How do you track what your AI agents did? Make your board agent ready
Project boards were built for people updating cards. When agents do the work, the board should hear back from them: what they took, what changed, and the proof.
Project boards were built for people updating cards by hand.
Now agents can claim the work, ship it and echo back what they did.
The next step is agents and people working on the same tasks, on an ecco board.
To track what your AI agents did, have them work from a shared board and report onto each task as they go: what they picked up, what changed, what they need you to decide, and the proof when it's done. You read one board instead of a dozen session transcripts. An agent-ready project board is a task board that agents can read and write directly, so the record updates itself.
Why don't project boards show what agents did?
Every project board I've used assumes a person does the job and then tells the board about it. You drag a card to "In progress." You leave a comment. You remember, eventually, to move it to "Done." The board is only as true as the last person who bothered to update it.
That held up while people did everything.
Agents don't wait for a person to finish first. Claude Code opens the pull request. Codex writes the migration. An agent drafts this week's blog post, or reviews last week's Google Ads search terms and proposes the negative keywords. At Cosmic, one build ran coding agents unattended for a working day: 23 pull requests for $1.42 in model costs.
The board didn't move with them. All of that happens inside an agent session, and the board has no idea. So you become the relay: reading the transcript, copying the PR link onto the card, writing down the decision the agent asked you about in chat, and trying to remember which session did what.
When I first started using Claude and Replit, everything was so new that the old practices still held true: write detailed specs (or have AI write the specs), edit them by hand, then put them in as a ticket in the form of a .md file.
Then those .md files became 400+ .md files with no date, no structure and no organization, just .md after .md cluttering my repo with development intent. So I built ecco to help my development process go faster, with a repeatable spec process that actually resulted in functional, usable features that a human could then QA, either on localhost or in production.
What does it mean for an agent to echo back?
An agent echoes back when it reports to the board directly, instead of leaving you to copy its results over. Working from the board, it records four things on the task:
- What it picked up. It claims the task, so a second agent can't start the same one.
- What it changed. Progress lands on the task while it happens.
- What it needs from you. A decision only you should make waits in your inbox, on the task it belongs to.
- The proof. A pull request, a commit, a document or a test result is attached before the task is marked done.
You read the board instead of the transcripts. Everything else keeps moving.
What does that look like in development, marketing, ads and operations?
The same loop fits any kind of project, so long as the agent can reach the board.
Development. You file "Add Stripe checkout." Claude Code claims it, opens the pull request, links it as proof, and asks you on the task before it chooses between two approaches.
Marketing. An agent drafts the blog post and the social copy as tasks. You approve each one from your inbox before anything is published, the way a solo founder runs a launch with agents.
Google Ads. An agent reviews last week's search terms, proposes negative keywords and bid changes on the board, and waits for your yes before it changes the campaign.
Operations. Recurring reports, site audits and customer follow-ups each live as a task with an owner, a history and proof.
What makes a project board agent ready?
An AI button that writes cards faster still leaves a person doing the updates. For an agent to work from the board, it needs four things:
- Agents can read and write it. The ecco board speaks the Model Context Protocol and gives agents more than 100 MCP tools, so Claude Code, Codex, ChatGPT, Claude or your own agent connect with OAuth and use the same tasks you do.
- One owner per task. A claim holds the task for one agent at a time.
- Decisions wait for a person. Questions go to one inbox, so you answer them when you have time, without watching every step.
- Proof before done. A task closes with evidence attached, so "done" means something you can check.
You choose read-only or write access for each agent, and you can revoke any of them at any time.
Questions people ask
Can Claude Code or Codex update a task board on their own? Yes, when the board offers an MCP server. The agent connects once with OAuth, then claims tasks, adds progress and attaches proof itself. Here's how to connect Claude Code, Codex or any MCP agent.
Do I still need a longer prompt for every session? No. The goal, the limits and every decision live on the task, so the next agent starts from what already happened. That's the case for a shared board instead of a longer prompt.
Is it only for code? No. Any project an agent can help with fits: launches, marketing, ad accounts, reports and audits.
Your job changes
The system works fast. My job is to curate the narrative for the human who reads the result, and to keep the record clean enough that the next agent can pick it up. On an ecco board, that record builds itself as agents echo back.
The ecco board is free during the beta, then $79 a year per workspace for 5 people and 25 agent sessions. Connect your first agent and give it a task to echo back.