Behind my house is an old metal garage.
Thirty feet by fifty feet.
It belonged to my dad.
There is nothing futuristic about it.
Metal walls.
Concrete floor.
Big doors.
A working garage built for working with actual things.
And sometimes I stand inside it and see something that isn’t there yet.
So come with me for a minute.
Not into the garage as it exists today.
Into the garage I’m trying to build.
⸻
The door rolls upward.
The old walls are still there.
The concrete is still there.
My dad would recognize the building immediately.
But somewhere along the way, the garage has become something else.
A laboratory.
A workshop.
A machine shop.
A research environment.
An engineering studio.
A place for electronics.
Software.
Robotics.
Fabrication.
Publishing.
Whatever strange thing needs to exist next.
There are workbenches around the room.
Machines occupy one section.
Electronics another.
Parts wait beside test fixtures.
Books and diagrams share space with motors, cameras, sensors, raw material, prototypes, and things that haven’t decided what they are yet.
There are computers here.
But computers aren’t really the point.
⸻
Somewhere in the architecture is one serious machine.
The center.
OmniShell.
Not because I want to build the world’s largest chatbot.
Quite the opposite.
The idea behind OmniShell is that intelligence should not remain trapped inside the little rectangle where the conversation happened.
Research systems.
Models.
Software tools.
Databases.
Engineering programs.
Machines.
Sensors.
Eventually robotics.
They all possess different capabilities.
What interests me is the architecture that allows those capabilities to work together without pretending they are all the same thing.
The central computer is simply where much of that coordination lives.
The garage is where it eventually meets reality.
⸻
Imagine I’m standing at a workbench holding a mechanical part.
Something about it bothers me.
I don’t open seventeen applications.
I don’t spend twenty minutes reconstructing where I left the relevant files.
I say:
“I’ve been wondering whether there’s a better way to make this joint.”
That sentence begins the work.
Existing research is found.
Previous experiments appear.
Materials are compared.
Possible geometries emerge.
Arguments form for and against them.
Simulations eliminate bad ideas cheaply.
Different systems challenge one another.
And somewhere underneath all of it, the Architecture Wall keeps asking questions that matter enormously to me:
Where did this idea come from?
What supports it?
What contradicts it?
What remains uncertain?
Is this established?
Proposed?
Inferred?
Speculative?
Has reality tested it yet?
Because I don’t want to build machines whose purpose is to make artificial intelligence sound certain.
I want infrastructure that helps intelligent systems remain accountable to what we actually know.
⸻
Then the question moves.
Research becomes engineering.
Engineering encounters manufacturing.
Manufacturing encounters the actual machines sitting twenty feet away.
Can this part be cut?
Can the tool reach it?
Can the bearing be replaced?
Is the tolerance realistic?
What does it cost?
What happens when someone has to repair it three years from now?
The digital design does not live in an imaginary world.
Eventually someone has to pick it up.
⸻
This is where the garage becomes important.
I am interested in reducing the distance between:
I wonder if…
and
Let’s find out.
That distance is currently filled with extraordinary amounts of friction.
Searching.
Organizing.
Translating between tools.
Rebuilding context.
Moving information manually.
Finding the right capability.
Remembering why a decision was made.
Discovering that the simulation assumed something the machine cannot actually manufacture.
Losing the result of an old failure and repeating it three years later.
None of those problems are glamorous.
Collectively, they consume enormous amounts of human attention.
I want to see how much of that distance can be collapsed.
⸻
Imagine a question begins on Monday.
Research reveals what is already known.
Strong possibilities survive criticism.
Weak ones disappear.
One promising idea moves into engineering.
A testable version is designed.
Manufacturing constraints reshape it.
Parts are fabricated.
Sensors are prepared.
Software is written.
And eventually there is something sitting on the workbench that did not exist when the week began.
Not because an AI magically invented it.
Because many different forms of capability were coordinated around a human question.
Then reality gets a vote.
⸻
Maybe the prototype works.
Good.
Now we know something.
Maybe it fails.
Better than pretending.
The result enters the Architecture Wall:
Contradicted.
Why?
What assumption failed?
Was the model wrong?
Was the measurement wrong?
Did we misunderstand the material?
What experiment would distinguish the remaining explanations?
Failure doesn’t disappear into embarrassment.
It becomes information.
The next experiment begins with more knowledge than the last one had.
⸻
That accumulation may be the part I care about most.
I want the thousandth experiment to inherit something from the first 999.
Not just files.
Reasoning.
Lineage.
Why something changed.
What evidence caused the change.
What test it passed.
What test it failed.
Which uncertainty remains unresolved.
The history of making a thing should not have to disappear simply because the thing has been made.
⸻
And the strange part is that I don’t want this infrastructure only for engineering.
Walk farther into the garage and there might be a book being laid out.
A DJ Field Guide illustration.
A new FluxMintDigital tool.
Software.
Research.
A learning system.
Something for a client.
An experiment I have not imagined yet.
Those things look unrelated if you organize the world by industry.
They look much more related if you organize it by capability.
Research.
Reasoning.
Writing.
Design.
Simulation.
Fabrication.
Testing.
Memory.
Coordination.
Creation.
OmniShell isn’t interesting to me because it can become one particular application.
It is interesting because I want to know whether a substrate can understand enough about available capabilities to help connect intent to whatever combination of them the work requires.
⸻
People sometimes make the Tony Stark comparison when I describe this.
I understand why.
Robots.
Machines.
An intelligent system.
A workshop full of strange projects.
But the holograms were never the interesting part of those fictional laboratories.
The interesting part was the collapse of distance between imagination and execution.
An idea could move through computation, modeling, engineering, fabrication, testing, and revision almost as one continuous creative act.
Hollywood skipped most of the architecture because architecture makes terrible movie pacing.
I don’t get to skip it.
That’s the experiment.
Build the layers.
One by one.
⸻
Eventually I imagine useful robotics entering the environment.
Not humanoid servants wandering around pretending to be people.
Machines doing machine things.
A robotic arm holding an assembly.
A mobile platform moving material.
Machine vision inspecting a part.
Automated fixtures repeating a test.
Sensors collecting measurements humans would otherwise have to record manually.
And the closer computation gets to the physical world, the more disciplined the authority becomes.
A simulation can fail ten thousand times without hurting anyone.
A machine moving a heavy piece of steel cannot.
Intelligence should be able to explore widely.
Physical action requires boundaries.
That distinction is part of the architecture too.
⸻
If all of this works, something interesting happens to my own role.
I move upward.
Not above the work.
Into the part of the work that remains hardest to automate.
The strange question.
The intuition that something is wrong.
The judgment that one possibility matters more than twenty others.
Taste.
Purpose.
Responsibility.
The decision that something is worth investigating at all.
Maybe I stand at Dad’s old workbench someday.
Hands dirty.
A tablet beside me.
Half of something I haven’t named yet sitting under a work light.
And I say:
“Okay. Here’s something weird.”
Then everything capable of helping with that question begins moving.
⸻
That is the garage I’m trying to build.
It doesn’t exist yet.
Pieces of its architecture do.
Others are being designed.
Some may fail completely.
The final version will almost certainly look different from what I can imagine today.
Good.
Reality should be allowed to edit the dream.
But I know the direction.
I want a place where software can eventually touch steel.
Where research can become experiment.
Where experiment can become evidence.
Where evidence can change the architecture.
Where a book and a robot can emerge from the same underlying commitment to curiosity.
Where failure becomes memory instead of disappearance.
Where increasingly capable forms of intelligence remain connected to human judgment and physical reality.
Where the distance between wondering and finding out becomes a little smaller every year.
⸻
Sometimes I stand in that old building now and look around.
Same walls.
Same concrete.
Same garage.
And I think about how much of the future begins that way.
Not with something futuristic.
With an ordinary place.
An unfinished system.
A person who doesn’t yet know whether the idea will work.
And a question that refuses to leave.
The garage doesn’t need to prove that I was right to imagine it.
I want to build enough of it to find out.
That’s the experiment.