🤖 AI Agents Weekly: Agent 3, ChatGPT Developer Mode, MCP Registry, Writing Effective Tools for Agents, Qodo Aware
Agent 3, ChatGPT Developer Mode, MCP Registry, Writing Effective Tools for Agents, Qodo Aware
In today’s issue:
Replit Introduces Agent 3
A guide to write effective tools for agents with agents
The second wave of MCP
Anthropic introduces web fetch tool
OpenAI Introduces ChatGPT Developer Mode
Defeating nondeterminism in LLM inference
MCP Registry has been launched
Top AI papers, product updates, and tools
Top Stories
Replit Introduces Agent 3
Replit has launched Agent 3, its most autonomous coding agent yet, designed to both build and self-test applications while also creating other agents and automations. It marks a big jump from Agent V2 in autonomy, speed, and cost efficiency.
Key updates:
Built-in app testing – Agent 3 uses a live browser to interact with the apps it creates, checking buttons, forms, APIs, and login flows. It summarizes results and fixes issues automatically. A “Max Autonomy” mode (beta) lets it run unsupervised for up to 200 minutes, managing tasks and monitoring its own progress.
Agent + automation builder – For the first time, Agent can generate other agents and workflow automations. Users can create bots for Slack, Telegram, or integrations with Notion, Linear, Dropbox, Outlook, etc., using a guided setup for secure credential handling.
Extended autonomous runtime – Sessions can now last several hours, freeing users from manual oversight while the Agent builds, tests, and improves apps.
Flexible app creation flow – Users can start with frontend-only prototypes for rapid iteration or build full-stack apps immediately, switching modes as needed.
Agent 3 is available to all free and paid Replit users, opening up new levels of autonomous coding and workflow automation.
Keep reading with a 7-day free trial
Subscribe to AI Newsletter to keep reading this post and get 7 days of free access to the full post archives.