CTO @ CompleteCoach.fit / AI engineering / Data / Product Management
Melbourne, Victoria
- Tweets 457
- Following 469
- Followers 64
- Likes 662
Michal Szalinski retweeted
🚨 ESTO ES UNA MINA DE ORO
Anthropic ha publicado +400 agentes y skills listos para usar en casi cualquier tarea:
→ github.com/msitarzewski/agen… (158k estrellas)
Unos 170 agentes organizados como una empresa: ingeniería, diseño, marketing, ventas, producto, testing, seguridad y soporte.
Con packs adicionales para finanzas, salud y desarrollo de videojuegos.
Compatible con más de 15 herramientas.
→ github.com/wshobson/agents (40k estrellas)
202 agentes, 184 skills y 105 comandos repartidos en 94 plugins para Claude Code.
Especialistas en lenguajes de programación, revisión de código, depuración, infraestructura, seguridad y testing.
Instala solo los plugins que necesites.
→ github.com/anthropics/knowle… (26k estrellas)
Los plugins del propio Anthropic, organizados por áreas de trabajo: ventas, atención al cliente, gestión de producto, marketing, legal, finanzas y datos.
→ github.com/anthropics/claude… (18k estrellas)
Agentes oficiales que puedes poner en marcha: un resumen diario de Slack y GitHub, clasificación programada de incidencias de Sentry, bots para Slack y Linear, un agente de atención al cliente y un analista de datos financieros.
→ github.com/addyosmani/agent-… (103k estrellas)
De ingenieros de Anthropic.
25 skills que cubren todo el ciclo de desarrollo: interview-me, redacción de especificaciones, desarrollo guiado por tests, pruebas en navegador, revisión de seguridad e integración y despliegue continuos
Yo empezaría por claude-quickstarts.
Cada agente es un ejemplo completo y funcional que puedes ejecutar y desmontar para entender cómo funciona.
Aprovecha estos recursos oficiales gratuitos para tus próximos proyectos👇
89
787
5,207
495,696
Absolutely brilliant. -20% token use
🚨 A Netflix engineer just open-sourced the fix for the most expensive problem in AI: wasted tokens
It's called Headroom, and it might be the smartest fix to high token usage
Your agent reads a 10,000-token log file to find one error.
You paid for all 10,000 tokens.
The answer needed 1,200.
Headroom sits between your agent and the LLM and compresses everything before the model sees it.
JSON, code, logs, RAG chunks, each gets its own specialized compressor.
And it's reversible: the originals stay on your machine, so nothing is lost.
The results speak for themselves:
→ Up to 95% fewer tokens
→ Same accuracy on benchmarks
→ Zero changes to your code
Setup takes one minute:
1. pip install "headroom-ai[all]"
2. headroom wrap claude
Done.
Works with Claude Code, Cursor, Codex, and anything OpenAI-compatible.
Everything runs locally, fully open source.
The cheapest token is the one you never send.
🔗 Github repo: github.com/headroomlabs-ai/h…
So, are you compressing your context, or just paying the bill?
3
Ok, so I actually really like my Grok @Bot setup. It managed to get through a lot of work and was mostly autonomous within my project/framework setup. BUT ... the limits are kind of useless. Why does the weekly limit get consumed so fast when the monhtly stays at only 12%. @poteto please help ...
1
20
OK, @poteto Grok Bot is awesome. It built a whole team using dr eggbot and then applied my framework to it. Pretty impressed so far
6
I'm currently deliverying approx 200-250 PRs per month, so I thought I'd give Grok @Bot my framework and ask it to build me a high-performing GrokTeam. What could possibly go wrong here?
4
200 PRs this month (and climbing) delivered with high quality and near total trust for my AI agent team. I hardly ever have to read code these days. Just state intent, approve design and visually verify at the end. It’s a dream to watch.
4
Agent to agent communication is next level… once you have a team of AIs working for you, you’ll understand.
1
8 agents. 1 team leader, 1 env controller, 1 release mgr, 5 devs.
I think I’ve finally cracked the code on getting the AI team shipping PRs.
Tracking at 240ish PRs per month now. Maybe more if I push it and don’t run out of codex credits/resets.
Gpt-6.1-sol speed has improved a fair bit since 24hrs ago. Thanks Tibo @thsottiaux
2
My favourite prompt for @OpenAICodexCli
Stop fucking around and ship PRs. You’re not even using your available devs at all. What the fuck are you doing?
1
1.5B token today ...
1 Dot BotMaster, 1 Codex Team Leader, 1 Codex Prod Support Eng, 1 Codex Env Controller, 4 Codex devs
That's my team. And it's still not moving more PRs than I do by running the show myself directly in Codex.
16
First observations of using @OpenAIDevs Dot as a Chief of Staff with codex agents. The setup is similar to pstack but with enhancements and running 7 codex agents all enabled to run multi sub-agent sessions. Team of 20-30 overall.
- Dot as a coordinator is generally useless.
- Within 24hrs it completely lost the ability to use tools to communicate with all agents except one outside the project context
- Messed up delivery
- Only progressed some items off a 17 item backlog and not to completion
- It generally ignores agents complaining they are stuck and just need approvals or confirmation which was given prior and is clearly in the instruction set
- What's the point of a Dot if it can't communicate to an authorised computer? It tried to use a spare chat as a bridge to communicate to other agents. Why not just communicate directly?
So, hasta la vista ... this Dot experiment is a failure @OpenAI ... no amount of free credits fixes that problem
1
58
One Bot Master (dot), seven Codex agents, dev-stack and a 130,000 line complex monorepo and gpt-6.1-sol xhigh. What could possible go wrong here?
26
I hate context compaction. It’s like your agent gets a lobotomy every time it happens. LLMs would be so much more powerful if we could solve for that issue
6
Do you want a proven AI delivery framework?
Do you want to spend less tokens? How about have a best practice devops/gitops?
Self-verification loop that ensures your AI agent actually gives you what you intended?
All done. You need dev-stack. All the skills, tooling, workflow, git workflows to make you a 10x engineer.
I personally use this to deliver 200+ PRs per month.
github.com/MikeS071/dev-stac…
Thanks Lauren @poteto for her initial work on pstack which is integrated into dev-stack. Though I worked on making the process more seamless to cater the requirements-> arch/design -> build/test -> verify -> fix -> deliver loop working in an automated way so you the human can be almost hands off (other than specifying the scope)
Let me know what you think
60
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.
2,232
5,270
53,278
10,189,428
Michal Szalinski retweeted
GPT-6 Sol and Luna are out. Not only are they a very significant improvement across the board, but also in writing and general "you know when you try it" quality.
We are also permanently reducing the API price by 50% making both of them viable for a ton of new usecases and making your usage go further too, even on the subscriptions.
And one more thing. We are loading a banked reset into all accounts of our Plus, Pro and Business users. Let's go!
openai.com/index/introducing…
We have been focusing on efficiency and intelligence for all.
Very proud of the team. Only possible when you have incredible models at the top end of the capability that you can then use to make a big difference in everything else.
2,970
1,599
25,161
3,288,450
CTO - Chief Tokens Officer 👮😂
I need one of those asap
instagram.com/reel/DdffxgmOa…
56