Pricing the layer that buys in a supplier's metered units and sells in units of its own. The pass-through-or-recast decision, the era lineage behind it (per-call APIs, OEM licensing, the agent harness), and who owns the meter when the unit on your invoice belongs to someone else.
[ The two doors ]
← scroll to view full figure →
The harness is this decade's name for a layer software has priced twice before.
A vendor that ships an agent harness wraps a large language model in tool dispatch, memory, sandboxed execution, and guardrails, and sells finished work to its customers. It buys its critical input in a supplier's metered units. Every software layer built that way eventually arrives at one binary, and this hub is the record of that binary: what it is, where it has appeared before, and what each answer costs.
Pass-through keeps the supplier's value metric and bills in the unit you buy. It is cost-plus by another name: legible on day one, easy to defend, and it denominates your product in someone else's meter. Recasting chooses a unit of your own, defined on what the layer delivers to the customer. It is the path to value-based pricing, and it is the harder door, because your input costs become yours to manage rather than your customer's line item to audit.
Keep the supplier's metric and bill in the unit you buy: calls, tokens, compute hours, marked up to carry margin. Fits while the layer is thin. Strains as capability accretes, because the invoice describes what you consume, not what the customer receives.
Define your own value metric on what the layer delivers. The path to value-based pricing, and the harder door: input volatility becomes your problem to manage, and the unit has to be one your buyers can understand and estimate.
A company that never chooses has usually walked through door one by default, because the supplier's meter was already running. The test: does the unit on your invoice describe what the customer receives, or what you consume?
The vocabulary churns; the decision does not. Each generation renames the layer, and the rename is what stops companies from recognizing a decision an earlier layer already faced.
Per-call API pricing held while a call meant interchangeable work, and failed when business logic moved into the layer. OEM licensing has asked the denominational question for decades: whose units is the grant denominated in. The harness is the current sighting, a runtime layer buying in a model vendor's tokens and selling something else. Three vocabularies, one decision.
The dated record behind these reads lives in the SPP Pricing Observatory, where vendor pricing moves are verified, dated, and maintained as arcs, and where harness vendors already have arcs on the books. Articles in this hub render that ledger rather than restating it, so their dated evidence stays current as the ledger is re-verified.
Start with Pass Through or Recast below: the binary, the lineage, and the test to run on whatever the layer is called next. The supporting articles each take one consequence: what an agent harness is and what it costs to run, the era when a call stopped being interchangeable work, who owns the meter when billing consolidates, what happens to application pricing when the model layer stops absorbing costs, and who pays for the run that produces nothing useful.
Sibling hubs carry the neighboring questions. The AI economics core, credits, agentic pricing strategy, and monetization, lives in AI Pricing. The unit-and-grant decision itself, including the OEM canon, lives in Licensing. Use this hub when the question is what happens to a pricing architecture built on top of somebody else's metered unit.
If you are standing in front of the two doors now, describe the layer you are pricing via Talk to an Expert and a pricing expert who has seen this decision under its earlier names will reply.
SaaS, platform, copilot, agent, harness: the names change and the decision does not. Pass the underlying cost through, or recast the metric on what the layer now does.
An acquired product ships with a value metric someone else chose. The acquirer either re-runs that decision or inherits it, one deal at a time.
Read →
An agent harness is the runtime layer that turns a language model into an agent. What the term means, where it came from, and what you…
Read →
Per-call API pricing survived only where a call stayed one uniform unit of work. Everywhere the layer thickened: pass through, or recast the metric.
Read →
When a customer supplies the model, a unit with inference blended into it keeps counting and stops covering. What the unit grants, and who supplies the…
Read →
An agent burns real compute and returns something unusable. Somebody pays for that run, and the meter decided who before anyone argued.
Read →
Application vendors priced AI features against inference costs the model layer was absorbing. That absorption is ending, and the vendors exposed are the ones whose licensing…
Read →
Three independent usage-billing vendors were absorbed by payments platforms in six months. The meter behind your AI pricing now has an owner paid on the flow.…
Read →If your product buys in a supplier's units and sells in units of its own, the pass-through-or-recast decision is on your desk whether or not anyone has answered it. Describe the layer you are pricing and a pricing expert will reply.
Talk to an Expert →