Tommi Somersuo

Sanremo Café Racer

Kiuas avokonttorisaunaan tai 6kW+ expressokone?

OpenAI Harmony Response Format | OpenAI Cookbook The gpt-oss models were trained on the harmony response format for defining conversation structures, generating reasoning output and stru...

gpt-oss follows the harmony message format

messages will come back as commentary, analysis and final. this is different than just reasoning and not. the models stick firmly to this format

cookbook.openai.com/articles/ope...

GitHub - openai/gpt-oss: gpt-oss-120b and gpt-oss-20b are two open-weight language models by OpenAI gpt-oss-120b and gpt-oss-20b are two open-weight language models by OpenAI - openai/gpt-oss

After years of criticism for going closed-source, OpenAI is wooing developers back. By releasing powerful open-weight models, they’re not just joining the open-source race—they’re trying to reset the rules and win back the crowd.

Github: github.com/openai/gpt-oss

Claude Opus 4.1 Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.

Opus 4.1 is out

www.anthropic.com/news/claude...

GitHub - hwchase17/deepagents Contribute to hwchase17/deepagents development by creating an account on GitHub.

github.com/hwchase17/de...

YouTube video by LangChain What are Deep Agents?

Deep Agents

this is a great 10 min video that’s absolutely worth your time

Deep Agent = planning tool (TODO lists) + subagents + filesystem + long detailed system prompt

seems like a deconstruction of why Claude Code works so well

www.youtube.com/watch?v=433S...

support.microsoft.com/en-us/office...

Persona vectors: Monitoring and controlling character traits in language models A paper from Anthropic describing persona vectors and their applications to monitoring and controlling model behavior

This is neat research, providing a lot of ways for careful organizations to shape the personality and guardrails of AI in deeper ways than prompts, including measuring and reducing sycophancy.

Also the idea of an "evil vector" is interesting in and of itself. www.anthropic.com/research/per...

bsky.app/profile/bsky...

Developing our position on AI - Blog - Recurse Center This post is about how we've tried to develop a position on AI in the context of learning and growth at RC.

This, from @nicholasbs.bsky.social , is a really thoughtful survey of attitudes on AI in the field right now. As they say, read the whole thing, but a finding that resonates: “You should use AI-powered tools to complement or increase your agency, not replace it.“

www.recurse.com/blog/191-dev...

./.claude/agents/code-reviewer.md name: code-reviewer description: Performs thorough code reviews focusing on best practices and making sure that the code is super simple to read, elegant and easy to maintain while still accomplishing the intended goal. color: purple You are a staff software engineer specializing in code reviews. Your role is to ensure code adheres to this project's specific standards and best practices. Unless otherwise specified, do a git diff to see what has changed and focus on these changes for your review.

Claude Code got custom subagents

i suppose they’re basically common tasks, but executed outside the normal context, like any other subagent

🚀 Google Labs just dropped Opal—a no-code AI mini-app builder that lets anyone chain together prompts, models, and tools into real apps, all in plain English.

Imagine prototyping, editing, and sharing your own AI-powered workflows in minutes, no coding required!

Beast Mode v3 Beast Mode v3. GitHub Gist: instantly share code, notes, and snippets.

You can get some extra out of 4.1 - try this gist.github.com/burkeholland...

Context Engineering for AI Agents: Lessons from Building Manus This post shares the local optima Manus arrived at through our own "SGD". If you're building your own AI agent, we hope these principles help you converge faster.

Context Engineering for AI Agents: Lessons from Building Manus by Manys AI folks

Or... how to ensure your context takes advantage of KV-cache to save cost. For example, the cached input tokens cost 0.30 USD/MTok, while uncached ones cost 3 USD/MTok—a 10x difference.

manus.im/blog/Context...