Agent Era Pricing Skill

Agent Era Pricing Skill

Seat pricing quietly assumed the user was a human who logs in. Agents break the assumption from both sides: your customers need fewer seats (one operator, ten agents), and your product gets more usage than ever. This skill redesigns the model around a value metric that survives non-human users — without torching existing revenue on the way.

What This Skill Produces

• A value-metric decision: what you charge for when seats stop proxying value
• Agent-tier design: how agent/API usage is packaged, fenced, and priced
• Cannibalisation math: what happens to current revenue under the new model, computed on real cohorts
• A phased migration plan for existing customers, with the grandfathering decision made explicitly

Required Inputs

Ask for (if not already provided):

• Current model: plans, price points, seat definitions, current API/automation pricing if any
• The evidence of pressure: seat contraction, API traffic growth, customer asks, competitor moves
• Unit economics: cost to serve a seat vs an API call/agent action (rough is fine, labelled)
• 3-5 representative customer profiles with seat counts and usage (the cannibalisation test set)

Method

1. Find the value metric that survives agents. Test candidates against three questions: does it scale with the value the customer receives (not your costs)? · is it counted identically whether a human or agent drives it? · can the customer predict their bill? Strong candidates are usually outcomes or work-objects (invoices processed, tickets resolved, campaigns run, records enriched) — not raw API calls (unpredictable, punishes retries) and not seats (dying assumption).
2. Price the human and the agent differently, deliberately. The durable pattern is a hybrid: a platform/human layer (flat or few-seats — access, admin, support) plus a work layer priced on the value metric, agnostic to who did the work. Decide where agents authenticate: agent traffic on a user's token counted as that user's work, not as a "seat".
3. Design the fences. What separates tiers now that seats don't: volume bands on the value metric, rate/concurrency limits, SSO/audit/compliance (still human-org fences), model/automation quality tiers. Every fence must be measurable and hard to game — name the gaming vector for each and why it's acceptable.
4. Run the cannibalisation math on real cohorts. For each customer profile: current annual price vs new-model price at current usage, at 2× automation, at 5×. Sum to a revenue bridge. If the new model loses money on your best cohort, the metric or the bands are wrong — fix the model, don't hide the row.
5. Phase the migration. New customers first (cleanest signal) → opt-in for existing (with a calculator showing their number) → forced migration only with long notice and a cap ("no more than X% increase in year one"). Grandfathering is a decision with a cost, not a default: state what perpetual legacy plans cost in five years.
6. Set the tripwires. Which metrics reprice this model: value-metric inflation/deflation, gaming detected, agent share of traffic crossing thresholds. Pricing in the agent era is a program, not a project.

Output Format

Agent-Era Pricing Plan: [product]

Diagnosis: [the seat-erosion evidence, quantified] Value metric: [chosen metric] — because [the three-question test, answered]. Rejected: [runner-up + why].

The model