Your AI Agent Stack Is Volatile. Your Token Commitments Shouldn't Be.

Jason Lemkin published something important on SaaStr this week. His argument, grounded in board meetings with AI-native companies, is that prompts are portable — switching an AI agent vendor takes days, not months. One-year contracts are the new default. Structural churn is embedded in every deal, even the large ones. He's writing for founders and vendors. But the most consequential reader of that piece is a CFO. Because if your AI agent stack is volatile — if every vendor relationship is effectively a rolling one-year decision — then you have a serious token treasury problem that nobody is talking about yet. --- ## The Commitment Timing Problem Most enterprise AI pricing offers a version of this deal: commit to volume upfront, pay a lower per-token rate. Reserve capacity, capture the discount. It's structurally identical to how you'd lock in a commodity forward rate or negotiate reserved cloud compute. The model works when your consumption is predictable and your vendor relationships are stable. Lemkin's observation breaks that assumption. If your organisation is running twenty-plus AI agents — as many enterprises now are — and each one sits on a one-year contract with an implicit churn option, then your committed token capacity is exposed to a volatility event every time a contract comes up for renewal. Churn an agent vendor mid-commitment period and you're either eating unused reserved capacity or scrambling to redeploy it. Sign a new vendor without modelling the capacity implications and you're now overcommitted across two providers simultaneously. Neither outcome shows up on the engineering dashboard. Both show up on the balance sheet. --- ## The CFO Opportunity Inside the Churn Story Lemkin frames this as a threat to vendors and an opportunity for buyers. He's right on both counts. But there's a third dimension he doesn't address: it's also the moment when finance finally needs to govern the AI agent stack as a portfolio. When you have one AI agent on a single annual contract, that's a procurement decision. When you have twenty agents across multiple providers, each on one-year terms, each with its own token pricing structure, each generating its own consumption curve — that's a treasury problem. The questions that matter aren't "is this agent performing?" They're: - What is our total committed token capacity across all providers right now? - Which commitments are at risk if we exercise the one-year exit option? - Where are we overcommitted relative to actual consumption? - What is the cost of switching at renewal versus the cost of staying? These are questions a finance team should be asking. They're questions the current toolset — built for engineers counting tokens, not CFOs managing exposure — cannot answer. --- ## Portability Cuts Both Ways Lemkin makes the case that prompt portability gives buyers leverage. That's true. But leverage without visibility is just the option to make a poorly-informed decision faster. The enterprises that use this moment well aren't the ones that switch agents most aggressively. They're the ones that build the financial governance infrastructure to know — before every renewal cycle — exactly what their committed exposure looks like, where the switching economics actually favour a move, and how to redeploy capacity without creating new financial risk. Token treasury isn't just about reducing spend. It's about managing the portfolio of commitments that accumulates when your agent stack is volatile by design. That infrastructure doesn't exist yet. Building it is the work.