Every agent run, tool call and token has a direct cost that scales with customer activity. If your value metric does not track that cost, adoption erodes your margin and your best customers quietly become your least profitable.
For AI-native teams · Any stage · agent or LLM-first
Every agent run has a compute cost. Our pricing does not move with it.
Our most active customers are our least profitable accounts.
Procurement keeps anchoring us to the cheapest single-agent tool they can find.
These are structural tensions built into the category itself. Each one has a design answer, and the companies that build it first set the anchors everyone else negotiates against.
Flat rate squeezes margin as usage grows. Pure consumption kills forecasting. Hybrid is the answer, and the hardest model to design well.
Charging for the result is where the pricing power is. Until the outcome is cleanly measurable, the interim proxy you pick creates incentives you will live with for two years.
There is no established anchor for what a fleet of coordinated agents is worth. Left alone, buyers anchor to the cheapest comparable single agent.
SOC 2, HIPAA and EU AI Act readiness carry real cost and real value. The window to price them closes as they become table stakes.
Pricing against the salary the agent replaces hands procurement a budget line to compare against, then inverts the moment the agent underperforms. Augmentation framing is safer than replacement.
The timing argument is real, and it is specific to this market:
Pricing norms in agentic AI are being set right now, by whoever moves first and gets it roughly right
Companies that can iterate on pricing through 2026 will set the anchors in their category
Everyone else spends the next five years negotiating against anchors somebody else chose
Margin erosion from unpriced compute compounds with adoption, so the best-case growth scenario is also the worst-case margin scenario
Three design moves separate the agentic companies with pricing power from the ones discounting to survive procurement.
A value metric designed so revenue moves with the compute the customer consumes and the outcome they receive, instead of tracking a headcount your own product is shrinking.
Committed tiers with rollover rights and overage logic, so buyers get budget certainty and you get a forecastable floor with expansion built in.
Package security, audit and regulatory readiness as a priced capability while the window is open, before they become a free checkbox in the enterprise tier.
Both engagements were AI-first products pricing what the model actually does.
Credits were priced at cost-plus with no link to the value a review delivered, and buyers could not understand them. We mapped credit cost to review complexity, built a spend-modelling tool and added pre-committed enterprise tiers with rollover rights. Conversion and ARPA both rose.
A diagnostic AI was priced per clinician seat, a model with no relationship to the outcome delivered. We built a value model in NHS commissioning language and a base-plus-outcome structure tied to auditable data. First outcome-based contract signed within two procurement cycles.
12–16 weeks · Any stage · agent or LLM-first · NDA on request
Bring your cost per agent run and your current price card. Thirty minutes is usually enough to see whether your metric tracks your cost, and what it would take to make it.
Not you? Executive view · Growth view · Product view · Finance view
