Last week Salesforce made a move that inspired a collective spit take across the valley. After months of hyping usage-based AI pricing (charging customers per conversation, per workflow, etc) they announced they’re returning to seat-based licensing for Agentforce. That’s right: seats! Apparently, even the robot needs a chair (or at least a line item).
Marc Benioff framed the shift as customers wanting predictability, which is polite CEO-speak for “CFOs are not signing off on invoices shaped like roller coasters.” If Salesforce is stepping back from pure consumption pricing, then something bigger is happening beneath the surface. Is the industry’s consumption-hype cycle winding down?
For years, everyone talked about usage pricing like it was gravity itself: an unstoppable force shaping the future of software. It was the fairest model, the most modern model, the model that perfectly aligned vendor and customer incentives. But then reality showed up with a strong cup of coffee and an LLM bill.
AI usage is not just variable, it’s wildly unpredictable. One support ticket might trigger 30 LLM calls or 3,000, depending on prompts, workflow design, or how philosophical your chatbot is feeling at any given moment. The unfortunately faith-based finance teams trying to model these costs quickly find themselves relying on guesses and prayers. Companies want AI innovation, not AI invoicing surprises, and that’s when the gloss comes off consumption pricing.
Salesforce’s return to seats is the industry’s collective reminder that predictability never goes out of style, especially when AI usage patterns resemble abstract art more than financial logic. AELA (their seat-based model with usage caps hiding under the hood) is essentially a way of saying, “Finance teams, we hear you. You want clarity. Well, here it is.”
Gartner’s data tells the story: Customers are adopting AI agents in a “very predictable, steady trajectory.” And steady adoption requires predictable costs. No CIO wants to defend a budget spike caused by a chatbot that decided to vibe out for twelve hours. The pricing pendulum hasn’t swung backward into 90’s nostalgia. It has swung toward the practical center, the place where adoption accelerates because customers finally understand what they’re signing up for.
The smartest companies aren’t treating pricing like an ideological battle, they’re blending models. Start with a base of predictable commitments (seats, tiers, platform fees) so customers have something solid to plan around. Then add usage-based guardrails underneath (credit pools, fair-use limits, burst allowances) to make sure vendors aren’t subsidizing runaway AI workloads. The point isn’t that one model is “better than” the other. It’s that your pricing model needs to match your customer’s buying motion and risk tolerance.
We’ve been tracking this at the Subscribed Institute since 2023, and the data is unambiguous: Companies using hybrid consumption models—mixing recurring commitments with usage-based charges—outperform pure consumption and pure subscription models on YoY ARR growth. The optimal sweet spot? About 25% of revenue from usage models, balanced with recurring revenue streams.
This hybrid approach recognizes that real customer relationships happen across multiple dimensions: predictability, flexibility, responsibility, and measurable results. Hybrid pricing wins because it meets customers where they are, not where a theory says they should be. Anyone still shouting that pure consumption pricing is the only “modern” model just isn’t thinking clearly (or has yet to be flattened by an inference whale).
Too many vendors adopted consumption pricing because AWS made it look glamorous, or because investors loved the sound of usage-based expansion, or because competitors were bragging about their per-token metrics. But pricing isn’t a fashion choice, it’s a strategic foundation. It has to reflect how your product delivers value and how your customers want to buy.
Consumption pricing is powerful in the right contexts, awkward in others, and downright risky in lots of early AI deployments. Salesforce realized this and did the most grown-up thing a company can do: it changed its mind publicly. That’s not a retreat, it’s a recalibration.
Hybrid models will continue to dominate not because they’re stylish, but because they reflect the complexity of delivering real value. And in the end, customers don’t care how wonderfully flexible your pricing model is. They care whether they understand it, trust it, and can scale it.
So stay true to your own pricing philosophy. Companies chasing consumption pricing may realize it isn’t for everyone. In a world already full of AI uncertainty, the most powerful thing a vendor can offer isn’t tokens or credits or seats. It’s clarity. Salesforce just reminded everyone of that. The companies that listen will win.
