The Short Answer
Don't rip out per-seat pricing. But don't sit still either. The right move right now is a layered model, piloted on new deals, with a clear trigger for when to go further. The advisors disagree mainly on how aggressively to layer — and that disagreement actually resolves cleanly once you know one thing about your own business: is AI's value uplift uniform across your customers, or wildly uneven?
That single question is the hinge the whole decision turns on, so start there.
1. Resolving the Real Disagreement
Three advisors (Architect, Philosopher, Explorer) converge on "layer a usage/outcome component on top of seats." One (Maverick) says "just raise the seat price, don't add complexity." These aren't actually opposed — they're optimized for different situations:
- If AI lifts value roughly evenly across your customer base (everyone gets a similar productivity bump), Maverick is right: raise seat prices and move on. Usage metering adds complexity and procurement friction for no real gain — you'd just be re-deriving the same number through a more complicated formula.
- If AI value is skewed — some customers/workflows get a real 10x, others barely touch the AI features — a flat price hike either overcharges your light users (churn risk) or leaves serious money on the table with your heavy users (Philosopher's segmentation point). In that world, you need a usage or credit layer to price fairly and capture upside.
Most products with genuinely differentiated AI features are in the second bucket, at least in the early years of adoption — usage is almost never uniform right after a new capability ships. That's why the majority view (layer, don't just reprice) is the safer default, but Maverick's caution is the right check against over-engineering pricing before you have evidence of that skew.
Practical test: pull usage logs today. If AI feature adoption/output is tightly clustered, raise seat price. If it's bimodal or long-tailed, build the usage layer.
2. What Everyone Actually Agrees On (the safe core)
Strip away the framing differences and all four advisors converge on the same operational plan:
- Keep per-seat as the chassis. It still correctly prices things that scale with humans regardless of AI: access, permissions, admin, audit, collaboration. Don't let a shiny new metric erase pricing for things seats still legitimately buy.
- Add a second dimension for AI-driven value — credits, tokens, resolved tasks, automation runs, outcomes — whatever unit maps most cleanly to what the AI actually produces.
- Pilot on new logos/new tiers first. Never reprice your installed base in one motion. Existing customers are far more sensitive to model changes than prospects evaluating you fresh (Philosopher's point, and it's the single highest-leverage risk-reduction move in all four responses).
- Build metering infrastructure now, regardless of timing. You will need usage data no matter which path you take — to raise prices credibly, to design a usage layer, or eventually to justify full outcome pricing. This is a "do it now" item with zero downside.
3. The Question Nobody Fully Resolved: Will Seats Actually Shrink?
Here's a genuine tension worth surfacing rather than papering over. Philosopher warns that customers will quietly consolidate work onto fewer AI-augmented seats, eroding your seat revenue even if you do nothing. Maverick counters that seat counts rarely collapse cleanly — productive AI use tends to spawn new seats (oversight, prompt engineering, compliance review, parallel experimentation). Explorer's Jevons Paradox point cuts in yet another direction: cheaper output per unit of effort often increases total consumption rather than shrinking headcount, because demand for the underlying work turns out to be elastic, not fixed.
All three can be true in different markets. The diagnostic question: is the total amount of work your customers want done fixed, or elastic?
- Fixed workload (e.g., "we need exactly 500 contracts reviewed a year") → seats will consolidate, Philosopher's erosion scenario plays out, and you need the usage layer sooner.
- Elastic workload (e.g., "we could always use more content/code/analysis if it were cheaper") → Jevons kicks in, customers do more work rather than buying fewer seats, and Maverick's "seats rarely collapse as cleanly as advertised" holds.
You won't know for certain until you watch it happen — which is exactly why this belongs on your dashboard, not in a debate.
4. Your Actual Watch List
Don't wait for certainty, and don't act on vibes either. Track these signals over the next 1–2 quarters:
- Seat count vs. usage/output trend on existing accounts. Shrinking seats + flat or rising output = revenue erosion is already happening, not a future risk. That's Philosopher's "correction, not a bet" threshold — move faster.
- Usage distribution across customers. Tight clustering vs. long tail tells you raise-price vs. add-layer (Section 1).
- Competitive movement. If direct competitors ship usage/outcome pricing and win deals with it, procurement's "no box to check" objection (Explorer's point) starts dissolving fast — the market trains buyers, not you alone.
- Willingness-to-pay signals from new deals where you've already piloted the layered model — conversion, expansion, and churn data, per Architect's framework.
5. Decision Framework
Move toward usage/outcome pricing (beyond a light layer) when most of these are true:
- AI materially and measurably increases customer ROI
- Usage/value is unevenly distributed across your base (not uniform)
- You have clean, trustworthy usage telemetry
- Existing accounts show seats shrinking while output holds or grows
- Competitors are successfully selling non-seat models
Stay closer to seats-plus-price-increase when:
- Uplift is fairly uniform across customers
- You lack reliable usage data
- Procurement/forecasting predictability is a major deal-closing factor
- Your total addressable workload looks elastic, not fixed (Jevons regime)
Bottom Line
You're not jumping the gun by thinking about this now — you'd be jumping the gun by replacing per-seat pricing wholesale before you have usage data to justify the new model. The move that's actually available to you today, with no real downside, is:
- Raise seat prices where uplift is broad-based and hard to game.
- Layer a usage/credit metric on new deals where value is skewed toward power users.
- Instrument everything now so you're measuring the real signal (seat/output divergence) instead of reacting to the "10x" narrative.
- Let the data — not the AI hype cycle — tell you when to go further.
The seat isn't dead yet. But it's no longer the whole story, and the companies that win this transition will be the ones who let evidence, not urgency, decide how fast to move past it.