re:compoundFor the people who sign the work
AI made professional work fluent before it’s reviewable. re:compound makes it reviewable again — a ten-week installation for one senior reviewer and one producer, delivered by the founders, inside the AI environment your firm already approves.
The installation is three things: a working method for the pair, plain-language instructions placed inside the AI environment your firm already approves, and the shared documents that travel with the work and show what it rests on.
Ten weeks. One reviewer, one producer. Their real work. Fixed fee.
Year one: nine completed installations. All nine chose to extend.
“I can’t find anything wrong with it. It all looks good, I guess.”
Where this practice was built
Built from the founders’ careers — not re:compound engagements.
01 · Why review broke
Senior review used to read more than the content. Roughness, seams, hesitation, and the producer’s account of the path told an experienced reviewer where the thinking was strong, thin, or not yet owned. AI decoupled those signals from the work: fluency now arrives before understanding, and the more capable the tools get, the easier it becomes to do the wrong work excellently.
Your producer’s first take goes into the model and comes back polished, confident, and agreeing — a first guess wearing the appearance of external validation. Each side defers to the other’s authority, and the independent signal that review used to run on is flattened before the work ever reaches you.
The collaboration produces a decision-shaped artifact almost immediately. From then on, the team iterates the words rather than the understanding — polishing an answer to a nearby problem while the artifact acquires an authority it never earned.
You inherit the fork: reconstruct the inquiry yourself, or sign what you cannot efficiently inspect.
“I don’t have a feel for the labor cost numbers. But they fit. I approved the memo.”
“Who owns the restructuring plan end to end? Can I even ask that?”
“It looked right, so I sent it up. I couldn’t have told you where it was thin.”
Said to us — from both sides of the review boundary
Work that is polished and apparently complete, but whose evidence, assumptions, uncertainty, alternatives considered, and human ownership cannot be efficiently inspected.
02 · What this is
An engagement delivered by the founders that installs a disciplined way of producing, reviewing, and relying on AI-native work in one senior working relationship — at many firms, a partner and an associate — on live client work they already have. The fee is scoped and fixed up front — if the ten weeks slip, the fee doesn’t — and sized to be absorbed by the budget of the work the pair is already doing.
The moves the pair uses to frame the real problem, steer the model without deferring to it, decide when the inquiry has earned an artifact, and run a review that changes the work. Taught inside live deals, cases, and client work under real deadlines — one to two hours a week of structured time; the rest happens inside work already underway.
A shared surface that travels with each deliverable and shows the reviewer what the artifact rests on: the question being answered, the claims bearing load, the evidence and its provenance, the alternatives still live, and what the producer actually owns. Generated by the producer’s AI from the production work itself, then corrected and owned by the producer. It falls out of how the work is produced — which is why the template alone is worth little, and why what we install can’t be copied from the artifact. It’s shown in full below.
Plain-language operating frameworks the pair’s AI runs — versioned and maintained like software, installed as instructions inside the AI environment your firm already approves. Nothing new for IT to procure, integrate, or security-review. Licensed to the firm — and it stays after we leave.
Weeks one and two, we observe. Through week seven, we run it with the pair on their live work. By week ten, they run it without us — that is the acceptance test. Habits built under deadline, uncertainty, hierarchy, and real stakes.
At week ten you either extend, as every completed installation so far has, or we part ways and everything installed stays.
04 · The review event
Before the conversation, the reviewer’s own AI reads the artifact, the State of the Work, and the native evidence — applies the reviewer’s private lens, which is never transmitted to the producer — and prepares a Partner Attention Brief: the conversation most likely to improve the work, the decision, and the producer’s judgment. Senior attention stops being spent reconstructing smooth artifacts and goes where it has the highest expected value.
Illustrative example — fictional company and data. Generated by the reviewer’s AI from the artifact, the State of the Work, and the native evidence — the reviewer’s lens applied, never transmitted. Each brief adapts the four review modes to the work at hand — here as Strongest, Understand, Pressure-test, Decide, and Develop.
The inquiry moved the evidence bar from site averages to cohort-level cuts and killed the price-realization trunk cleanly. Name that discipline before applying pressure.
What in the site-level data first raised the mix suspicion? The producer may have contact here the memo doesn’t carry.
A2 reads more confident than its support: the payroll sensitivity is inferred, and the artifact’s language is firmer than the verification status. Open E07 in native form.
Q03 is the call only the reviewer can price: two days for the two-site cut, against a Thursday committee. The producer asked for exactly this decision.
If the trunk holds, transmit the distinction doing the work — structural vs. mix-driven stability — so it travels into the next deal.
Ranked and short — a long list of clever concerns is not governed attention.
Ask the genuine question the surface can’t answer: why the retention evidence changed the frame.
Add the comparison, counterfactual, or reframe that only experience supplies.
Inspect the load-bearing claims — and the native evidence beneath them — directly.
Transfer the judgment distinction behind the direction, not just the correction.
AI removed many of the task reps through which junior judgment used to form. The review loop is the redesigned apprenticeship: the producer practices governing an inquiry, marking evidence and uncertainty, and defending a frame under pressure — and each review transfers a named judgment distinction they carry into the next piece of work. Honestly surfaced uncertainty is treated as governance, not weakness — which is exactly what makes it rational to surface.
A review can be exacting — kill a weak branch, reject unsupported work, require direct verification — and still create value when it finds no defect: a sharper distinction, a better next interval, a producer who owns more than they did an hour before. And the direction travels both ways; the system preserves the possibility that the producer teaches the reviewer.
The production system earns the work. The review system earns reliance on it.
05 · Why we’re called re:compound
Most firms end each engagement with a document archive. The installation ends each cycle with three balances that carry forward.
What changed, what remains live, which distinctions travel. The next return starts higher — not from a folder of final versions.
What mattered, what didn’t, where the model over- or under-weighted. The reviewer’s judgment gets a longer reach each cycle.
Capability accrues to your people — not to a vendor, and not to a tool.
06 · The first engagement
Ten weeks. One reviewer and one producer — at many firms, a partner and an associate. The titles vary; the recurring working relationship is the unit, because that is where consequential work is framed, produced, challenged, relied on, and learned from.
The live work, the review moments, the reviewer’s interventions, the AI use, and what the culture rewards or suppresses.
Configure the assets to the firm’s work and language; run live production and review with the pair. One to two hours a week of structured time — the rest happens inside the work itself.
Support withdraws as the pair’s independent capability rises, until the discipline is simply how they work.
07 · The record — and the exit
In our first year, eleven organizations across mid-market private equity, law, and specialty advisory have started this installation. Nine have run to completion — and all nine chose to extend: additional partners, additional producers under the same partner, or a firm-wide phased program. Two more are in flight now. We deliver personally and take a small number at a time, so every one begins the same way: a conversation to establish fit.
The assets stay. The capability stays. re:compound leaves.“I feel I can sign my name to the work again. I know where it came from and I know my team owns the results.”
08 · The firm
One in the architecture of expert judgment — McKinsey, Bridgewater, Deloitte. One in clinical operations, where unanswered questions, unclear ownership, and failed handoffs carry consequences. Every installation is delivered by the two of us.
About re:compound
A practical first step
A thirty-minute conversation with Bud is the first step: the pair, the work, the timing. No deck, no demo environment, no discovery process — book it directly on his calendar.
Talk to us →