An AI Transition Field Guide

What is changing?

On February 3, 2026, roughly $285 billion left software, financial, and asset-management stocks in a single day, on the release of an AI tool, not on any change in reported numbers. The dispersion told the story: LegalZoom fell 20% and Thomson Reuters 16%, while Salesforce and ServiceNow fell about 7%. The market did not sort on "is this priced per seat?" It sorted on "can the model replace the need for people to do this work?".

For two decades software helped a human do the work. What is emerging does it. We call it Software-as-a-Worker: it changes what you sell, who you sell it to, and the budget you compete for. The labor budget is an order of magnitude larger than the software budget, so the move rightward is a prize, not only a threat.

Where am I exposed?

Incumbents are not doomed. They are trapped. Every asset you hold has a liability welded to its back.

Score seven questions 1 to 3 (pricing exposure, agentic readiness, data position, data rights, outcome accountability, org insulation, talent exposure) for 21 points. That lands you in one of four positions: Exposed (7-10), more fragile than the financials show; Instrumented (11-14), the most common and most deceptive, because the assets look strong so urgency reads low; Dual-Engine (15-18), a separate unit already running; or AI-Native (19-21), where the product works and the sales motion needs to catch up.

What do I decide differently?

Before any plays, four legitimate routes exist to execute your play: Build (fund from operations, keep the equity), Buy (acquire an AI-native Engine 2), Partner (be the data and execution layer under someone else's agent), or Harvest (run for cash and sell while the multiple holds). The route choice has a clock: Harvest and Buy both price off a seat-revenue multiple that is compressing.

When do I act?

There is more than one clock, and they run at different speeds. The revenue clock is slow and contractual (18 to 36 months to reprice at renewal). The talent clock runs out first: your seats are locked in, your people are not. The competitive and capability clocks you do not control at all.

What do I measure?

The old metrics measure the era you are leaving: DAU, feature adoption, seat count. Watch six internal signals quarterly instead. The first four tell you how fast the new engine is coming up; the last two, how fast the assets funding it are running down.

1. 01Executable core. Share of core jobs an agent can run with no human in the UI. Target: above zero, rising every quarter.
2. 02Work-metered revenue as a share of total. Target: 2 to 5% and growing.
3. 03Outcome-priced revenue as a share of total, tracked separately. Target: above zero in at least one segment.
4. 04Training-usable data captured under permissive terms. Target: rising month over month.
5. 05Latent attrition in the named critical population. Target: measured twice a year, trigger set in advance.
6. 06Engine 1 erosion: seat NRR and seat-heavy headcount at your top 20 accounts. Target: tracked, trigger set in advance.

What would prove us wrong

A view that cannot be disconfirmed is not worth much, so we watch our own falsifiers:

• Seat revenue holds through 2027 while seat-heavy headcount keeps falling. Then slow Play 1; keep Plays 2 and 4, which pay off anyway.
• Outcome pricing stays stuck in support deflection through 2028. Then stop at the metered stage; it is a complete model.
• Inference costs fall far enough that COGS stops binding. Then accelerate; this is the pleasant way to be wrong.
• Your own top-20 accounts show no seat-heavy decline over four quarters. Then you have more time than the thesis implies; re-run it every quarter.

If none of your signals ever move against you, you are almost certainly measuring the wrong things.

What do we carry forward?

The assets that make the crossing possible are the ones you already own. The job is not to escape the old model; it is to take apart its liabilities without breaking the assets attached to them. Two must survive: your data, the fastest-moving asset and the one you cannot buy back, and your people, the only asset that can resign.

Your strongest people leave for AI-native challengers you cannot out-pay, so stop competing on comp and compete on what they cannot get there: real domain problems, real customers, and a decade of proprietary data to build against. The quieter risk is the stayers, frustrated on the core and staying only because the market is soft. Gallup put the share of US employees watching or seeking at 51%, even as actual turnover held steady: deferred attrition, on a clock you do not control.

The answer is a talent contract, its commitments declared to the whole company as much as possible rather than whispered in the halls:

1. 01A published route into Engine 2, run on a cadence with a return path, so a rotation is not a one-way career bet.
2. 02A real Engine 1 mandate: rebuild work, not maintenance. Fund it and say so, or your engineers will know inside a sprint.
3. 03A skills guarantee with hours, budget, and production work attached. What they want is not to be obsolete in three years.
4. 04The truth about the roles that shrink. People leave faster from a plan they suspect than from a hard one they are told.
5. 05A named critical population, kept internal. Treating everyone as critical protects no one; name the people who pair domain depth with the ability to build agentic systems, know what each is optimizing for, and review the list quarterly at executive level. You declare the discipline, not the names.

In every platform shift, the value goes to whoever builds the trains, not the rails. As an incumbent you already own the track, the stations, and the passengers. The only question is whether you put a train on it before someone else runs one down your line.

The mistake we watch companies make is not moving too slowly on AI features. It is moving fast on features while leaving the business model, the architecture, and the org chart exactly as they were. That produces a Type 1 company: busy, full of demos, and structurally identical to the thing being disrupted. It is SaaS with better marketing. It is not SaaW.

The six questions are the whole method, in order: know what is changing, know where you are exposed and where you are strong, choose your route and run the plays it calls for, respect that the clocks move at different speeds, instrument the transition, and protect the assets that make it possible. The incumbent who reaches the far side with its data, its trust, and its bench intact is the one the challengers were never able to become.