back to all use cases

agent-lab.

we run a team of ai agents and needed to keep track of what was approved.

a paper model of a small village of grey and black boxes with square windows. one window, near the middle, is red.

the problem.

who
our own studio.
before
working with a team of ai agents means many roles, sessions and proposals at once. it's easy to lose track of who did what, and of what was actually approved.
what was needed
one place to see which agent is working, and to approve every change first.

how we thought about it.

  1. agents only propose; nothing changes until we approve it.

    because with many roles at once, it's easy to lose track of what was actually approved.

from sketch to screen.

  1. sketchmany roles, sessions and proposals at once, drawn as one map.
  2. measuresized for a laptop first, and on a phone.
  3. wireframeteam map, live sessions, project memory, blueprint, in grey.
  4. screena local dashboard. agents only propose; we approve.

what we built.

agent-lab, a local dashboard for our claude code agents: a map of every role on the team, the live sessions, the project memory, and a blueprint where agents only propose and nothing changes until we approve it.

agent-lab's teams screen, in portuguese: a row of cards, each a built-in team of agents with a duplicate button.
real screens of our own tool, in portuguese.
agent-lab's map, in portuguese: a tree of the team's configuration, with the project node open on a list of every agent role.
the map: every role on the team.
agent-lab's teams screen at phone width, in portuguese: the header, a sliding menu and the team cards stacked one under the other.
the teams screen at phone width.
agent-lab with a fictional demo project.

how we use it.

day to day at the studio: we see which agent is working, start new projects from our team templates with a preview before any file is written, and review what agents propose before it becomes direction.

agent-lab's new project screen, in portuguese: step 1 offers four project types as radio cards; steps 5 and 6 show the preview and a create-files button that stays off until a preview exists.
a new project from a team template, with a preview before any file is written.
agent-lab's blueprint screen, in portuguese: a line saying agents only propose and nothing changes until you approve it, a badge for the served version, tabs for current version, proposals, history and propose a change, and no open proposals.
the blueprint: agents propose, we approve.

status.

what
ai agents
status
used every day

our own tool, built and used every day. not for sale.

see how we work. see all use cases.

have something like this? tell us.

sending a message doesn't commit you to anything.

send us a message.

next use case

reformer social club.