How to do AI Agent billing in 2025

The Agentic Playbook

and

Jul 03, 2025

If you feel like you can’t find the right answers to agent pricing, it’s because they’re being written in real time. We’re in the midst of the biggest shift to disrupt the market in 25 years, and every founder we speak with has the same questions. Luckily, we’re obsessed enough with billing to compile all the answers.

Agentic AI is software, and for the past 25 years, software has been SaaS. So, it has been (understandably) difficult for the world to accept that Agentic AI is not SaaS.

Agentic AI is to SaaS what SaaS was to CD-ROM. The rapid change you’re seeing in approach to pricing is indicative of the market waking up to this distinction.

AI Agents priced as SaaS makes no more sense than SaaS being priced as CD-ROM.

Outcome based pricing billed on AI native platforms is the final destination everyone will eventually get to. We’ve said it since the beginning, and we’re watching the debates finally start to tip in our direction. Since we spent the last 6 months interviewing over 160 agentic founders, we were a little early to the party.

The good news is we’ve had plenty of time to sort out the details for you.

What pricing models actually work for AI Agents?

Though SaaS was born in 2000, Windows was still sold on a shelf in a box until 2012. Not because Benioff was wrong, but because he was early. Physical products in retail stores was the monetization model the software industry knew and had access to.

Today’s Office Max equivalent, “the devil you know” so to speak, is a category of billing platforms that predates the first agentic company. Billing agentic AI on these platforms is like selling Salesforce at Office Max today.

Stripe, Chargebee, and Zuora weren't built for the AI economy. They’re built on traditional SaaS assumptions that break with agents.

Seat-based pricing: Agents don't need "seats" because one user can deploy dozens of agents

Fixed SKU structures: Can't handle dynamic, context-dependent agent workflows

Usage = API calls: Misses the actual work your agents perform

Monthly billing cycles: Agents deliver value continuously, not in subscription intervals

75% of AI agent companies we analyzed struggle with pricing because they're forcing agent economics into SaaS billing boxes. They’re the devil you know, so you try to make it work. (Spoiler: it doesn’t work!)

Instead of pricing like SaaS, price like labor. That’s what agents are doing.

Per workflow completed (document processed, call handled, analysis delivered)

Per outcome achieved (meeting booked, issue resolved, lead qualified)

Per unit of business value (cost saved, revenue generated, time recovered)

As Rob Litterst from PricingSaaS puts it: "AI agents are making it so that you actually can do jobs... SaaS could never actually do a job for you." The shift from selling tools to selling outcomes is already happening, with companies like Intercom and Salesforce leading the charge.

We broke down each of these models in detail in The Complete Guide to AI Agent Monetization. We’ve seen every one of these models implemented and have tracked what works and what doesn’t across all 160+ companies we’ve consulted.

Every founder starts with usage-based pricing. Half realize it’s killing their margin within the first 30 days.

How do you make AI Agents profitable?

The biggest killer of agentic companies is margin erosion, not lack of revenue.

ARR has always been the lifeblood of SaaS. Since cost of delivery held steady at scale, ARR and EBITDA had a strong positive direct correlation.

That correlation flies out the window with Agentic AI. Costs can spike unpredictably and margins can vanish overnight if your customers write too much to your agent.

Here's what breaks traditional margin tracking:

Model costs fluctuate: GPT-4o costs 33% more than GPT-4.1, but customers pay the same.
Usage varies wildly: One customer's "simple" task burns 10x the tokens you expected.
Hidden infrastructure costs: Voice processing, image analysis, vector databases - all invisible to traditional billing.

Stripe and friends help you collect money. They don't help you spot a bleeding margin when your inference costs explode.

We’ve seen founders celebrate pilot wins… only to panic when infra bills hit 5 figures with no matching revenue. Adam Schoenfeld candidly shared with us: "I do think it's going to be something we have to deal with because... there is a version of the future where the costs actually go up, where like the agents have a lot more different tools and those tools have their own costs and there's all this variability between customers."

This margin compression is happening everywhere, and most don't catch it until it's too late.

The metrics that actually matter for agent profitability:

Agentic Margin (AM): Revenue minus all agent operating costs per customer
Agentic Margin Ratio (AMR): Your true profit percentage after AI infrastructure
Task Monetization Ratio (TMR): What percentage of agent work actually generates revenue

Real example: One customer told us their "profitable" AI support agent was actually losing $0.40 per conversation after accounting for model costs, voice processing, and infrastructure. They were scaling their way to bankruptcy.

It’s gutting to realize your most successful agent is actually hemorrhaging money. That’s why we wrote a complete guide to margin management: The Agentic Margin: What It Costs vs. What You Earn from AI Assistants. We want to help you catch the bleed before you scale it.

How do you grow and scale an agentic company?

Does it feel like all your customers are blown away by what your agent can do, but then start to go cold after a few months? Most of the founders we spoke with saw a major drop-off 6 to 9 months after launch. It can be super scary, but there is a path forward.

Vibe Revenue is a new challenge that software never faced in Office Max or SaaS, but is giving agentic companies a damning false hope.

AI Agents are so hot right now everyone wants to try them… once. Then, never give you money ever again. The result is an initial influx of money followed by a steep drop off 6 to 9 months after launch. That’s Vibe Revenue.

As Sequoia's Pat Grady perfectly captures this phenomenon: "The difference between having a magic moment and solving an end-to-end workflow... That's the difference between the vibe revenue and the real revenue."

Initial contracts come up for renewal. Novelty wears off. Customers evaluate ROI with cold, hard metrics. Without genuine momentum, you're facing the renewal cliff without a paraglider. We’ve watched it happen like clockwork to companies of all sizes.

The warning signs you're in vibe territory:

High initial adoption but shallow engagement depth
Customers trying your product vs. depending on it
Usage metrics that look great but don't translate to business outcomes
Contracts that feel more like "experiments" than strategic investments

What separates vibe from value?

Deep workflow integration (not just cool features): Each integration point creates another reason to renew
Expanding value over time: Agents handling increasingly complex, important tasks
Measurable business impact: Clear ROI that executives can defend to the board

Real example: One customer told us their AI agent went from "nice-to-have demo" to "mission-critical infrastructure" when they shifted from showcasing features to demonstrating $50K/month in cost savings with detailed attribution.

We wrote extensive instructions on how to prove outcome and sell results in Vibe Revenue: A Mirage of AI Success.

The AI economy doesn’t fit on SaaS infrastructure

If you’re building agents that work 24/7 to deliver measurable business outcomes and replace entire job functions, you’re not building SaaS. You can’t bill for SaaS.

AI agent companies using outcome-based pricing see 4-8x higher contract values than those trapped in SaaS models. They convert pilots faster, retain customers longer, and scale without the margin compression that kills their competitors.