Nick Bobrowski
@nicko_ai
Followers
69
Following
2K
Media
57
Statuses
251
🔥 Just shipped Agency Swarm v1.0.0-beta.1. Completely rewrote it on OpenAI’s new Agents SDK. No more black-box Assistants API - we now have direct control over threads and runs (read: you can use third-party models). Plus native Search, Computer Use and o3/o4-mini support.
2
0
1
RT @bradlightcap: gpt-5 is much better at instruction following than previous models. API users spending time optimizing their prompts are….
cookbook.openai.com
GPT-5, our newest flagship model, represents a substantial leap forward in agentic task performance, coding, raw intelligence, and steera...
0
125
0
RT @YayaSoumah: A few notes on GPT-5 vs Claude Models. In order to properly learn how to best leverage a specific model, you need time, a….
0
3
0
I’ve never seen so many bugs in the ChatGPT app. At this point, I believe most of the development is handled by AI, and humans only review a fraction of the changes. The only way to avoid serious issues is to give AI coding agents eyes:.
linkedin.com
I gave my AI coding agents eyes. Now they can see what they’re building — and fix their own UI bugs. The secret? Selenium screenshots + GPT-4.1 visual analysis in my test suite. Here’s the workflow...
0
0
0
🧠OpenAI open-sourced their deep research agent stack — O3/O4-mini quality in ~50 LOC. Includes full prompt logic, agent chaining, and tool use. Build your own research agent now. 🔍
1
0
0
This.
Immigrants in Portugal are just the scapegoat for a dysfunctional government and corrupt system that doesn't build housing, makes it impossible for Portuguese to start or run businesses. This means no jobs created so no job opportunities for them, and no houses built so no places
0
0
0
Testing it with our new Agency Swarm framework v1.0 (beta) built on top on Agents SDK (+orchestrator-workers pattern + extra production-ready features).
Our evals product just got 10x more powerful! You can now evaluate agentic workflows end-to-end using the Responses API. For example, you can enable web_search and a couple of remote MCP servers and then evaluate the models ability to use all 3 tools to solve some complex task.
1
0
1
RT @itsalexvacca: Humanity's progress is accelerating insanely fast:. Stone Age→Farming: 100,000 yrs.Farming→Steam: 12,000 yrs.Steam→AI: 20….
0
2K
0
RT @hidecloud: Claude 4’s ability to follow complex, multi-step instructions and work through problems with clear chain-of-thought reasonin….
0
1
0