The capability-adoption gap is the principal's window

In early February, during a back-office diligence on a mid-market services business, we ran two numbers against each other. The in-place SG&A run rate. And the SG&A run rate that would hold if the finance, customer-service, and compliance functions were rebuilt on a modern AI-native toolchain, operated by a small internal team. The gap was thirty-two percent of the in-place line, on a twelve-to-eighteen-month horizon. The asset was priced on the in-place line. The next buyer will price it on the rebuilt one.

That arithmetic holds, to varying degrees, across most mid-market operating platforms we are seeing in diligence this year. What follows is a principal’s seat view of why the arithmetic is open, what creates it, and how the window closes.

What shifted

Three observable facts, independent of any vendor claim.

METR’s public task-duration measurements, the length of real-world work a model can complete end-to-end without human correction, have tracked a sharp rise across 2025. The original series, published as Kwa et al., Measuring AI Ability to Complete Long Software Tasks (arXiv:2503.14499, March 2025; accepted at NeurIPS 2025), documented a seven-month doubling time across the 2019-2025 window. METR’s Time Horizon 1.1 update (29 January 2026) shortens the post-2023 doubling to roughly 4.3 months on the revised task suite. The doubling time sits somewhere between three and seven months, depending on the window chosen. Models released in early February 2026 sit outside the published curve pending fresh measurement. Auditors using them seriously read the jump as another step in a sequence they have come to expect.

The frontier labs have been visibly candid about using the latest generation to accelerate work on the next. OpenAI positions GPT-5.3 Codex as an agentic software-engineering tool. Senior Anthropic staff have spoken publicly about Claude writing material portions of the code at the lab. Whether that amounts to intelligence explosion or an incremental productivity shift is a separate question. The observable effect inside code we look at is consistent either way: the rate at which usable capability arrives is decoupling from the rate at which humans author it.

Institutional procurement, meanwhile, runs on contracts, renewal cycles, and board approvals written against last year’s capability. Most mid-market operating platforms, and most of the service providers underneath them, are running on tools eighteen to twenty-four months behind. That lag widens every quarter the frontier accelerates and the procurement cycle stays the same length.

Why the lag is structural

The useful question is why it persists. A lag that short would close fast if closing were cheap.

It persists because the procurement cycle is structural. A mid-market operating business with a three-year board plan, a two-year software-licence renewal, and a two-year chief-operating-officer tenure moves on a slower clock than a ninety-day restructure requires.

The restructure costs political capital a CEO holds in small supply at that point in a tenure. The quarterly-earnings arithmetic posts the severance line before the savings line. Listed mid-caps feel this the hardest. Private operators that have kept the same COO for a decade feel it less. Family-office-held platforms with a patient operating frame feel it least.

The principal’s advantage in this cycle is that every one of those frictions falls away at principal level. A family office setting the twelve-quarter cost base, a founder-led fund re-underwriting an acquired platform, or a patient principal restructuring a recently acquired asset is operating in the one seat from which the restructure is politically available and financially rewarded inside the same twelve months.

Where the asymmetry sits

Four places in the next twenty-four months, from audit work.

Inside operating expense, as upside. The SG&A line of a mid-market services, back-office, or compliance-heavy platform is the immediately addressable surface. The question is what share of the current base is structurally re-scopable on a twelve-to-eighteen-month horizon, operated by a small internal team on a modern toolchain. Across the diligences we have run in the first quarter of this year, on a small, hand-selected sample, the observed range sits between twenty-five and thirty-five percent of total SG&A for services businesses. Industrial platforms and regulated operators come in lower, between ten and twenty percent, bounded by compliance perimeter rather than capability. The numbers are specific to the deals in front of us and belong read that way, as sample evidence rather than a general market claim. A buyer pricing the acquisition against the in-place line is underwriting the upside directly. A principal already holding the business, and holding the 2024 cost base through this cycle, is leaving the spread on the table for the next buyer.

Inside acquisition comparables. A platform bought eighteen months ago at a labour-heavy multiple is a different asset today. The platform stayed the same. The buyer’s cost of operating it changed. A principal running a revised multiple against the post-restructure cost base is pricing the asset correctly. The market comp set, anchored to last-cycle assumptions, still reads the old price. In the deals we have seen close since January, the spread between the two prices runs between fifteen and twenty-five percent of enterprise value. That is the negotiating range.

Inside long-term service contracts. Any multi-year contract with an embedded labour-cost assumption, whether IT managed services, facilities management, customer-support outsourcing, or middle-office functions, is structurally mispriced against the counterparty’s real cost base. A principal paying that contract is funding a margin the counterparty has already released. The renegotiation case writes itself, once the principal runs the math before the renewal window opens.

Inside capital events over the next twenty-four months. Refinancings, recapitalisations, exits, platform acquisitions, large procurements. Each event is priced either against the current frame or last year’s. A decision against the current frame captures the asymmetry. A decision against the old frame hands it to the counterparty. The difference locks in when the event closes.

The range of outcomes

The window is open for somewhere between two and six more quarters before the market comp set catches up. Four factors will compress it. Published KPIs on named restructures. A marquee sell-side transaction priced on the post-restructure cost base. Rating-agency methodology updates on labour-cost assumptions in services. The first Fortune 500 PR campaign claiming a fifteen-percent SG&A reduction.

If the frontier plateaus this year, the window closes on the existing gap. If the frontier continues at the current rate, the gap widens, and a larger arbitrage opens against platforms that have already completed the first restructure. Across either path, the principals who act in the first window own the cost base from which the next cycle runs.

The posture for the next four quarters

Four decisions are worth running against the portfolio this quarter.

1. Which portfolio operating businesses carry an SG&A line that is structurally re-scopable, and what is the delta between the in-place base and the rebuilt one. The internal number, held separate from the vendor pitch.

2. Which platform assets, held directly or acquired in the last two years, were priced on a labour constant that has moved since. What does the revised multiple look like at exit, and what would trigger the market comp set to agree.

3. Which long-dated service contracts are structurally favourable to the counterparty today, and on what timeline does the renegotiation window open.

4. Which capital events over the next twenty-four months coincide with the window in which the frame has moved and the market still reads the old price.

These are capital questions first, technology questions second. Each is answerable inside a principal’s quarter, once the work is held across capital, technical, commercial, and operational frames at once, at principal level.

The window is open. On the working assumption, it stays open for between two and six more quarters. The principals who do the arithmetic now own the cost base of the next cycle.