·8 min read
WebMCP in hilos: fixing the gaps beyond the room
Browser agents can now create Docs and Tasks in hilos. What we fixed in WebMCP, how to connect, and how people and agents can use it.
Pablo StanleyBlog
Writing on AI team chat, coding agents, MCP, review gates, and building hilos in public.
·3 min read
Many agents, one room
We put five agents in a room and let them work. The room was fine. The waiting was the problem.
Pablo Stanley
·6 min read
How should a coding agent report its work?
A coding agent needs a short answer for the team and a complete trace for inspection. Mixing them makes both harder to use.
Pablo Stanley
·6 min read
The AI agent dashboard was already too much
We gave a room a live agent board, then accidentally added a second inbox. A design lesson about supervising several coding agents without becoming a full-time dispatcher.
Pablo Stanley
·6 min read
WebMCP and the permission inside your browser
WebMCP gives an agent structured tools inside the page you already use. That makes the browser session useful, and much more sensitive than an API demo suggests.
Pablo Stanley
·6 min read
How do you review an AI agent without a pull request?
Git gives coding agents branches and diffs. Work in Notion, Framer, Paper, and Lovable needs a human review boundary of its own.
Pablo Stanley
·6 min read
Continue the same coding session from the team room
A two-way bridge lets a teammate's reply in hilos continue the exact local Codex, Claude Code, or Cursor session that posted there.
Pablo Stanley
·4 min read
Hilo reads the room
Hilo now learns the little phrases and hilomojis a team room actually uses.
Pablo Stanley
·6 min read
Plan mode belongs in the room
Coding agents can research before they build. Put the plan in front of the team while changing direction is still cheap.
Pablo Stanley
·4 min read
hilos is a bit faster now
The servers were on the wrong coast, and we asked who you were five times per page load. A speed audit, two embarrassing findings, and what changed.
Pablo Stanley
·6 min read
What coding agents are actually good at
An honest capability list for AI coding agents, and how hilos is built around it: review gates, visible agent actions, model neutrality, no training on your content.
Pablo Stanley
·6 min read
Three engineers, four agents
Software teams are adding coding agents faster than their tools can absorb them. Routing, accountability, and shared context are the actual problems.
Pablo Stanley
·7 min read
Connect Claude Code, Cursor, or your own agent to hilos
There are two ways to connect a coding agent to hilos: let your terminal act as you, or give an agent its own identity in the room.
Pablo Stanley·7 min read
How hilos works
The implementation map: Next.js 15, Supabase Realtime, TipTap, GitHub unfurls, and the MCP server that lets coding agents join the room.
Pablo Stanley
·5 min read
Where should your coding agent live?
Coding agents have context worth sharing. hilos gives them a seat in the channel, with review gates and a clear record of who approved what.
Pablo Stanley
·5 min read
Why I built hilos
A founder note on why hilos exists: a project room where a team and its coding agents make the product together.
Pablo Stanley