Agentic AI founders.
You're building an agent and you have to decide what to charge for before your first real contract locks it in.
Masterclass in Agentic AI Pricing & Monetization
Free · 3 hours · Live · Certificate on completion
Cursor charges $20 a month for each developer. Sierra charges only when its agent resolves a ticket. Different pricing metrics work for different agentic companies. Agentic companies also have unique differentiators that matter to how they monetize. In three hours and the only course of its type, we help you work out how your agent differentiates, what pricing model fits it best and frameworks to help guide your decisions.
Hosted by
Ajit Ghuman & Akhil Gupta
Taught by the authors of Monetizing Agentic AI and Price to Scale
Experience from
What is Agentic AI pricing?
Agentic AI pricing is the practice of charging for AI agents that do work on their own. SaaS pricing charges for access to a tool. Agentic pricing has to hold four things together at once: the value the customer receives, the inference cost of producing it, the output the agent creates, and the systems that meter and bill autonomous work.
What we cover in three hours
Part 1 · Goals and segments
Who you actually sell to, split by needs, value, behaviour and willingness to pay. In our experience this is the single biggest reason a pricing model underperforms.
Part 2 · Packaging
Building offers that fit each segment instead of one oversized plan. This is where most of the money is made and the hardest thing to fix later.
Part 3 · Harness audit
The five layers above the model. You'll mark what you own, what you rent, and what a competitor could rebuild in a weekend. That sets your ceiling.
Part 4 · Pricing metric
Seat, usage, output, outcome or hybrid. You'll score your product on the three dimensions of the spectrum and read the metric it points to.
Part 5 · Price point
Setting the actual number: which goal wins, what data to trust, and how to build a full price range through packaging and discount rules.
Part 6 · Making it run
The seven-layer monetization stack, which layer breaks first when usage pricing meets infrastructure built for seats, and how to design so the next change is cheap.
Who should be in the room
Who this is for
You don't need a pricing background. You do need a product or a practice you're actually responsible for, because the exercises use it.
SaaS leaders.
Your product works, and AI is quietly removing the seats you price against.
Services and agency leaders.
You want to move from billable hours toward outputs and outcomes, and you're working out what that does to your economics.
Product, pricing and finance teams.
You own packaging, metrics, rates, metering or billing, and you're the one who has to make the model actually run.
Course access and materials
What's included
- One live 3-hour session
- Five price list teardowns
- A harness audit worksheet for your own product
- An exercise scoring your product on the AMS
- Live Q&A with both instructors
- Monetizely pricing templates
- Slides and recording, yours to keep
- A certificate you can add to LinkedIn
From strategy to execution
The 5-Step Pricing Framework
Each decision creates the inputs for the next. The result is a pricing model that is commercially sound and operationally real.
01
Direction
Set goals & KPIs
Define the growth, adoption, margin, retention, and market-position goals pricing must support.
02
Market
Segment customers
Separate buyers by needs, value, behavior, and willingness to pay—not only by company size.
03
Offer
Design packaging
Shape clear offers around customer jobs, agent capabilities, controls, and service levels.
04
Economics
Choose metric & price
Select seat, usage, output, outcome, or hybrid pricing and validate the right rate structure.
05
Execution
Operationalize
Connect metering, entitlements, billing, sales enablement, governance, and iteration.
01–02 Understand the business and buyer
03–04 Design the offer and economics
05 Make the model work at scale
Choose the right value metric
The Agentic Monetization Spectrum
Place the agent on three dimensions, then move toward the pricing model its autonomy, scope, and economics can support.
Human anchoredAgent led
Predictable
Per seat / access
Balanced
Hybrid / usage
Value aligned
Output / outcome
01
Zero Human Ability
How much human involvement does the agent still need?
02
Operational Domain
Does it handle a task, a workflow, or a broader domain?
03
Output / Cost Curve
How quickly does output value outpace compute cost?
Harness design
Harness design and your price ceiling
The model is only one component. The harness around it is the product: orchestration, coordination, memory, permissions and the controls that turn intelligence into reliable work.
You'll mark which layers you own, which you rent, and what a competitor could rebuild in a weekend. That tells you where your differentiation lives and sets the ceiling for the pricing metric you can defend.
The stronger the harness, the further pricing can move from access toward output and outcome.
The book
Where the material comes from
We published Monetizing Agentic AI in 2026. The class works through three of its chapters in depth: harness design, the pricing framework and spectrum, and monetization engineering. The rest of the book covers agentic economics, the frontier companies and the services conversion, and we don't cover those live.
Every registrant gets the pricing chapter free.
Monetization engineering
Choosing the metric is only half the job
A pricing model doesn't really exist until you can meter it, rate it, enforce it and bill it. That work has a name, we call it monetization engineering, and it's where a lot of good pricing decisions quietly die.
Here's the shape of the problem. Picture a single button in an agentic marketing product labelled "Generate Campaign." Behind that one click, the agent drafts a plan, writes five emails, generates three images and produces landing page copy. It consumes roughly 15,000 input tokens, 4,000 output tokens, three image calls and twenty database lookups.
One click. One line in the product. And underneath it, a genuinely hard revenue question with no obviously correct answer.
Charge for tokens with a markup and your customer can't predict their bill, which makes procurement difficult. Charge a flat fee per campaign and you absorb the variance every time someone regenerates. Charge a platform fee and you smooth it for both sides while weakening the link between what you charge and what it costs you. Each option moves risk somewhere different.
There's a second problem stacked on top. Pricing doesn't hold still anymore. A new model halves your cost, or doubles it. Companies routinely move from per-seat to per-token to per-workflow to per-resolution inside six to eighteen months. Every one of those shifts ripples through metering, entitlements, rating and billing.
That's why this is engineering rather than configuration. The team that hardcoded a plan name into fifty microservices last year now spends a quarter untangling it before the new model can ship.
We walk all seven layers and show you which one tends to break first.
Seven-layer flow
QuoteCash
01
CPQ
DealConfigures the offer, price and contract terms.
02
Entitlements
RuntimeControls what each customer is allowed to use.
03
Metering
RuntimeCounts the tokens, calls, workflows or outcomes consumed.
04
Rating
RuntimeApplies pricing rules, tiers, credits and discounts.
05
Billing
FinanceTurns rated activity into invoices and payments.
06
Revenue recognition
FinanceDetermines when the resulting revenue counts.
07
ERP
RecordPosts the final financial result to the system of record.
Your instructors
Ajit Ghuman
Co-Founder and CEO, Monetizely
Ajit wrote Price to Scale, now in its second edition, and co-wrote Monetizing Agentic AI. He ran pricing at Twilio, Narvar and Medallia before starting Monetizely, and he now works with software companies from seed stage through post-IPO. In this class he takes the framework and the metric.
Akhil Gupta
Co-Founder and COO, Monetizely
Akhil co-wrote Monetizing Agentic AI and is the author of its chapters on harness architecture and monetization engineering. He's spent 16 years building high-throughput enterprise systems and running the teams that build them, which is why he teaches the two parts most pricing courses leave out: what actually sits above the model, and the metering, entitlements and billing architecture your usage or outcome model needs before it can run. He's a UC Berkeley-certified CTO and a graduate of Delhi College of Engineering.
Per-seat pricing broke in two places
For about twenty years, pricing software was one line of arithmetic. You counted the seats, multiplied by a price, and billed monthly. It was simple enough to quote on a napkin and stable enough that finance could model it in a spreadsheet. Every billing system, CPQ tool, comp plan and investor model got built on that one assumption.
If you're pricing an agent today, you've probably felt that model straining. It's worth being precise about why, because it actually broke in two separate places.
The cost side inverted.
In SaaS, your tenth user and your ten-thousandth cost roughly the same to serve. That's where 75 to 85 percent gross margins came from. Revenue scaled and costs didn't.
Agents don't work that way. Every task burns tokens, calls tools, and runs up infrastructure cost that tracks usage instead of seats. The expensive part is often invisible too, since reasoning models generate thousands of internal tokens the customer never sees but still pays for.
One customer on a flat monthly fee can consume fifty times what another consumes. Your heaviest users, the engaged ones you most want to keep, can cost ten to a hundred times what your light users cost. Margins that look healthy across the whole book can be sharply negative on specific named accounts, and you won't see it until you go looking.
The value side inverted.
If your agent genuinely replaces work, your customer needs fewer people doing that work. Fewer people means fewer seats. So the better your product performs, the less you collect.
Microsoft ran into this with Copilot at $30 per user per month, and responded by bundling more value into higher-priced enterprise tiers, which raises the price of the seats that remain. That buys time. It doesn't resolve the mismatch between how the product creates value and how it captures it.
"We realized we were running a utility company but billing like a magazine subscription."
The public markets have already reacted. Roughly $1 trillion came out of software stocks between mid-January and mid-February 2026. This class is about the work on the other side of that: rebuilding the model from segments up.
Five companies. Five different answers.
Frameworks are easier to trust once you've seen them tested against real numbers. These five companies all sell AI agents, they're worth more than $60 billion between them, and no two of them charge the same way. We work through each one in the class.
Cursor · $20/mo Pro, $40 per user/mo Business
Cursor charges by the seat, which looks conventional until you notice what the tiers actually separate. Code generation and codebase context are roughly the same on every plan. What changes between Free, Pro, Business and Enterprise is admin, governance and security.
That's a deliberate choice. A solo developer doesn't need single sign-on, and an enterprise security team doesn't care whether the free tier caps completions. It's a useful example of what per-seat looks like when the human is still doing the work and the AI is assisting.
Devin · $20 entry, ACUs at $2.00 to $2.25
Devin runs a hybrid: a platform fee plus usage, metered in Agent Compute Units. The entry price came down 96%, from $500 to $20, which tells you something about how fast this market moves.
The interesting detail is what the ACU is anchored to. It's priced against compute cost rather than the value of the output, and Cognition has been open that reliability is the reason, since complex tasks currently succeed 15 to 30 percent of the time. It's a good case study in pricing a product that's still proving itself, and in what has to change before the anchor can move.
Harvey · About $1,200 per lawyer per month
Harvey prices against lawyer time rather than against other software. With a 20-seat minimum, the annual floor lands near $288,000, which sounds high until you set it against what an Am Law 100 firm bills per hour. At that rate the product needs to save a few hours a month to pay for itself.
The packaging is enterprise-only and built per client. That suits the largest firms well. It also means mid-size firms and in-house legal teams, who want two or three practice areas rather than the full platform, don't currently have an obvious way in. We use Harvey to talk about what happens when the rate is right and the reach is narrow.
11x · About $5,000 a month
11x anchors Alice to the cost of a human SDR, roughly a $60,000-a-year role. Anchoring to a labour budget instead of a software budget is the move that makes agentic pricing interesting, and 11x is a clean example of it.
The market is now testing where that anchor sits. Alice handles part of the SDR job rather than all of it, and competitors have priced against narrower scopes, with Agent Frank at $499 a month and AiSDR at $900. It's a useful case for talking about how you defend an anchor once the field fills in.
Sierra · Per resolution, roughly $150K to $350K in year one
Sierra charges when its agent resolves a customer issue, which is about as close to outcome pricing as this market currently gets. Outcome pricing needs three conditions to hold: attribution has to be clear, measurement has to be immediate, and the value per outcome has to be high enough to matter.
Customer service happens to meet all three. Most categories don't, which is why we spend time on when this model travels and when it doesn't.
Final decision
The metric is the hardest decision to reverse
Three hours, no cost, with two people who do this work every day. You'll leave with a metric chosen for your own product and the reasoning to back it up.
Reserve your seatCursor$20 per developer / month
SierraPer resolution
Harvey$1,200 per lawyer / month