From FR Advisory.
A declaration of standing, because it determines how you should read what follows. I spent twenty years running commercial organisations inside SaaS, enterprise technology and private equity portfolios. I no longer work in the technology industry, I hold no position in any company named in this series, I have no clients in it, and I am not selling anything to anyone in it.
That is why the names are in it.
In 1899, electric motors turned less than five per cent of the machines in American factories.
Edison had opened central generating stations in New York and London in 1881. Eighteen years of available power. Eighteen years of installed technology. Less than five per cent.
Not because the electricity failed. Because of what factory owners did with it.
A nineteenth century factory was built around a single steam engine. The engine turned a line shaft running the length of the building, and belts dropped from that shaft to every machine. Everything about the building followed from that one constraint. Multi-storey, because you stacked floors around the shaft. Cramped, because proximity to the drive was everything. Machines positioned by where the belt reached, not by the sequence of the work.
When electricity arrived, owners bolted electric motors onto the shafting that was already there. It was cheaper than rebuilding, and it required nobody to admit the building was wrong.
They kept the shafts. They kept the layout. They kept the building. They changed the power source and nothing else, and for two decades the productivity statistics showed almost nothing.
The gain only arrived when somebody put a motor on each machine, tore out the shafting, and laid a single storey factory out around the flow of work rather than the geometry of a dead engine. Productivity growth in American manufacturing accelerated by about five percentage points in the 1920s against the decade before, and roughly half of that is attributable to that one change.
Four decades between the technology and the return.
Every hour of that gap was operating model.
You have bought the motor and kept the shaft
MIT studied enterprise AI initiatives and found ninety five per cent returning nothing measurable.
That number has been quoted into meaninglessness this year, usually by people selling the remedy. So let us be precise about what it actually is.
It is not evidence that the technology does not work. It is the 1899 factory, measured.
Here is what your industry has done in the last eighteen months. Appointed Chief AI Officers. Rewritten roadmaps. Stood up centres of excellence. Run pilots in every function. Cut headcount to fund the licences. Renamed the product with the letters A and I in it. Briefed the board. Briefed the analysts. Put it in the earnings call four times.
Now name one thing on this list that changed.
How value is priced. How consumption is metered. What the sales organisation is paid for. Who owns expansion. What metrics are handed to the market. How revenue is recognised. Whether anyone can prove what the customer actually realised.
Those are the shafts. Every one of them is still bolted to the floor, and every one of them was designed for a business that sold seats to people who did the work themselves.
You changed the power source. You changed nothing else.
The evidence, with names
This series will name companies, because a thesis that cannot name anyone is an opinion.
In one week in October, a pricing tracker recorded two hundred and thirty pricing changes across two hundred and twenty two companies. Twenty of them were price rises. Roughly two hundred were something else entirely, and that something else was almost always moving what the money buys rather than what it costs.
GitHub cut what a Copilot Enterprise seat includes by forty four per cent on the first of September. The price stayed at thirty nine dollars. That is a forty four per cent price increase that appears on no price list, because procurement checks prices and nobody checks exchange rates.
Figma took its agent out of beta onto a shared credit budget spanning three products, where the cost of a run varies with the prompt and no rate card exists for what a credit buys.
Surface Labs cut a three and a half thousand dollar plan from fifteen thousand agent actions to five thousand. The price held.
Centrical launched a per user licence plus a shared credit pool. Two value metrics in one contract. The old model with the new one bolted on, shipped deliberately, as a product decision.
Harvey, the best funded application company in legal AI, built its own model, scrapped it when the frontier labs outperformed it, and went shopping for primary law instead.
Salesforce concluded it would not own the intelligence layer, made Claude a preferred model inside Agentforce, and kept the trust boundary, the permissions and the customer record. The correct decision, and an admission that the application is no longer the defensible asset.
And in October, ServiceNow grew agentic AI consumption thirteen fold, raised its AI contract value target from one billion dollars to one point seven billion, beat on revenue, and lost twenty seven per cent of its market value in a day, the worst in its history, because a subscription era backlog metric guided one and a third points light.
Read that last one twice. The company doing the most on agentic AI was punished the hardest, by a measurement system built for the model it is trying to leave. Its chief financial officer said on the call that the metric does not necessarily capture the momentum in the business. The market sold anyway, because that is the number it has.
In the same quarter, its chief executive told investors customers like the predictability of seat based pricing and that token economics could improve revenue per user five fold. Revenue per user. The denominator is still a person. And ServiceNow's own headcount fell, on its own internal AI efficiency.
A company proving seat compression on itself, while selling seats, while being repriced for the one part it got right.
That is the whole of this era in a single earnings call.
The capital has started to move, and it is moving against you
Boards have been able to ignore this while the money stayed polite. The money has stopped being polite.
Public enterprise software multiples are down roughly forty eight per cent from 2024. Private software is still clearing at eighty seven to ninety two cents on the dollar of net asset value. Thirty eight percentage points separate what the public market says these businesses are worth from what private holders carry them at, and only one of those two marks is set by the seller.
Private credit lent to software companies grew from under eight billion dollars in 2015 to over five hundred billion by the end of 2025. Nineteen per cent of all direct lending. One in three private credit funds is in it. Every pound of it was underwritten on the belief that recurring revenue is safe collateral, which is a statement about the commercial model and nothing else.
One point two trillion dollars is stuck in zombie private equity funds, up thirty eight per cent in a year. Over a quarter of private equity's software portfolio companies were bought five or more years ago and cannot be sold. Thirty per cent of limited partners now describe the assets being moved into continuation vehicles as distressed or challenged. Partners Group has started capping application software exposure at ten to fifteen per cent when it buys fund stakes.
Exits are not being delayed. They are being abandoned, quietly, one investment committee at a time.
The people who held the view that technology adoption was the variable are revising it in public. They are arriving at the same conclusion from the opposite direction, and they are arriving with capital rather than opinions.
Why this is existential rather than difficult
There is a comfortable reading available, and leadership teams have been taking it all year. It goes: the technology is early, the returns will come, we are investing ahead of the curve, hold the line.
Electricity was early for forty years. The factories that waited did not get a late reward. They got bought by the ones that rebuilt.
The pattern does not vary. Steam, electricity, the internet, SaaS. Four times the technology arrived first. Four times the operating model arrived years later. Four times the companies that died were not the ones with worse technology. They were the ones that installed the technology and preserved the model, because preserving the model required admitting nothing, costs nothing this quarter, and is always defensible in the room.
What makes this cycle faster and more brutal is that the customer now has the same technology you do. In every previous transition the buyer waited for the vendor. This time the buyer can build the thing, and every seat you sold is a line item your customer is actively working to remove.
So the clock is not forty years. It is closer to four.
The agentic commercial model is not a proposal. It is a destination. Pricing on realised value, metering as core infrastructure, compensation on outcomes, customer success as proof rather than renewal management, revenue architecture built for variable consumption, and an organisation flattened because the work between the data and the decision has gone.
Every company that survives will arrive there. The only live question is whether yours arrives under its own direction or under new ownership.
What this season will do
Eight issues. No scene setting, no balance for its own sake, and no vendor spared.
01. You Changed Everything Except the Business Model. The indictment, in full, with the receipts.
02. The Dynamo Gap. Four transformations, one pattern, and the graveyard it produces every time.
03. The Punishment. ServiceNow, and the trap underneath it. The market currently has no instrument that rewards the crossing, which means cowardice is the dominant strategy and it is priced accordingly.
04. The Capital Turns. Marks not taken, exits cancelled, the thirty eight point gap, and what happens when half a trillion of credit meets a revenue line nobody believes in.
05. The Scoreboard. ARR, NRR and cRPO were built for subscriptions and are now actively mispricing the era. What replaces them.
06. The J-Curve. If you cross, this is the order, and this is what breaks in which quarter.
07. Built Around. Leaders are not being removed. They are being routed around, and they are always the last to know.
08. Inevitable. What arrives on the other side, and who is on it.
The position
Your AI is not the problem. It has never been the problem.
The problem is that you have spent two years and a great deal of money bolting a new power source onto shafting laid down for a business that no longer exists, and calling the result a transformation.
The factory owners of 1899 were not stupid. Bolting the motor on was cheaper, faster, lower risk, and it let everyone avoid the conversation about whether the building was wrong. It was the sensible decision every year, right up until the companies that tore out the shafting took the market.
You are making the same decision, for the same reasons, with better slides.
Change the model, or be bought by someone who did.
The Agentic Commercial Model. agenticcommercialmodel.com
Sources
- Paul A. David, "The Dynamo and the Computer: An Historical Perspective on the Modern Productivity Paradox," 1990: Edison central generating stations in New York and London from 1881; electric motors at under 5% of factory mechanical drive in 1899; slightly over half of mechanical drive capacity by the early 1920s; manufacturing TFP growth accelerating roughly 5 percentage points in 1919 to 1929 against 1909 to 1919, around half attributable to secondary electric motor capacity; the group drive and unit drive distinction and why sunk shafting delayed the rebuild
- MIT research on enterprise AI initiatives reporting 95% with zero measurable return
- PricingSaaS Pulse, week to 11 October 2026: 230 pricing changes across 222 companies, 20 increases, 9 cuts
- GitHub Copilot AI credits change, 1 September 2026: Enterprise entitlement cut from 7,000 to 3,900 credits, seat price unchanged at $39
- Figma AI credit updates, agent generally available from 6 October 2026: shared credit budget across Figma Make, the agent and Weave tools, cost varying with prompt complexity, no published per credit rate
- Surface Labs published pricing: Agentic Marketing Operations at $3,500 per month including 5,000 agent actions, reduced from 15,000 per pricing page tracking, $0.20 per additional action
- Centrical pricing change, October 2026: per user licences plus a shared AI credit pool
- Facts Are Friends, "The Data Giants' Dilemma": Harvey abandoning its proprietary model after frontier models outperformed it
- Anthropic, "Salesforce and Anthropic expand partnership," 14 October 2025, and Salesforce Q2 FY27 earnings
- ServiceNow Q3 2026 results and earnings call, 28 October 2026: revenue $4.106bn up 20.5%, subscription revenue $3.988bn up 21%, AI net new ACV up 65%, 2026 AI ACV target raised from $1bn to $1.7bn, agentic AI production consumption up more than 13x year on year, Q4 cRPO growth guided at 18.5% against approximately 19.8% expected, shares down approximately 18%
- PitchBook, 12 May 2026: median public enterprise software EV/TTM revenue down 47.7% from 2024; secondaries clearing at 87% of NAV per Jefferies; Partners Group capping application software exposure at 10 to 15% when buying LP interests
- McKinsey Global Private Markets Report 2026: GP-led secondaries pricing at 92% of NAV; 30% of LPs describing continuation vehicle assets as distressed or challenged
- Bank for International Settlements, BIS Quarterly Review, 16 March 2026: private credit lending to SaaS from under $8bn in 2015 to over $500bn by end 2025, 19% of direct loans, one third of funds exposed
- Institutional Investor, 28 July 2026, citing Treo Asset Management: $1.2tn in zombie private equity funds, up 38% year on year
- S&P Global Market Intelligence, 6 May 2026: over 27% of private equity software portfolio companies acquired five or more years ago