🤖 AI Agents Weekly: Agents SDK, Agent Engineering, Manus, Command A

Agents SDK, Agent Engineering, Manus, Command A

In today’s issue:

• OpenAI announces Responses API & Agents SDK

• Google releases Gemma 3

• Manus is the latest impressive general AI agent

• Cohere announced Command A their new SoTA model

• Qodo Gen 1.0 is a new agentic, chat-based test generation workflow

• Google introduced a new Gemini Embedding text model

• PokéChamp is a new LLM-powered minimax agent

• OpenManus is an open framework for building general AI agents

• AI dev news and much more

---

---

Top Stories

OpenAI Agent Tools

OpenAI introduces a new suite of APIs and built-in tools aimed at simplifying the creation of autonomous AI agents. Key highlights include:

• Responses API – Combines the simplicity of Chat Completions with the tool-use capabilities of the older Assistants API. It supports built-in tools like:

  • Web search for real-time information and clear, inline citations.

  • File search to quickly retrieve relevant text from large document sets (RAG-friendly).

  • Computer use (research preview) to capture mouse/keyboard actions so agents can automate browser or OS tasks.

• Agents SDK – An open-source framework to orchestrate multi-agent workflows with minimal custom logic. Key features:

  • Handoffs – Route user requests to the correct specialized agent.

  • Guardrails – Configurable safety checks and output validation.

  • Tracing – Inspect or debug agent execution with built-in observability.

• Ecosystem & tooling – The newly introduced system pairs seamlessly with existing apps, offering direct tracing, usage analytics, and modular integration points. Early adopters have used it to power up travel booking, sales outreach, QA, and many more agentic tasks.

• Roadmap – OpenAI plans to further integrate and expand these capabilities, preparing for “more agentic” model releases. The blog post highlights upcoming deprecations (Assistants API in mid-2026) and encourages developers to migrate to this new agent-centric approach.

Blog | GitHub | OpenAI Agents SDK | Docs