Show HN: TaskPeace – a task queue my AI coding agents pull work from over MCP
TaskPeace is a task queue that lets AI coding agents pull work items directly over MCP instead of a human dispatching each task.
11 articles · 3 categories
The finishable daily brief
Friday, Jul 3, 2026
11 articles · 3 categories
read top to bottom · then stop
In 30 seconds
Today's signal was mostly about coding-agent tooling maturing past the demo stage: a task queue that lets agents pull work over MCP, a linter for the instruction files that steer agents, an MCP memory layer for codebase queries, and a library for running LLM work as bulk agentic jobs.
Alongside the tooling, practitioners pushed back on how agents get framed and trained — one critique rejects the "coworker" metaphor outright, while Vercel's Andrew Qu argues agents need skills, sandboxes, and agent-readable sites rather than chat-style prompting, and a new presentation details Agent RFT, OpenAI's reinforcement fine-tuning approach for reasoning models.
New open-source tools target the operational plumbing coding agents need in production: pulling work over MCP, remembering codebase context across queries, keeping steering files honest, and running bulk agentic jobs.
TaskPeace is a task queue that lets AI coding agents pull work items directly over MCP instead of a human dispatching each task.
codebase-memory-mcp is an MCP server that gives coding agents persistent codebase memory so repeat queries don't re-scan the repo from scratch.
Skillsaw is a linter for Agents.md-style instruction files, catching drift between what a coding agent is told and what the codebase actually looks like.
Lotus is a library for running LLM and agentic workloads as optimized bulk jobs instead of one request at a time.
A comparison lays out OpenCode, Pi, and Goose as three distinct layers of today's AI coding agent stack rather than competing replacements.
Practitioners argued over how agents should be framed and steered, from rejecting the "coworker" label to Vercel's case for skills and sandboxes over chat, plus a look at how OpenAI trains reasoning models to use tools via reinforcement learning.
MIT Technology Review pushes back on framing AI agents as "coworkers," arguing the metaphor obscures how these systems actually fail and need supervision.
Vercel's Chief of Software explains why its eve agent framework is built around skills, sandboxes, and agent-readable websites instead of chat-style prompting.
Speakers detail Agent RFT, OpenAI's platform for fine-tuning reasoning models through real-time tool interactions and custom reward signals to solve credit-assignment problems.
Agent Draw demonstrates an agent sketching in real time as you speak, built on the TLDraw canvas, as an example of agent UX beyond text chat.
Cloudflare detailed the internal data platform now running its own billing and security AI agent, while Google DeepMind moved into a new application vertical by partnering with a film studio.
Cloudflare's Town Lake unifies operational, billing, security, and business data, and its Skipper AI analytics agent now handles roughly 91K billing queries across that platform.
Google DeepMind and film studio A24 announced a research partnership exploring AI tools for filmmaking, extending frontier-lab research into a new creative vertical.
You are caught up for this edition