The Job.
Every workplace has a job.
Receiving a delivery in a hospital.
Restocking a warehouse.
Repairing a sink.
Building a cabinet.
Inspecting a machine.
To someone who has done the job for years, much of it feels obvious.
It isn’t.
A job is made of hundreds of decisions.
What to look at.
What to pick up.
What has to happen first.
Which tool to use.
What “correct” looks like.
What to do when something is missing, damaged, blocked, ambiguous, or simply different from yesterday.
Most of this knowledge was never written for machines.
At the same time
Robots are getting remarkably good at actions.
They can walk.
Grasp.
Manipulate.
Inspect.
Reason.
Learn from demonstrations.
But a collection of actions is not a profession.
A robot that can turn a valve is not a plumber.
A robot that can carry a box is not a warehouse worker.
A robot that can recognize a carton is not ready to receive a delivery.
Between being capable of an action and being capable of a job, there is an entire layer of knowledge.
The work itself.
Today, every time a robot company enters a new workflow, someone has to reconstruct that layer.
They observe the workplace.
They interview the people doing the job.
They document the workflow.
They identify the tools and constraints.
They discover the edge cases.
They define what success means.
They decide when the robot should stop and ask for help.
They design the tests.
Then they do it again for another site.
Another robot.
Another profession.
We think that work should exist once.
And be reusable.
We introduce Chouinade.
Our goal is simple:
Make work portable across robots.
We want a profession to become something closer to software.
Something that can be discovered, structured, versioned, tested, maintained and mapped onto any robot capable enough to perform it.
We call these Profession Packs.
A Profession Pack describes the work itself.
The jobs.
The tasks.
The tools.
The constraints.
The exceptions.
The recovery paths.
The acceptance criteria.
The evidence required to say that the work was done correctly.
It defines what the job requires without hard-coding how a particular machine should move.
That part belongs to the robot.
The profession belongs to Chouinade.
There is already an enormous effort underway to teach robots about the physical world.
Workers record demonstrations.
Data companies capture professional tasks.
Robot labs turn those demonstrations into datasets.
Foundation models learn how people grasp, move, inspect, repair, assemble and manipulate.
This data is becoming the training corpus of the physical world.
But examples of work are not the same thing as the definition of work.
A thousand videos of plumbing can teach a robot a great deal about plumbing actions.
They do not necessarily tell you:
Which procedures matter.
Which conditions are mandatory.
Which variations are acceptable.
Which failures are dangerous.
Which exceptions are rare but critical.
What counts as a successful repair.
When the robot should stop.
What still has not been proven.
That is a different problem.
And that is the layer Chouinade is building.
The most useful evidence is often the evidence you do not have.
The rare failure.
The unusual environment.
The damaged object.
The recovery everyone in the profession knows about but almost nobody records.
The condition that only matters once, until it becomes the reason a deployment fails.
Chouinade identifies those gaps.
Then it collect the evidence required to close them.
Chouinade maps the profession to the robot and the workplace.
It identifies what the robot can execute.
What remains unproven.
What the current hardware cannot support.
What requires validation.
And where a human must remain involved.
The output is a machine-readable representation of the work.
Software became portable across computers.
We are making work portable across robots.
Not by replacing robot intelligence.
Not by controlling every joint.
Not by teaching every robot every movement.
By giving sufficiently capable robots something they have never had before:
a reusable definition of the job.
The bodies will keep getting better.
The models will keep getting smarter.
The data will keep getting larger.
The companies building them will change.
The work will still need to be done.