Outcome-Based Pricing for AI Agents – Who Captures the Value

01 /The Concept

The founder of Outreach – which he scaled to $250 million in annual revenue and a $4.4 billion valuation – stepped down as CEO last fall to launch something he apparently considers an even bigger opportunity.

Paid is a platform the founder describes as a "business engine for AI agents" – the infrastructure through which AI agents generate revenue for their creators.

The platform has three components.

The first handles monetization. It enables developers to charge users in multiple ways: per seat, per resource consumed, or even per outcome delivered. Any combination of these can be layered into a single pricing plan.

Outcome-based billing currently works by analyzing the agent's responses to determine how much actionable output it delivered – for instance, counting how many qualified leads were returned in response to a search query. Future versions will likely integrate with CRM systems to assess final outcomes: were those leads actually converted to signed deals, and for how much?

The second component answers whether the AI agent is actually profitable for its developer. The problem is that developers pay for every API call to a model like OpenAI. If they charge on an outcome basis but the agent burns too many calls per result – or the revenue-share rate is too low – they can end up losing money on volume. This component lets developers see exactly where they're making and losing money across all tasks, so they can tune their algorithms or adjust pricing.

The third component helps users understand whether the AI agent is actually working for them. It calculates the value a user received versus what they spent, and automatically sends periodic ROI reports to users – with copies to the developer, so both parties understand whether the agent is earning its keep.

This may be the most important component of all. Without it, developers have no way to prove to users that outcome-based fees are justified.

Integrating all of this requires just a few lines of code added to an existing AI agent. From there, developers can start experimenting with pricing models and iterate.

Paid is currently in beta, and the startup has just raised its first €10 million.

02 /Why It Matters

The founder published what he calls an "AI agent developer manifesto" declaring that traditional SaaS pricing is dead. That's a bit ahead of the curve – but the direction is clear.

The two classic SaaS models – per seat and per resource – break down for AI agents. Charging per seat stops making sense when a single user can run dozens of agents simultaneously. And charging per resource feels like leaving money on the table: AI agents don't just compute – they can own entire job functions that companies currently staff with humans who earn salaries, bonuses, and commissions.

This means AI agent monetization needs to be rethought from the ground up: charge for outcomes, and prove to users that the outcomes are worth paying for.

The AI agent market was worth $3.5 billion in 2023, grew to $5.1 billion in 2024, and is forecast to reach $47 billion by 2030. The financial rails for that market are being built right now.

Paid addresses one dimension of the financial stack – how agents get paid. But the ecosystem needs more.

For example: if AI agents replace human workers who previously used SaaS tools, those agents will need to pay for those tools themselves. Skyfire ([related review](/review/a-teper-nauchi-ih-platit)) is building infrastructure to enable exactly that – opening SaaS products to AI agents as a paying customer segment. It raised $8.5 million in its first round, then pulled an additional $500K from a16z Crypto because the payment layer uses cryptocurrency.

And if humans can hire freelancers, AI agents will eventually need to as well – sourcing human help for tasks they can't handle on their own. Payman ([related review](/review/kogda-ne-my-im-platim-a-oni-nam)) raised $3 million in its seed round combining a freelancer marketplace for AI agent work orders with payment tooling. It has since pivoted to focus purely on the payment layer. Meanwhile, Y Combinator graduate Abundant ([related review](/review/tri-modeli-budushhego-dlja-ii-botov)) is building the marketplace layer – a network of humans that AI agents can automatically engage when they hit the limits of their own capabilities.

03 /Opportunities

The first takeaway is practical advice for AI agent developers: don't undersell yourself with subscription fees.

Build agents that take a cut of the results they generate for clients. That's the most attractive model right now, and it will quickly feel normal to users. If you've built an AI system that picks high-performing stocks, don't sell a subscription – take a percentage of the returns above what the user would have earned from an S&P 500 index fund.

The second direction is broader: building the financial infrastructure that lets AI agents earn and spend. "Earn" – using platforms like Paid. "Spend" – paying for SaaS and freelancers, using platforms like Payman and Skyfire.

Either way, the timing is right. AI agents are the most dramatic technological shift happening in real time, and the startups that will win are the ones that figure out how to ride the wave – not the ones that wait to see which way it breaks.