A note from us: University students receive our Saturday Deepdive for free when they register with their university email address at: https://getsuperintel.com/plus-whitelist

Dear Readers,

This week we are doing something a little different. Instead of breaking down the news, we sat down with one of the people quietly rewiring how knowledge work gets done. Akshay Kothari co-founded Notion and runs it as COO, and in barely a year his company has gone from "the app where you keep your notes" to what he calls an agent-first workspace, a place where people, custom code, and AI agents all work side by side on the same project.

That shift is bigger than a product update. Notion's bet is that the messy infrastructure built for humans to collaborate, multiplayer editing, version history, the ability to roll a mistake back, is exactly what lets agents run for days without breaking things. Peter and I pushed Akshay on the hard parts: what an agent actually does for a non-technical team, where the human-versus-token math is heading, and why an $11 billion company still insists it is the challenger. It comes back, again and again, to one conviction: the point is not to replace people with tokens, but to make people ambitious enough to do far more. It is one of the sharpest conversations we have had on where this is all going.

Enjoy the read,

Kim & Peter

A Superintelligence interview with Akshay Kothari, Co-Founder & COO of Notion. Hosted by Kim Isenberg and Peter Thum.

Kim: Hi, I'm Kim Isenberg, Editor-in-Chief of Superintelligence, joined by my co-founder and co-host Peter Thum. Today we have a guest we've been looking forward to for a while: Akshay Kothari, co-founder of Notion. Akshay, thank you for taking the time.

Akshay: Thank you for having me. Big fan of your pod.

Kim: Most people know Notion as the place where they keep notes, docs, and wikis, but that story has shifted a lot over the last year. Back in February you opened up custom agents, and since then teams have built more than a million of them. Then on May 13th you launched the developer platform, and you now describe Notion as an "agent-first workspace" where people, custom code, and AI agents all work side by side in the same project. So when did it become clear internally that Notion was something fundamentally different?

Picture this. A buyer opens ChatGPT and asks for a recommendation in your category. Your competitor's name comes up. Yours doesn't. And that buyer never makes it to your website.

That's happening right now in markets everywhere. And most teams don't know it's happening because it never shows up in their analytics.

HubSpot AEO shows you exactly where your brand stands in AI search, where competitors are getting recommended instead of you, and tells you specifically what to fix. No expertise needed.

Try it free for 28 days. Just $50 a month after.

Akshay: It's interesting that people think of Notion as note-taking. The original idea was always about democratizing software-making. We felt the industry had swung too hard toward rigid, specialized software, and we expected the pendulum to swing back toward horizontal platforms where you can do many things in one place. We took inspiration from Lego: the same bucket of blocks can become a castle or a car. Roughly 20 to 25 building blocks make up most software, so we thought, if we gave you those blocks, would you build something custom-fit for you?

But people don't wake up wanting to build software. They wake up to do their job. We'd built this dream platform and for a few years nobody came. So we used the building blocks ourselves and shipped a few templates: a to-do list, a task manager, a mini CRM. Those put Notion on the map. They gave a community of builders a starting point, and once people had a to-do list they'd go deeper and realize they could change everything about it. They remixed the templates, got proud of their creations, and told their friends. That's how Notion grew.

So note-taking was a wedge into the market, a way to get distribution. The long-term dream was always a platform where multiple people work together and build software that helps them collaborate. Now, over the last three years, AI has completely changed the way we work. We've gone from AI assisting you to agents doing end-to-end work. And we realized this past year that all the infrastructure we'd built for human-to-human collaboration is a perfect fit for agents too. If you and Peter both write a document, you can see which parts each of you wrote; add an agent, and you can inspect what the agent wrote as well. Everything humans produce is great context for agents to do real work, and everything agents produce (documents, databases, workflows) is easy for humans to inspect and give feedback on. Over the next five or ten years, humans and agents will coexist, and every company will have to decide what infrastructure they use. Notion has evolved to enable not just humans but agents. The goal isn't to replace humans; it's to augment everyone who uses the platform.

Kim: I love that angle. You said teams have built over a million custom agents. Give me the most concrete example of an agent doing real work for a non-technical team today.

Akshay: There are so many, and when you look at who uses them, every function of the company is represented. The one that really took off internally came from our People team. They gave it a mascot and called it "Smilers." It uses all the knowledge in our workspace to answer questions in a Slack channel called "Environment Ask," where people ask about anything they need in the office: where's the printer on the ninth floor, where's the health insurance info, what happened to that benefit. Historically the same people answered the same question thirty times as new employees joined.

Now Smilers has access to all that information. If it knows the answer, it responds in seconds. When it doesn't, it creates a ticket the People team can answer, and that interaction becomes part of the knowledge base. It's a self-healing knowledge base: a question only needs to be answered once. This saves the team hours every day, and with ten offices worldwide, someone in Tokyo no longer waits twelve hours for an answer. The coolest part is that no engineers were involved. Each custom agent has three aspects: the trigger, the context, and the instructions, the "job description." And the job description isn't code or a flowchart; it's plain English, or your language. The agent just follows it.

Kim: It reminds me of Andrej Karpathy saying a year or two ago that the most popular coding language would be English. It's coming true. But most productivity tools just make old habits faster. Beyond Smilers, does Notion's AI actually change how people work?

Akshay: When a new technology arrives, the first few years are usually spent replicating what we already did. With the printing press, the first things we printed looked like handwritten manuscripts; only after typography did we realize we didn't need to imitate handwriting, and that setting type was better. AI might be in the same spot. We're about three years in, and today we're mostly looking at our day-to-day busywork and asking AI to take it off our plate so we can spend more time on strategic work. As we shed busywork, we'll reimagine how we spend the rest of our time and how we interact with agents.

There's a great meme I'll send you: on one side a humanoid robot in a field cutting corn one stalk at a time, the way a human would; on the other, a combine harvester sweeping through the field. Right now we're still on the left, doing things one by one the way humans do. Over time we'll move to the combine and invent new ways of working that make more sense in this new world.

Kim: I couldn't agree more that we're still early. But let's stay here. The pitch is developers and agents building in Notion side by side, with Workers at the center. What couldn't you do before that you can do now?

Akshay: You have to think about how to give agents the right context. For people who live in Notion, Notion itself is good context, but that's not enough at larger companies that use many other tools. Over the last couple of years we built fifteen or twenty connectors (Google Drive, Asana, Salesforce), but there's a long tail of tools, including internal ones. We wanted to let you connect any data source, and that's what Workers does: it uses AI to write a small piece of code that syncs a Notion database with any other source, refreshing roughly every fifteen minutes. It's efficient because you're not burning GPU context every time; you convert GPU work into CPU work once, write the data, and that source becomes context for everything you do.

One example: our social team uses Sprout to track posts across channels, but Sprout isn't connected to the product releases and knowledge we have in Notion. So someone built a connector to the Sprout data that syncs every fifteen minutes, and now I can build a social-media-analyst agent on top of it. I can ask which tweet got the most impressions last week, then ask it to look at an upcoming release and create a launch plan based on what performed well. I've connected Notion content, product bugs from another system, and social data that used to be hard to reach. People do this for personal use too, connecting Spotify and Strava, or Salesforce and Gong.

Kim: As far as I know, external agents now work natively inside Notion too. Is that right?

Akshay: Right.

Kim: So when autonomous agents write and reorganize a workspace, which failure modes worry you most, and what are you building in to guard against them?

Akshay: The profession changing fastest is software programming, because coding agents are getting extremely good. But the bigger reason programmers can run agents for days isn't just the models; it's the infrastructure. GitHub lets many people contribute to the same codebase, resolve conflicts, review work, and roll things back, all at once. One way to think about Notion is that we're bringing that same infrastructure to the rest of knowledge work. How do we let our CFO, our marketing leader, our People leader run an agent for days? That requires a multiplayer collaboration layer, the ability to inspect what agents do, and the ability to roll back. We're very focused on enabling that for the future of knowledge work.

Kim: You've said these tools won't stay developer-only for long. Is the endgame a workspace that extends itself, and how do you keep it simple for non-developers? I think of Codex, with OpenAI merging ChatGPT and Codex into a kind of super-app that's easier for non-coders. How do you think about that?

Akshay: The developer platform lets you do three things. First, connect to any data source through Workers. Second, build any action: if your context is in Notion and you want a button to refund a customer in Stripe, you can build that inside Notion. Third, bring any agent, so you don't have to use the Notion agent; you can use the Claude agent or others. The platform does require some technical knowledge: spinning up a coding agent, writing a little code, working in the terminal. The next step is to let mainstream users do all of that just by speaking English inside Notion. The step after that is to package it up. When we look at the workflows people build, there's a lot of consistency. Lots of companies want that social-media setup, so we can ship it as a packaged agent: it connects to your Sprout, pulls the context into a database, and gives you an agent that queries it. You essentially get a social-media analyst as an intern. That's a real shift: instead of handing you project-management software, I hand you an agent that runs end-to-end. We'll take the most common things people do and give them to you out of the box.

Kim: Fantastic. Let me pass it to Peter to switch perspective.

Peter: We've talked about the product opportunities AI opens up, but I'm curious about the competitive angle. After eleven years you're an $11 billion incumbent, so what you have to worry about is different than when you set out. When we first started talking you described yourself as a startup. How do you think about the shift from a small company to a player defending territory it's won?

Akshay: I'd argue we're still a tiny startup. When you compete with big tech worth a few trillion, and now labs also worth trillions, $11 billion is about one percent. I don't go to sleep thinking we're the incumbent; I feel like we're still the challenger, and we operate like one. The shift over the last two years, from a SaaS company to an AI company, has been a real fight to survive and to thrive. We just made the top ten of the AI 50. The most interesting part is that the opportunity is an order of magnitude larger. As a share of US GDP, software spend was about two percent, but knowledge work is about forty percent. Software can now actually do the work, not just organize it. In this new world we deliver the work: you pay for what a custom agent does, and you decide whether it was worth the money. It takes twenty or thirty years for a new technology to find its most useful applications, and we're maybe ten percent into AI. A new set of companies will become the new guard of knowledge work; everything is up for grabs, which is the coolest part. We hope that in a decade or two Notion becomes a default tool, but we're far from it today.

Peter: You described your space as a horizontal platform with a database that can integrate other people's interfaces fairly easily. The old question was "make or buy"; for you it's "make, buy, or integrate," and you can do that more freely than competitors. Do you see all this activity as fodder for that decision, rather than as the threat that four people with an AI will steal your market share?

Akshay: In the enterprise there are always multiple winners. Consumer is harder, more winner-take-all. But we provide real value. My pitch is this: if you believe humans and agents will collaborate well over the next ten years, then companies that leverage agents will move much faster than those that don't. So as a leader you have to decide whether you're building the right infrastructure to leverage agents across the whole organization, not just engineering. And you have to keep long-term optionality. You don't want to bet your entire future on a single model or company. What matters most is the best dollar-per-intelligence your organization can leverage.

It's a fascinating new paradigm. We used to think in headcount; soon, maybe as early as this year, you'll think about headcount alongside token cost, and you'll ask what you actually got for that spend. We're moving fast from "token-maxing" to "what outcomes do I want?" Notion is the infrastructure that gives you that visibility: a multiplayer platform that adheres to your security and privacy, gives you access to all the models, lets you version-control the work, and lets you decide which agents to use, built on the system of record we've developed over a decade. We're an applied AI company and a fan of a multi-model world, because more choice means better intelligence per dollar.

Some companies will build it themselves, and power to them. But the few I've talked to are realizing how hard it is to keep their agent harness current with the latest coding agents. We're three and a half years in, and Notion has rebuilt its agent harness roughly seven times, every six months like clockwork. So if you're a shoe manufacturer building your own harness to save money, every six months you'll need a team to rebuild it. That's the real build-versus-buy question: build it yourself for more control and bet on one model company, or buy infrastructure from Notion that gives you both control and optionality.

Peter: I'll ask based on what you just said. To what degree are you thinking about helping companies manage the relationship between value created by employees and value created by tokens? How much does optimizing that balance play into your role as an intermediary?

Akshay: It starts with your own value system, and Notion's is very much "long humans." There are two schools of thought in AI. One is the AGI school: build something better than humans, and fifty percent of people lose their jobs. We're not on that tree. Our heroes are Douglas Engelbart and Alan Kay, for whom computers were always about augmenting human intelligence. It's the bicycle for the mind; maybe with AI today it's the electric car for the mind. So for us the question isn't "can I do everything I do today with half the people?" It's "can I be two times more productive?" You get there because the half of your time spent on work you don't want to do can now go to agents. People give humans too little credit for what they could do with more time. We assume everything gets automated and we'll just sit on a beach, but people become more ambitious instead. It's already playing out that way: we've gone from ChatGPT being miraculous to needing only a week with a new model before we say "this can't do that yet." It's accelerating and recursive, but I'm bullish on humans constantly figuring out how to use this to do more.

Peter: One reality companies face is that the answer is a moving target. There's no fixed ratio of people to tokens; depending on your industry, management, and goals it can change very quickly, and you don't know where you'll succeed or fail.

Akshay: There's going to be a real debate here. For a long time, tokens were an insignificant share of the budget. As that gets to ten or twenty percent, the first wave is people asking "what am I getting for this?" Some outcomes are clearer than others. For some things people will say, "I don't think you need to spend $1,500 to triage your email."

Kim: We're having that debate right now. I think it was Uber's CEO or COO saying, "We're spending so much on tokens, where's the return?"

Akshay: You can't entirely blame people, because every CEO has been saying "token-max, use AI all day." So naturally there's a swing back: "what is all this cost doing for my company?" One thing we do is give you the option to choose. An agent can use frontier models, but it can also use open-weight models, which can be one-twentieth the cost. Smilers doesn't need the latest model; an open-weight model handles it reasonably well. Other companies come at it from a finance lens. Ramp just published their thinking, essentially asking, as a CFO, "twenty percent of your spend goes to this, do you even know what you're getting?" That forces the question. The best companies will find an optimal setup where it's not about replacing humans with tokens, but leveraging humans to think more ambitiously and tokens to do more, things that run overnight and move the company forward. It'll be constant iteration for months, maybe years.

Kim: I'd love to stay on token-maxing, but we have more questions. Notion AI, Copilot, and Gemini are all converging on the same user. What's the Notion advantage that survives once AI features become a commodity?

Akshay: You're right that products are converging faster than ever. In the previous decade there was always some specialization to jump into; now you can start with a niche use case and quickly realize you need the power of agents, and to get that you need the context, so you end up building the whole thing. So what differentiates Notion? It goes back to our values and mission. We're excited about "malleable software": you speak a few words in English and get a piece of software you can then customize and make your own, the way you'd move blocks around. That's not for everyone, and in some ways it's art; we want this thing to exist. It was the dream of the computing pioneers of the '70s and '80s, so it's not even our idea; we're just trying to bring it to life. That matters to us more than the business or the valuation.

Kim: Five-person teams now hit real ARR with no outside capital. How real is that threat? Is specialization beating integration right now?

Akshay: That's always been true; we worried more about upstarts than big tech. As things converge, the question becomes what's actually valuable, and it might be your community, brand, or distribution. Just having the best product idea matters less, because everyone's building the same thing. You have to think about network effects, brand, distribution, and values. Someone said recently: if I'm making a product decision and it's not informed by my values, it's not worth it, because the agent can probably do it better. If I give my agent all the context across Slack, Notion docs, and Salesforce, it already has more context than I do. So part of it is asking what we stand for, what our mark on the world is.

Kim: Peter?

Peter: You started with product-led growth and became a company with one foot in enterprise and one in consumer, where you'll face a lot of vertical competition. Does the question of what your values are get driven by who you're trying to serve over time? Does competition ultimately dictate a reconsideration of your principles?

Akshay: Great question. There's been a big unlock with custom agents. Traditionally there was only one way to grow: product-led to $100 million, add sales to reach a billion, then go enterprise to reach ten billion, because large companies pay more. So Notion always assumed we're an enterprise business long-term, because that's the only path to tens of billions. What's shifted is that the industry is moving from selling software to selling work, and that changes the growth equation. Back to the five-person startup: at our most expensive plan that's maybe $1,500 a year in seats, max. But that same startup today could be a $50K or $100K account, because they can use custom agents to do real work that's worth paying for. So every company, even startups, has two paths: keep growing toward enterprises, because half the world works there, and decide what you're selling, the old software or the work. If you're selling work, there's an order-of-magnitude-larger opportunity ahead.

We've always grown bottoms-up: individuals, then startups; the startups became mid-market companies, which pulled us into the enterprise, and now several customers spend over a million a year with us. But instead of betting the whole company on enterprise, we work with the enterprises that want to lean into AI. Every engineer is a trade-off: push the product to the frontier, or build the long tail of enterprise asks? We want to serve the most innovative teams at large enterprises without depending on growing wall-to-wall in every account. We keep the product at the frontier, attack anything that blocks us at the enterprise, and trust that today's startups, who become tomorrow's enterprises, will keep pulling us in.

Peter: SaaS companies used to build a roadmap, take it to market, and build on success in steps. To what degree, because of AI, is the gold now more like a truffle hunt, where the value sits in small teams coming up with ideas you can leverage across the platform, versus ideas coming from your own team thinking about users' needs?

Akshay: TBD, but the world I want to live in has more people contributing than a select few. My coworker Max has a good line: if AI is building all the software that's easy to build, where's the quality software? Over the last year or two, which piece of quality software is clearly better than what we had before? In a world where everything is easily buildable, it comes back to basics: the people who care, who have a real point of view, who can imagine how this technology shapes our future, still have a role. You can imagine agents getting good enough to build tasteful things on their own, but that hasn't happened yet, so I'm optimistic humans and teams keep playing that role.

One thing we see: in a thousand-person company, maybe ten people are AI-forward and really know how to prompt and build custom agents. What helps that knowledge diffuse is that a custom agent in Notion is a document you can share. Peter, you build an agent and share it with Kim; now Kim is a co-creator and can learn how you built it. When you're both happy with it, you roll it out company-wide, and someone can duplicate it and make it their own. Suddenly what one person built diffuses across the organization, which beats people using agent products solo. We extended our multiplayer collaboration to agents, so you can co-create them. One data scientist at Notion built a custom agent called Data Scout, synced with Snowflake tables and able to write scripts, and now everyone has a personal data scientist. Every team I've worked with complained about not having enough data scientists, and now one of them made roughly ninety percent of that capability available to the whole company. So I'm bullish: there's a lot we can do as a community to educate people, but each person will build wonderful things.

Kim: I won't start a philosophical debate, but that Anthropic post on recursive self-improvement popped into my head, and the question of where the human is in the loop in five years. Five years is maybe too long, so let's say two years out, or even one given how fast this moves: what does the human-AI layer in Notion look like if you've gotten it right?

Akshay: One year out is fairly straightforward: we're building the same infrastructure software engineers have and giving it to the rest of knowledge work. You can frame it two ways: we're giving that infrastructure to knowledge work, or more knowledge work is becoming programmable, so you can use coding agents for things like legal review or people ops. We already see early signs. The question then is what people do with the freed-up time, and we want to help them become more ambitious and build the future they want.

Peter: Do you think the companies that don't succeed will be the ones that are dictatorial, or that won't listen?

Akshay: A couple of ways. One failure is not waking up to the technology, betting only on humans; that's wrong, because this is one of the most crucial technologies in decades. The other failure is the opposite: thinking only "what can I do now, and can I do it with agents and cut my workforce?" without raising your ambition. You'll become very efficient, but if you're not figuring out how to grow, that's tough. For SaaS companies it used to be amazing to do 3x, 3x, 2x, 2x. All those principles get thrown out now; companies can grow 25x in a single year. So the thing to ask is whether you can get your company onto the exponential, moving from year-over-year to month-over-month to week-over-week growth, by leveraging agents to do more. The more companies that get onto that exponential in the value they provide, the more chance they have to become a defining company of this era.

Peter: Leaders have to get into that messy middle you're building a platform to regulate, and the relationship between their team and AI won't be clear, maybe ever, certainly not in the next few months. That shift to measuring success weekly or monthly has to happen before you even know where you'll squeeze that success from.

Akshay: Exactly right, and that's why I'm excited. Someone said New York is about a week behind San Francisco, and the rest of the world maybe months behind. Some of our customers are clearly living in the future, and it's a question of how soon it diffuses. Our job is to take the learnings from the top of the class and provide infrastructure so even the median company can succeed. One concept we've talked about this year is "Fortune 5 Million": we're not building just for the Fortune 500; we want to bring the Fortune 5 Million forward and help them leverage AI to survive and thrive. One story to leave you with: a Chick-fil-A franchise owner built Workers into all the internal data he uses to run his franchise, figured out how to run the most efficient one using Notion, and was invited to speak at their summit to show other franchisees how he did it. Someone not technical, but technically curious, set up Workers, built some agents, and ran his business with them. Those are the customers that are the fuel to come to work every day.

Tell him the campaign. Viktor pulls last quarter's performance from Meta and TikTok, scrapes competitor ads, drafts the brief, posts it for review. You edit, he ships the creative requests to your designer. Inside Slack.

Get Started for Free

Kim: Where can people follow up, and what should they try right now? Give us a quick TL;DR.

Akshay: We're on Twitter, LinkedIn, and YouTube. We're @NotionHQ on Twitter, and I'm @akothari, with open DMs. Over the last six to nine months we've spent a lot of time not just building products but telling our story, so you'll see our product releases and our values come to life on those platforms.

Kim: Fantastic. Akshay, it was such a pleasure, and I learned so much. Thank you.

Peter: Thank you very much. Nice to meet you.

Akshay: Thank you for having me. Great to meet you.

Akshay Kothari is Co-Founder and COO of Notion, the workspace he has helped grow into an $11 billion company recently named to the top ten of the Forbes AI 50. He runs the operating side of the business, including sales, marketing, customer success, support, people, legal, and finance, and has become one of the clearest voices on what an agent-first company actually looks like from the inside.

His path there ran through an earlier exit. At Stanford, where he earned a master's in electrical engineering after a bachelor's at Purdue, he co-founded Pulse, the news-reading app that began as a class project and was acquired by LinkedIn in 2013 for around $90 million. He stayed on at LinkedIn as VP of Product and head of its India business, first invested in Notion that same year, and joined full-time as COO in 2018.

That blend of founder, operator, and early believer shapes how he talks about the road ahead. Kothari notes that Notion has rebuilt its agent harness roughly every six months, and argues that the collaboration layer built for humans, multiplayer editing, version history, and rollback, is exactly what makes agents safe to run across a whole organization. His through-line is that the goal is not to swap people for tokens, but to give every team, from the Fortune 500 to what he calls the "Fortune 5 Million," the leverage to be more ambitious.