For private equity
Where the value sits in a portfolio.
Reading AI readiness across companies you already own, in a form that compares. A company can only answer these questions about itself. A fund can answer them across everything it holds.
Why now
Four challenges inside the hold period.
A value creation plan written two years ago treated agents as a cost line inside the IT budget. By the midpoint of the hold they decide whether the margin case, the retention case, and the exit multiple survive contact with the market. Two of the four challenges below are already inside the portfolio. The other two arrive through the pricing unit and through every buyer's diligence at exit.
Each one is a race between the pace a company can move and the control it can prove. Each is decided in the operating model rather than the tool stack, so a portfolio company cannot buy its way out with the tools its competitors are buying.
Ungoverned agent sprawl is a self-inflicted risk to the thesis.
Under a cost-out mandate, agents arrive wherever a manager can show a headcount saving, ahead of any shared control. Within a quarter they sit on the repo, on customer data, and on the systems of record, with no shared identity and no record of what any of them did. The failure lands where it costs most, as a customer data incident during a sale process or a regulatory finding that stalls an add-on.
Uber spent its full 2026 AI budget by April and published the governance it had skipped by May. A portfolio company carrying acquisition debt has neither the balance sheet to absorb that overrun nor the platform team to build the fix. Book the saving once the evidence shows the work is safe.
Competitors will turn agents into a widening execution gap.
The competitive unit has moved from the feature to the speed of the learning cycle behind it. The gap compounds, because every faster cycle returns more feedback and more reusable knowledge for the cycle after it. The niche the deal was underwritten on buys less protection each quarter of the hold.
Retail ran the experiment first. Shein reads demand in days and reorders only what sells, while the traditional apparel chain still runs six to nine months from forecast to fulfillment. That cycle re-sorted a global category before agents existed. Agents now bring it to the back-office operations where most margin plans live.
AI-native entrants will attack the pricing unit.
A new class of company reaches one hundred million in revenue with a few dozen people, inside two years, AI-native from the first commit. Each generation ships leaner than the one before.
For a portfolio company that sells software, the pressure lands on the seat. Seat-based revenue assumes a human behind every license. When a customer's agents absorb the work of five licensed users, per-seat revenue compresses at renewal while the product stays fully competitive. The erosion shows up in net revenue retention two or three quarters before it appears in a lost deal, and at exit as a lower multiple on a revenue base the buyer no longer trusts.
Frontier-model providers are a longer-term threat to the exit.
Eighteen to thirty-six months out, inside most current holds, the frontier labs appear to have entered a cycle in which each model accelerates the next. As intelligence commoditizes, the providers move up into the application layer. In January two labs entered healthcare within a week of each other. In April one of them shipped a design product into the market of a company it supplies.
At exit the buyer will run one test. Strip the rented model out and ask what remains. A user interface and a customer list price as exposed. A domain model, a governed execution system, a proprietary evaluation corpus, and the record of what worked carry the multiple.
Govern execution so speed is safe, and prove outcomes so scale is earned. Above the models, build the layer the company owns, so the models stay replaceable underneath it. A company that builds that system in year one compounds on it through exit. One that builds it after an incident spends the rest of the hold on remediation.
Three tests for the next board meeting- Which agents are running in production, who owns each one, and what evidence exists of what they did last month
- What NRR looks like by cohort once seat counts are separated from price
- What the company would own if its model supplier became a competitor
A target that answers all three earns a premium. A portfolio company that cannot is a workstream. The four questions below are how a fund reads that workstream across everything it holds.
Where to start
Four questions the portfolio has to answer.
Each one is answerable from evidence rather than from a status update, and each one reads across the whole portfolio.
Where does value sit above the model layer, and where underneath?
Exposure is a position that can be read now, not a date to be predicted.
Which companies are actually ready, and which only look ready?
Confidence and capability produce very different readings.
Who inside a company has to own this for it to survive?
The wrong sponsor is the most common cause of a stall.
What gets built once, and what has to be built inside each company?
Some of this is portfolio infrastructure. Most of it is not.
Question one
Exposure is a position, not a forecast.
Every company sits somewhere relative to the models. A company building above them becomes more valuable as the models improve. A company selling what the models are about to do for nothing becomes cheaper on the same schedule.
How to read it
Ask what a company would still own if the model layer absorbed the feature it sells today. What remains is domain-specific data, embedded workflow, and customer trust. Those assets get stronger as the models improve. A feature does not.
Why it concentrates
Every software company in a portfolio meets this at the same time. Holding many companies does not spread the exposure the way it spreads most other risks, which is what makes it a portfolio question rather than a company one.
Question two
A grade that reads the same way in two companies.
Readiness is graded against evidence across eight domains. A company sits at its weakest domain rather than its average, which is why a strong showing in six of them can still leave a company at the bottom of the scale.
Weakest gate sets the grade
Measurement on its own is common now. A grade that reads the same way in two different companies is not, and it is the part that turns a status update into an allocation decision. It also separates a company that cannot, which is missing capacity, from one that will not, which is missing a sponsor.
Where the grade usually breaks
Security is the domain that stops a rollout, and the one where companies grade themselves highest.
Most companies can name the model they use. Fewer can name the identity their agents hold, or say what that identity can reach.
A company that cannot answer this will pass its own internal review and fail its customer's. In a portfolio, that shows up as a deal cycle that lengthens without anyone being able to say why.
Question three
The wrong sponsor is the most common cause of a stall.
Below a certain revenue scale, nobody is accountable for the operating infrastructure inside a company. AI then defaults to the product organization, where it is measured on what ships, and internal work loses that contest every time. These are conditions for a value creation plan rather than terms in a vendor conversation.
What to insist on
- A sponsor who can change how the company works
- Named dedicated people, confirmed in writing
- A team that is not also shipping the product
- One workflow chosen before any build begins
What we have seen
Protected time on top of a full role does not move a schedule. Named capacity does. Named capacity is the single largest predictor of whether the work lands, and that finding came out of live delivery rather than theory.
Question four
What travels between companies is method, not data.
Built once, held by the fund
- The readiness grade and how it is scored
- The enforcement pattern and the shape of policy
- The proof standard, so results compare
- A pattern memo showing which constraints repeat
Built and owned locally
- Business data, which never leaves the company
- The workflows themselves
- Systems of record, which survive as inputs
- The people who own the intent
Nothing about one company's business moves to another. Under a mutual non-disclosure agreement, each company's materials remain their own, and nothing moves between companies unless the owners approve.
How we engage
Three offers, in sequence.
Each is priced on its own and nothing later is automatic. The fund decides every next step on the evidence from the step before it.
The briefing
Half a day for product and engineering leaders across the portfolio. No preparation and no environment access. Teams leave with shared vocabulary and a rubric for locating the constraint in their own process.
The clinic
Up to six companies at a time, three people from each. Every company brings one capability heading into delivery, writes it against the same minimum contract, and leaves with a finding and an action list.
The pilot
One company and one live capability, run end to end through that company's own process. It closes on agreed evidence and a readout that separates what was observed from what was inferred.
The fund funds the part that produces the comparison. The company funds the part that changes how it works, because only the company can carry that change.
Where we sit
The layer underneath the build.
Many funds now have engineers embedded across the portfolio. We do not do that job, and we do not offer a competing version of it.
Build capability
- Sprints, cohorts, prototypes, shipped features
- Produces output
- Measured by what got launched
- Moves fastest at the start
Governed substrate
- Enforcement, record, proof, comparable readiness
- Produces evidence
- Measured by what can be repeated
- Compounds over time
The substrate is what keeps that output durable and comparable once several teams are moving at once.
What is actually at stake
Which of your companies does the work, and which becomes the data feed?
Agents are in production and the work is moving to whatever software can carry it. Some products will do that work. Others will be read at the API layer by one that does.
Every company you hold meets this at the same time. It is the one exposure a portfolio cannot spread.