Welcome back to Good Better Best.
Each week, we break down real pricing, packaging, and product moves from SaaS and AI leaders and share the ideas worth stealing.
This week Ulrik and I broke down Linear's pricing on Pricing Page unPacked. We found it fascinating that Agents now create half the work in Linear, and Linear takes no markup on the AI tokens behind it. This week, I broke down their model, and what you can learn from it.
Before we get there, a couple things you should check out:
▶ The AI Monetization Roadmap: Ulrik’s follow up to The Pricing Roadmap is available for Kindle pre-order. If you are a Kindle reader, reserve your copy here.
▶ AMA with Jas Garcha, Head of Pricing at Vercel: In this session, Jas will be taking questions on how Vercel is monetizing agents, and what other companies can takeaway. Register here.
Let's get to it.
This Week in Pricing, Packaging, and Product
This week PricingSaaS tracked 348 companies with changes (457 events). Top 50 highlights:
Anthropic added Claude Sheets and Claude Whiteboards in beta [Link]
Slack capped Slackbot at 15 messages per user each week on Business+ [Link]
Front raised the starting price of its Autopilot AI agent [Link]
Attio replaced Enterprise's custom quote with a self-serve Ultra plan at $1,000 a month that includes 20,000 workspace credits [Link]
Devin let Pro, Max and Team users sign in with their ChatGPT plan [Link]
Datadog launched End User Device Monitoring as a new product [Link]
RingCentral moved AI Receptionist from included to a paid add-on [Link]
HubSpot doubled Marketing Hub Professional's included AI prompts and answers [Link]
Google Cloud switched Conversational Analytics from shared token pools per tier to per-user allowances [Link]
PagerDuty cut included events on Essential while raising AI actions [Link]
Exa cut its Monitors API price from $15 to $2 per 1,000 requests and added a self-serve Enterprise tier at $12,000 a year [Link]
Aircall cut AI Voice Agents pay-as-you-go pricing from $0.49 to $0.29 a minute and revised minute bundles [Link]
Baseten replaced its Basic, Pro and Enterprise plans with a pay-as-you-go track and a contact-sales Production track [Link]
Dovetail scrapped its Free plan for paid Dovetail One tiers starting with Bronze at $4,800 a year plus credits at 15 cents each [Link]
Synthesia replaced the 3-minute-per-video cap on its Syren AI video agent with video quotas by plan [Link]
Luma AI now bills Video-to-Video on generated plus reference seconds while roughly halving rates [Link]
Instantly raised its Hypergrowth plan from $97 to $107 a month and added a built-in CRM [Link]
Mintlify moved Automations to Enterprise only and gave Pro a credit-based Autopilot feature that uses 250 credits per update [Link]
Ahrefs removed the public $1,499-a-month price from its Enterprise plan, which now shows only a Talk to sales button [Link]
Check out more updates on PricingSaaS →
A year ago, AI agents created 3% of the work in Linear. Today it's 50%.
Even more impressive, 95% of workspaces have Agents activated. Most SaaS players would kill for this kind of AI adoption.
Today, I want to break down how Linear is doing it.
Linear in 60 Seconds
Linear started in 2019 as a faster, cleaner alternative to Jira, built by ex-Airbnb, Uber, and Coinbase engineers. It grew almost entirely through word of mouth among developers and passed $100 million in ARR this year at a $2.5 billion valuation.
Now it's betting that agents, not people, will file most of the work, and it wants to be where that work lives. Here's the kicker: Linear takes no markup on the AI models doing all that work. It charges for seats and sells AI at cost.
Below I walk through exactly how Linear is charging for AI, why they’re doing this, the risk hiding in the model, and what they can do to fix it.
How Linear’s AI Pricing Works
There are two parts to Linear's pricing.
Part one is seats. You pay per person, billed monthly or yearly.
This is where Linear makes its money. Plans are generally differentiated by functional features, with each new plan getting more team capacity, Admin granularity, and AI intelligence. Importantly, all plans get access to Linear Agent, including Free.
Part two is AI credits. Two features use them: Coding Sessions and Loops. Coding sessions are when Linear's agent writes code and opens a pull request (a proposed code change). Loops are agent jobs that run in the background.
Using AI credits is fully opt-in. If a customer never adds a prepaid balance, they just don’t get access to the Coding Sessions and Loops features.
Some details on how credits work:
Credits are a prepaid balance shown in dollars and shared across the workspace.
The minimum top-up is $10, or $50 if you turn on automatic reloads.
Unused money rolls over from one billing cycle to the next, and expires 12 months after you buy it. It can't be refunded or moved to another workspace.
Linear lists no discount for buying in bulk.
Just to reiterate, prepaid credits are totally separate from per-seat costs, and do not draw from that balance at all. Also worth reiterating, credits only support two of Linear’s AI features: Coding Sessions and Loops.
Here’s how Linear describes how tokens are used for each:
A few important details to highlight:
Coding sessions charge for model tokens (the units AI models bill by) at the AI provider's published rates
On top of that, you pay for the computer the agent runs on (price scales with size of the virtual computer and is charged in 20-minute increments)
Loops typically cost $0.07 to $0.20 a run.
While Linear does claim to offer model tokens at cost, they make no promises about computer usage or Loops, which suggests they take a modest margin on both.
In short, Linear is betting on adoption in the short-term in hopes that it will lead to monetization in the long run.
The Bet and the Risk
The bet is working. Some stats that back it up:
95% of paid workspaces use agents
Issues with a pull request attached are up 7x since January
Net revenue retention is at 177% (insane for a company with a seat-based model)
Linear pitches itself as the tool built for teams where agents do the work, unlike Jira, which grew up around people filing tickets. Pricing tokens at cost fits that story. If agents are going to work somewhere, Linear wants it to be Linear, and they don’t want price to be the blocker.
It also helps that most agent work in Linear is free. Only Coding Sessions and Loops draw on credits. Everything else agents do is included in the seat, so there's no meter to make teams think twice before trying them.
In our episode, Ulrik pointed out two more reasons “at cost” is smart here:
It ends discount conversations. With no markup, any discount means losing money. When a buyer asks for a deal on credits, there's nothing to negotiate.
It's more believable than free. Free AI looks like it can't last. An at-cost price is one developers can trust to hold for years.
The risk is the ceiling. Agents already create half the work in Linear, but revenue still comes from people at $10 or $16 a seat. If agents keep doing more, value shifts to the part Linear doesn't mark up. Seats could stall or even shrink as teams need fewer people to ship.
The ceiling is already low. Ulrik ran the math: 100 people on Business is $19,200 a year, and above that there's only Enterprise. By our math on Linear's own numbers, $100M across 40,000 paying companies is about $2,500 per company per year.
Ulrik raised a great question on our podcast: With this model, how could a customer like OpenAI ever pay Linear $10 million a year? Right now, it’s not possible.
How to Make Linear’s Model Better
So how does Linear raise that ceiling without breaking its promise to developers? Here are two moves we'd make:
Keep tokens at cost, and charge for what only Linear can do. Tokens are a commodity, and "no markup" is the promise developers trust. Linear's value sits on top: running loops at scale, controlling which agent can touch which code, and reporting on what agents shipped. That's where a $10 million customer would spend.
Open every door, and charge for control. Today Slack and email intake start on Business, and web forms are Enterprise only. Ulrik would flip that: let everyone bring work in from anywhere, then charge for sorting it, assigning it, and controlling who can do what. As he put it, limiting the ways in is like "restricting how many people can go into your store."
None of this needs to happen tomorrow. Selling AI at cost is how Linear got agents doing half the work, and 177% NRR says customers are happy to keep paying for seats. But the longer agents grow while seats stay flat, the harder the switch gets. Ulrik's warning applies to anyone using at-cost pricing to drive adoption: don't push off making money forever.
Patterns worth Stealing
Even if agents aren't doing half the work in your product yet, these moves are worth borrowing:
Put the commodity at cost, and say so plainly. "No markup" on tokens builds trust with buyers who know what AI costs. Your margin can live in what only you provide.
Don't meter every AI feature. Include everyday agent work in the core plan, and save credits for the heavy, costly jobs. It keeps adoption high and the bill predictable.
Use “at cost” to end discount talks. With no markup, there's nothing to negotiate.
Show credits in dollars, and let them roll over. Nobody has to learn a new unit or rush to spend it.
Give away the ways in, and charge for control. The more work that flows into your product, the more valuable the controls become.
Linear is a good reminder that seat pricing isn't dead. Its buyers still prefer a simple per-seat license, so Linear meets them there and keeps charging for seats. But they’ve also laid the groundwork for what comes next: a shared credit balance, usage-based pricing for agent work, and buyers who already trust how it prices AI.
As AI usage keeps growing, Linear won't have to rebuild its model from scratch. They just have to turn the dial.
Until next time,
Rob
Sources
