Frameworks outnumber the agents

There are currently more frameworks for pricing AI agents than there are AI agents doing anything measurable, which is saying something given every startup founded in the last 18 months has described itself as "agentic."

𝗨𝘀𝗮𝗴𝗲-𝗯𝗮𝘀𝗲𝗱. Outcome-based. Seat-based. Hybrid.

There are 2x2 matrices, the matrices have arrows, the arrows have names, and one literally called "The Pathway to Growth" which sounds less like a pricing model and more like a newsletter written by someone who just finished a 10-day silent retreat and is now very excited to tell you about compounding.

I've been reading things this week that suggest the entire debate is happening in the wrong room.

𝗦𝘁𝗮𝗿𝘁 𝘄𝗶𝘁𝗵 𝗡𝗼𝘁𝗶𝗼𝗻. Akshay Kothari announced usage-based pricing for Custom Agents last month. Pay for work done. Clean logic. Except a user named William Hutter did the math publicly: "$500 just for email triage, a sports coach, and a morning brief agent. Nobody can afford that. I'll be out in 2 months and back to Openclaw."

Kothari's reply: "Openclaw is free?"

𝗡𝗼𝘄 𝘁𝗵𝗲 𝗩𝗖 𝗹𝗮𝘆𝗲𝗿. The operator post going around argues VCs are pushing hard for outcome-based pricing. Makes sense on paper, higher revenue ceiling, better alignment.

Except 𝗼𝘂𝘁𝗰𝗼𝗺𝗲-𝗯𝗮𝘀𝗲𝗱 𝗽𝗿𝗶𝗰𝗶𝗻𝗴 𝗱𝗲𝘀𝘁𝗿𝗼𝘆𝘀 𝗔𝗥𝗥. NRR becomes meaningless. Annual contracts lose value, buyers can stop volume and stop paying. VCs are pushing founders toward the pricing model that destroys the metric VCs use to value the company. Somewhere a spreadsheet is crying.

But neither camp is solving 𝘁𝗵𝗲 𝘂𝗽𝘀𝘁𝗿𝗲𝗮𝗺 𝗽𝗿𝗼𝗯𝗹𝗲𝗺. You cannot price an outcome you haven't defined, and you cannot define one without deciding who owns the risk when the agent gets it wrong.

That is the actual fight. The pricing debate is downstream of a liability debate nobody is having.

𝗘𝗻𝘁𝗲𝗿 𝗘𝗹𝗲𝘃𝗲𝗻𝗟𝗮𝗯𝘀, February 2026. First company to insure its AI voice agents. AIUC-1 certification: 5,835 adversarial tests across 14 risk categories. AIUC, its certifier, was co-founded by ex-Anthropic employees, the same lab whose model powers most of these agents.

The food chain is financing its own auditor. I don't know what to call that except 𝗲𝘅𝘁𝗿𝗲𝗺𝗲𝗹𝘆 𝟮𝟬𝟮𝟲.

Insurance requires actuarial data. Actuarial data requires standardized risk testing. AIUC-1 is that. Which means outcome-based pricing can only function after the risk quantification layer ElevenLabs just built.

That infrastructure arrived last month from a company nobody was watching.

𝗢𝘃𝗲𝗿 𝟵𝟱% 𝗼𝗳 𝗲𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲 𝗔𝗜 𝗽𝗶𝗹𝗼𝘁𝘀 𝗳𝗮𝗶𝗹. Stated reason: legal and security concerns. The bottleneck was always liability.

The arrow pointing toward "Outcome-Based" is not wrong. The actual route has an 18-month detour through contract law and adversarial testing that nobody put on the slide.

I have no idea what my agents cost on a given Tuesday.

That is a liability problem wearing a pricing model's clothes.