+ agents and the amount of fuckups they do is just incredible.
Harness matters, safety matters, sandbox matters and self-correcting loops are crucial.
The hard engineering just moved from a place to another.
Orquestra v0.1.0 is out: an open-source bot that runs Claude Code + Codex as a fleet, with parallel devs and cross-review.
Coming next: a self-improvement loop. Every 2 days it reviews its own runs and turns mistakes into rules in its skills, not logs.
github.com/jpvmm/orquestra









