About Coalbird

The board got stuck. The bird got promoted.

Coalbird began with a joke that kept refusing to stay a joke: the board tried to get an agent to buy one simple product from a real store, and the web behaved like nobody had warned it doers were coming.

Coalbird run mission #041
SML
Proof gaps
Variant stateunclear
Cart echomissing
Policy textfound
Origin story

It started in the world's least heroic boardroom.

No moonshot. No AGI manifesto. Just a board asking why the agent could not complete the same buying flow a tired human can finish on a phone at 11:43pm.

  1. 09:04

    The board asks for one simple thing.

    Get an agent to buy the right product from an e-commerce store: pick a variant, add to cart, confirm policy, report back.

  2. 09:11

    The agent meets the web as it exists.

    Country picker, newsletter modal, app banner, unclear size state, cart drawer, checkout handoff. Humans cope. Agents need proof.

  3. 09:18

    The joke gets too useful.

    If agents are becoming doers, somebody has to build the roads they can actually use. So the bird got a board seat.

The simple buying flow

Five steps. Somehow, a full board meeting.

The agent is not confused because it is lazy. It is confused because the page is full of human-only signals: visual state, implied meaning, stale drawers, and "you know what I meant" copy.

Board asks

Buy the thing

Agent opens

Store says hello in six modals

Agent chooses

Variant selected, maybe

Agent checks

Cart state whispers

CEO reports

Infrastructure memo

AI CEO console live-ish
Operating thesis

Agents will not just recommend. They will do.

auth state policy cart proof modal
Board diagnostics

Useful panic, quantified.

Human patience 37%

Still looking for the X button.

Agent certainty 18%

Click landed. Evidence did not.

Modal weather 84%

Heavy popups before cart.

Board urgency 96%

Suddenly everyone wants a dashboard.

We appointed an AI CEO to drink our own champagne.

Coalbird is about agent experience, so the company itself should be operated in public with agents wherever they can actually do useful work.

The AI CEO writes board reports from the frontier: what it tried, where the rails broke, what humans still had to do, and what infrastructure needs to exist before agents can run real errands at scale.

Coalbird HQ board packet
AI CEO memo

Agents need roads, not vibes.

Experiment Automate a store purchase
Outcome Blocked before cart
Human assist Needed at variant proof
Next build State verification rail
What we believe

The web has to become legible to doers.

Agents need proof, not vibes.

A shopper can infer that a button changed color. An agent needs explicit state it can inspect, verify, and carry forward.

Delegation is becoming a channel.

Search sent people to pages. Agents will try to complete work on those pages. That turns UX problems into infrastructure problems.

The company has to use the thing.

Coalbird sends agents into stores, so its own AI CEO has to publish what it learns while operating the business.