AI in private equity · the story of Galt Partners, a fictional firm
Adoption is a behavior curve, not a tooling curve.
The firms that advance are not the ones that buy better tools — they are the ones that change how they work.
The frame · electrifying the factory
Electricity took decades to matter. The motor wasn't the bottleneck.
Factories swapped steam for electric motors — and kept everything else.
The steam factory
One steam engine turned a central crankshaft; belts and pulleys carried the rotation to
every machine on every floor. Everything ran in sync — when the engine
stopped, the mill stopped. The building was vertical because of the shaft: machines
arranged around the power, not the work.
Point solution
The steam engine is swapped for an electric motor — on the same
shaft, in the same factory. Cheaper power; nothing else changes.
Application solution
The unit-drive workstation — each machine with its own motor, independent of the shaft.
Workers could now work asynchronously, no longer timed to the central
crankshaft — but still inside the old architecture.
System solution
Once work ran asynchronously, it could be organized around the
workflow — the horizontal factory: machines where the work wants them, materials
flowing along an assembly line. This is where nearly all the value was.
The horizontal factory came decades after the electric
motor was available. The bottleneck was never the motor — it was the willingness to rearchitect
around it.
The maturity curve
The axis that moves you is behavior.
Every firm is somewhere on this curve — note what the x-axis measures.
Step 0Point solution
Shadow AI
Everyone at Galt already uses AI — hidden, personal, unshared.
Officially the firm hasn't adopted it. Unofficially, it runs on it daily.
Behavior
Hidden, federated individual use — nothing shared, nothing compounding.
Tooling
Personal ChatGPT accounts, personal notetakers — confidential deal data in consumer accounts the firm can't see, doesn't control, and couldn't produce in an audit.
Still the vertical factory — the motors are just hidden under the desks.
Step 0.5The detour
Tool exploration — the detour
Galt responds to AI by shopping, not by changing how anyone
works. Impressive demos, stalled pilots, decaying usage.
Behavior
Unchanged — the purchase substitutes for the behavior change. Spend rises; the curve doesn't move.
Tooling
ToltIQ, BlueFlame, Rogo, Hebbia — wrappers on a frontier model; effectively token resellers.
Still the vertical factory — shopping for motors.
Step 1Point solution
Early adoption
A workhorse and a notetaker, out in the open — and
workflows start to crack under the new behaviors AI enables.
Behavior
Open, emergent experimentation — what improves, what's newly possible, what newly breaks.
Tooling
A workhorse and a notetaker — enterprise accounts, sanctioned, secured, firm-paid.
A better motor on the same shaft — the factory hasn't changed yet.
Step 2Application solution
Proficiency
The application solution, lived — shared knowledge bases,
skill files, and a working literacy of what models do what.
Behavior
Context is shared, common workflows are automated as skills — a shareable .md file — and the right model goes to the right job.
Tooling
The same core tools, deepened — a knowledge base per deal, connectors wired, a skill library accumulating.
Unit-drive workstations — new patterns, still inside the old building.
Step 3 · the final stepSystem solution
System rearchitecture
Not "where can AI help our process?" but "what should the process
be, given AI?" The firm rebuilds around abundant intelligence.
Behavior
Humans hold judgment, questions, conviction; AI carries search, summarization, volume.
Tooling
Whatever the redesigned process requires — chosen after the process is designed.
The trap
Waiting. Rearchitecting is uncomfortable — and the firms that move first compound
the advantage.