May 1, 2025 · 1h 11m · lennys-podcast
The rise of Cursor: The $300M ARR AI tool that engineers can’t stop using | Michael Truell
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Michael Truell, co-founder and CEO of Cursor (Anysphere), breaks down how his team built the market-leading AI code editor, scaled to $100M+ ARR, and architected custom models to redefine software engineering. He shares counterintuitive insights on startup execution, developer taste, hiring, and why AI will massively increase the demand for engineers.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Lenny holds 20.7% of the talking time here. How this is scored →
Lenny as informed peer 3.4 Guest teaching 4.6 Guest disagreement 1.2 Lenny pushing back 1.1
speaking balance: gold is Lenny, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 5:00 Rejection of prevailing AI programming paradigms
Michael forcefully dismisses both the status-quo text editing view and the popular hands-off chatbot paradigm as flawed form factors for software development.
Hardest push from Lenny ▶ 23:00 Pushback on background agent managementLenny challenges the ideal of agentic workflows by pointing out that managing flawed AI workers feels like being an overwhelmed engineering manager doing endless low-quality code reviews.
Biggest teaching moment ▶ 35:00 Dismantling the GPT wrapper assumptionMichael educates the audience on why standard foundation models fail at sub-300ms multi-file diff prediction, explaining how Cursor custom-trains internal models for autocomplete.
Lenny holds their own ▶ 37:40 Synthesizing model ensemble architecturesLenny demonstrates deep industry domain knowledge by connecting Michael's custom pipeline directly to OpenAI CPO Kevin Weil's ensemble-of-models framework.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Lenny as informed peer | Guest teaching | Guest disagreement | Lenny pushing back | Why |
|---|---|---|---|---|---|---|
| Episode Preview: The Vision and Growth Behind Cursor | 0 | 0 | 0 | 0 | Lenny delivers an introductory monologue and curated teaser clips highlighting Cursor's revenue milestones and technical vision. | |
| Sponsor Segment: Eppo Experimentation Platform | 0 | 0 | 0 | 0 | Host ad reads for Eppo experimentation platform and Vanta compliance automation. | |
| What Comes After Code: Redefining Programming with Intent | 5 | 6 | 3 | 1 | Michael reframes the industry debate by rejecting both conservative text-editing and naive Slackbot agent paradigms in favor of human-directed pseudocode abstractions. Lenny synthesizes this vision into a clear programmatic summary. | |
| The Evolution of Engineering Skills: Prioritizing Taste over Syntax | 4 | 6 | 2 | 2 | Michael defines why software engineering is shifting from syntax syntax-checking to logic design and taste. When Lenny probes whether this equates to 'vibe coding,' Michael clarifies that current vibe coding loses granular architectural control. | |
| The Genesis of Cursor and the Pivot from Mechanical Engineering | 3 | 5 | 1 | 1 | Michael recounts AnySphere's origin story, including their initial four-month exploration of CAD tooling for mechanical engineering before recognizing the immense ceiling in AI programming. | |
| The Advantage of High-Ceiling Markets in AI Innovation | 4 | 5 | 1 | 1 | Lenny extracts lessons about entering crowded but high-ceiling markets, while Michael outlines why full control over an IDE application enables radical user interface evolution. | |
| Structuring AI Interactions: Background Agents vs. Fast Foreground Loops | 4 | 5 | 2 | 2 | Lenny asks whether background AI workers in Slack will replace IDEs. Michael explains why tight foreground feedback loops and small iterative diffs outperform hands-off background agents for complex tasks. | |
| Rapid Iteration: From a Custom Editor to the VS Code Foundation | 3 | 5 | 1 | 1 | Michael describes the initial five-week sprint building a custom editor from scratch before user demand prompted them to rebase onto VS Code. | |
| Exponential Trajectories and Product-Led Execution | 4 | 5 | 1 | 1 | Michael attributes Cursor's rapid trajectory to product paranoia, relentless dogfooding, and ignoring traditional sales functions to focus purely on engineering velocity. | |
| Sponsor Segment: One Schema File Feeds 2.0 | 2 | 6 | 2 | 1 | Following a brief sponsor feature with Andrew Luo, Michael debunks the misconception that Cursor is a simple LLM wrapper by detailing their custom multi-file diff models. | |
| Pragmatic Post-Training on Open Source Foundations | 5 | 5 | 1 | 1 | Lenny references OpenAI's ensemble architecture concepts, while Michael explains how Cursor post-trains open-source weights to optimize speed and codebase indexing. | |
| The Trillion-Dollar Market for Automated Software Creation | 4 | 5 | 1 | 1 | Michael compares the automated software creation market to early search engines and PC operating systems, predicting massive R&D defensibility and a dominant platform winner. | |
| Why Incumbent Giants Face Structural Innovation Headwinds | 4 | 5 | 1 | 1 | Lenny asks why Microsoft Copilot lost early market dominance. Michael identifies structural incumbent challenges including bundling incentives and cross-departmental friction. | |
| Actionable Advice: Developing Model Taste and Pushing Limits | 4 | 5 | 1 | 1 | Michael shares tactical advice for engineers, emphasizing small task decomposition while running exploratory failure tests on sandbox projects to build intuition for model capabilities. | |
| Usage Anti-Patterns Across Junior and Senior Engineers | 4 | 5 | 1 | 1 | Michael analyzes how junior engineers over-rely on AI for wholesale generation while senior engineers excessively discount model capabilities. | |
| Cursor's Signature Two-Day Work Test Interview Process | 3 | 5 | 1 | 1 | Michael details Cursor's hiring evolution, highlighting their paid two-day in-person work trials that evaluate end-to-end craft and culture fit. | |
| Building a Lean, Product-Led Engineering Culture | 4 | 5 | 1 | 1 | Michael describes maintaining low headcount and high technical density, noting that hiring level-headed talent reduces organizational process overhead. | |
| The Multi-Decade Transformation of Knowledge Work | 3 | 5 | 1 | 1 | Michael reframes AI adoption as a multi-decade transformation that will surpass the impact of the internet by methodically automating deep vertical workflows. | |
| Cursor's Talent Search and Critical Open Roles | 4 | 5 | 2 | 2 | Lenny asks whether automated coding will eliminate software engineering jobs. Michael counters using Jevons paradox, arguing lower development costs will exponentially expand total software demand. |
Statements from this episode (31)
Truell: Cursor's goal is inventing an intent-based programming paradigm
“Our goal with Cursor is to invent, ah, sort of a new type of programming a very different way to build software that's kind of just distilled down into you describing the intent to the computer for what you want in the most concise way possible, ah, and really…”
Truell: Chatbot interfaces lack the precision needed for software engineering
“The problem with the chatbot style end of things is that lacks a lot of precision. If you want humans to have completely, you know, complete control over what the software looks like and how it works, you need to let them, you know, gesture at what they want t…”
Truell: Future coding will use terse pseudocode instead of raw code
“And so a world, you know, kind of after, You know, after code I think it looks like a world where you have a representation of the logic of your software that does look more like English, right? You have kind of written down, you can imagine in doctrine form, …”
Truell: Engineers will rely less on carefulness and more on taste
“I think we will get to a place where you will be able to be less careful as a software engineer, which right now is an incredibly, incredibly important skill. And yeah, we'll move a little bit from carefulness a little bit more towards taste.”
Truell: 'Vibe coding' breaks down when developers cannot understand scaling codebases
“By not understanding the details under the hood right now, you then very quickly get to a place where you're kind of limited. At a certain point where you create something that's big enough that you can't change.”
Cursor spent its first four months building AI for mechanical engineering
“There was a period of, you know, four months to begin with, where we were actually working on a very different idea, which was helping to automate and augment mechanical engineering and building tools for mechanical engineers.”
Truell: AI software has a uniquely high ceiling for leapfrog innovation
“I think the exciting thing about AI is, in a bunch of places, and I think this is, you know, very much still true of our space, and you can talk about how we think about that and how we deal with that, but you know, I think that just the ceiling is really high…”
Truell: AI coding tools must control the full IDE, not just plugins
“The extensibility that existing coding environments have is so, so, so limited. So if you think that the UI is going to change a lot, if you think that the form factor program is going to change a lot, you necessarily need to have control over the entire appli…”
Truell: Background AI agents best fit easily specified tasks like bug fixes
“And I think the background stuff, there's like a segment of programming that it's especially useful for, which is type of programming tasks where it's very easy to specify exactly what you want with, you know, without much description, and exactly what correct…”
Truell: Future IDEs will integrate external surfaces like Slack
“So I think that, you know, what the IDE is will totally change over time. And kind of our approach to, you know, having our own editor with premise non, it's gonna have to evolve over time. And I think that that will both include, you can spin off things from …”
Truell: Cursor's most successful developers use AI conservatively for scoped tasks
“So the customers we've seen have most success with AI, I think are still fairly conservative about some of the ways in which they use this stuff. And so I do think today that the most successful customers really lean on Things like you know, our next edit pred…”
Truell: Cursor's first prototype was built from scratch, not on VS Code
“The first version of Cursor was hand rolled. We now we use VS Code kind of as a base, like many browsers use Chromium as a base and it worked off of that. To begin with, we didn't, and built the prototype of Cursor from scratch, and that involved a lot of work…”
Truell: Cursor launched publicly three months after writing its first code
“And then we launched it out to the world within a couple of months from the first line of code. I think it was probably, probably three months.”
Truell: Cursor let sales and marketing fires burn early on
“Some of the normal things that people would maybe Reach for in, in building the company early on we really let those fires burn for a long time, especially when it came to things like sales and marketing.”
Truell: AI product builders operate between software and foundation model labs
“It's kind of a new form of product building where it's very interdisciplinary in that we are something in between a normal software company and then in between a normal software company and then a foundation model company in that you know we want to develop a,…”
Truell: Every magic moment in Cursor involves a custom in-house model
“At this point, every magic moment in Cursor involves a custom model in some way.”
Truell: Unlike writing, the next 30 minutes of coding is entirely predictable
“Code is this weird form of work where sometimes really the next five, 10, 20,
30 minutes of your work is entirely predictable from looking over your shoulder. And I would contrast this with writing.”
Truell: Cursor post-trains top open-source and closed-source foundation models
“Starting from the very best you know pre-trained models that exist out there. Often open source ones, you know, sometimes in collaboration with these big model providers that, that don't share their weights out into the world because the thing we care about le…”
Truell: AI entrenchment cannot protect companies from being leapfrogged
“And I, you know, I truly just think that the ceiling is so high that kind of no matter what, you know, entrenchment you build you can be leapfrogged. And I think that this resembles markets that are maybe a little bit different from normal software markets, no…”
Truell: AI defensibility comes from economies of scale in R&D
“If you're in a place where, you know, big investments and you know, having more and more great people working on the right path can keep giving you value, then you can get kind of these economies of scale of R&D, and you can kind of have, you know, deeply work…”
Truell: One company will build the tool creating almost all software
“I think that there will be one company and to be determined if it's going to be us, but I do think that there will be one company that builds the general tool that builds almost all the world's software. And that will be a very, very generationally big busines…”
Truell: Most of the original GitHub Copilot team has left Microsoft
“The group of people that worked on the first version of Copilot have, by and large, gone on to do other things at other places.”
Truell: Giant prompts are a recipe for disaster; break tasks down
“Instead, what I would do is I would chop things up into bits, and you can spend basically, you know, the same amount of time specifying things overall, but chopped up more. So you're specifying a little bit, you're getting a little bit of work, you're specifyi…”
Truell: Junior engineers over-rely on AI wholesale for end-to-end coding
“The junior engineers we see going a little too wholesale, relying on AI for everything, and we're not yet in a place where you can kind of do that end-to-end on a professional tool, you know, working with 1000 of other people within a long-lived code base.”
Truell: Senior engineers on average underrate AI and stick to existing workflows
“By and large, I would say, on average, as a group, the senior engineers underrate what AI can do for them and stick to their existing workflows.”
Truell: Cursor initially over-indexed on young candidates from top schools
“We maybe, we actually biased a little bit too much towards looking for people who fit the archetype of well-known school, very young, had done the things that were like, you know, high credential in those well-known school environments.”
Truell: Cursor uses a two-day on-site work test to evaluate candidates
“Core to how we interview, too, is, is actually we have people on site for two days and do, do a project with us, a work test project. And that has worked really well”
Truell: Most AI research ideas fade; simple ideas drive the progress
“A lot of the progress of AI can be attributed to some very simple, elegant ideas that have stayed around, and the vast majority of ideas that have been put out there haven't had staying power and haven't added a ton.”
Truell: AI shift will be more consequential than the internet over decades
“I think it's gonna be more consequential than the internet. I think it's gonna be more consequential than, you know, any shift in tech that we've seen since, since the advent of computers. And I think it's gonna take a while. And I think it's gonna be a multi-…”