How Notion made its AI worth paying for

I have a love/hate relationship with Notion. I remember first downloading it when it was the wonder child of Silicon Valley and watching endless YouTube videos to figure out how to “properly” set it up. I probably spent more time building a system in Notion than actually using it.

I love it because of its versatility, but I hate it because it feels like you need a degree to use it properly. Even with this love/hate relationship, I write every newsletter article in Notion. Maybe I just like complaining.

This week, I’m covering how Notion has added AI functionality on top of their software that allows users to create their own custom agents, a distinctly Notion choose-your-own-adventure product.

TL;DR

The thirty-second version of what Notion did.

Notion got famous as the new-era all-in-one workspace. It’s an app that lets a team replace its docs, wiki, and project tracker with one flexible place to work. Today, more than 100 million people use it.

Layered on top is Notion AI, the assistant baked into the workspace: it drafts and summarizes, answers questions about your own docs, takes meeting notes, and searches across everything you've stored.

Then in 2025, Notion Agents launched, which took their AI from just a chatbot to autonomous teammates that run entire workflows on their own. Users built over a million Custom Agents in the first two months, and Notion runs around 2,800 of its own agents nonstop internally.

Adoption of the AI offerings increased from 20% to 50% within a year, and last year they crossed $500 million in revenue, with roughly half of it coming from their AI.

Notion isn't trying to win the model race. Its edge is that your work already lives inside it, so the AI starts every task with your business knowledge. They are leveraging your context + model intelligence to go deep into the execution layer, the actual work. And because users define that work, Notion can't predict what it costs, so it had to rethink the SaaS pricing model and add usage pricing onto the subscription.

Under The Hood

The three strategies that make it work.

1. Get all the context

Notion's paid AI plan grew from 10-20% of customers in 2024 to over 50% in 2025. The reason is what Notion sits on top of: every doc, every wiki, every project, every meeting note. The AI race isn't won at the model; it's won in the context (I’m going to get that tattooed across my back).

When you fire up a ChatGPT conversation, it might remember bits of your past chats, but you still need to feed it the relevant files and data to get anything good out of it. It's exhausting.

But if your workspace software has AI that can quickly search over your entire knowledge base, using it is a no-brainer. That’s what Notion AI enables. It has notetakers, research agents, and task automation that sit on top of your work. No more needing to explain to Claude for the hundredth time about the experiment you want to run.

Obviously, Notion has all this context because it was built to collect it. It provides the tools and surface area, but users are the ones who fill it. This is different than a product like Netflix that serves content. But Notion's success is an incentive for other products to focus on capturing more user content. Maybe Netflix needs to start a forum so that I can comment on every Bridgerton episode.

2. Turn data into action

Having data is cool. But putting it to work is even cooler. Notion launched Custom Agents to automate routine (and tedious) tasks like triaging bugs or onboarding a new hire (reminds me of the time an unnamed company got my first and last name mixed up, and no amount of AI could automate their way out of that mess).

Over a million Custom Agents were created in two months.

One of Notion’s customers, Braintrust, has a marketing team that built a custom agent that scans competitor changelogs and transcripts from sales calls every morning, then updates the sales one-pager before anyone's even awake. The product team for Heidi, an AI care partner, saves up to 60 hours a month with an internal Q&A agent that fields product questions.

Ramp literally pays for Google's Gemini, and it's still migrating its docs and project tracking into Notion, specifically so its agents have more to work with. They've built over 300.

There's always going to be data living outside Notion, but the more of it they pull in, the more powerful the agents get. That's why AI Notetaker and Notion Mail make so much sense: every meeting and every email becomes part of the company's knowledge base. Honestly, I wouldn't be surprised if they built a homegrown Slack competitor one day, just to capture the conversations too.

3. Introduce usage-based pricing

Notion is a traditional SaaS company with standard subscription-based pricing, but AI costs money every time it’s run. That means that they are able to offer AI features only when they can predict the cost, like summarizing meeting notes. A meeting is a meeting, and regardless of length, the cost to transcribe and summarize is pretty well defined.

Agents aren’t predictable. Notion doesn't define the workflow. The user does. They build the agent, decide what it touches, and how often it runs. One person's agent summarizes a doc once a week. Another chews through hundreds of databases every morning. Notion has no idea what your workflow costs, because Notion didn't design your workflow. Below are some potential costs for an agent, but even Notion's just guessing.

As you can see, the range of costs is huge. So Notion does the only sane thing: it splits the bill. Notion continues to charge its subscription price for the software and certain AI features, like its notetaker. For the agents, it follows a usage-based model and charges $10 per 1000 credits. The more complex the workflow or the more often it is run, the more credits it consumes.

That’s also why they let you choose the model powering the agent. You can use a cheap model for routine jobs, a frontier model when you need the horsepower, and Auto if you'd rather not decide. This flexibility lets you control how many credits you burn per run (if it were me, I’d always use the best-of-the-best and pass the bill along to my employer).

The lesson for any SaaS company shipping AI: the moment your costs become user-defined, your pricing model has to change with them. Subscription pricing doesn't work when usage is unpredictable. Hybrid pricing, base subscription plus usage credits, is probably the new default.

Extra Credit

Notion CEO on Lenny’s Podcast

Ivan Zhao's vision for the product was to give anyone the ability to customize software to their needs. That definitely tracks. While the interview doesn’t cover their AI features, there are some great stories about pivoting and starting over.

AI Will Eat Application Software

Great read by a16z on how AI will not bring about the destruction of SaaS businesses (like Notion) and instead will strengthen some of their moats. They note:

• Why AI can do far more with high-quality, proprietary data

• Why adaptable processes (aka custom agents) are the strongest moat

• How to disrupt incumbents with counterpositioning

After I finished writing this piece, I, of course, asked Notion AI what it thought. No notes. Sometimes, even a superpower intelligence is at a loss for words.

This article is a bit different than my previous ones as Notion is both a consumer product and a B2B product. My question for you is there any software that you use at work that feels like a consumer product?

— Amaraj (aka the man with a black belt in Notion templating)

The Meme