Podorbit
DiscoverPodcastsGuestsCompanies
Podorbit

Tech podcasts · indexed by guest & company

PrivacyTermsSupport
support@podorbit.app
Cursor

Cursor

cursor.com →

People from Cursor

Michael Truell

CEO and Co-Founder

Aman Sanger

Co-Founder

Sualeh Asif

CPO and Co-Founder

Adam Ward

Head of Talent

Adam Weinstein

Co-Founder

Andrew Milich

Unknown role

Arvid Lunnemark

Co-Founder

Ashvin Nair

Model development lead

Federico Cassano

Researcher

Jacob Jackson

Unknown role

Sign in to see 1 more people from Cursor

Free, one click with Google. Browse every guest and company across the catalog.

Episodes featuring Cursor (14)

Lenny's Podcast: Product | Career | Growth/Aug 9, 2026/1h 31m

The playbook for building high-talent-density teams | Adam Ward, Head of Talent at Cursor

Adam Ward is the Head of Talent at Cursor, one of the fastest-growing developer tools in history. Before joining Cursor, he founded Growth by Design, an independent recruiting and talent strategy firm that helped build teams at the most ambitious AI and technology companies in the world. Adam has spent more than 20 years building elite, high-talent-density teams across the industry and is widely regarded as one of the most effective and creative recruiters in tech. In our in-depth conversation, we discuss: 1. Inside today’s “tale of two cities” talent market 2. Why the traditional recruiting funnel—what Adam calls the “funnel of doom”—leads to mediocre hires 3. Adam’s three-step playbook: scoping, mapping, and relentless pursuit 4. The worst question to ask when sourcing great talent 5. The rise of the forward deployed engineer—and how to become one 6. The biggest mistake founders make when hiring their first recruiter — Brought to you by: WorkOS—Make your app enterprise-ready, with SSO, SCIM, RBAC, and more Mercury—Radically different banking, now with Command — Where to find Adam Ward • X: https://x.com/wardadamp • LinkedIn: https://www.linkedin.com/in/adampward — Where to find Lenny: • Newsletter: https://www.lennysnewsletter.com • X: https://twitter.com/lennysan • LinkedIn: https://www.linkedin.com/in/lennyrachitsky/ — In this episode, we cover: (00:00) Introduction (03:01) The state of the hiring market (06:41) What roles are trending up and what roles are trending down (11:25) The three-step hiring framework that changes everything (21:58) Getting the right people to actually respond (25:42) Keeping talent a top priority (29:16) Inside Cursor’s recruiting operation (32:19) How to relentlessly pursue top talent (35:42) “Caring is free” (38:33) The conversation most companies underestimate (39:19) Rethinking the atomic unit of a search (40:12) Other recruiting tactics (45:04) What the on-site is really for (49:35) The importance of work trials (53:10) The offer stage (57:53) Negotiations and comp (01:01:50) The details that actually close the deal (01:07:23) What it’s like working with an exec team who values recruiting (01:09:42) Talent density (01:11:18) The first recruiting hire most founders get wrong (01:15:00) Building incredible teams (01:18:12) The time between offer and acceptance (01:19:48) What’s next for the recruiting function (01:21:00) Lightning round — Referenced: • Cursor: https://cursor.com • Trilogy: https://trilogy.com • Brie Wolfson on LinkedIn: https://www.linkedin.com/in/brie-wolfson-17758724 • Inside Cursor: https://colossus.com/article/inside-cursor • SpaceX: https://www.spacex.com • Joe Gebbia’s website: https://joegebbia.com • The rise of Cursor: The $300M ARR AI tool that engineers can’t stop using | Michael Truell (co-founder and CEO): https://www.lennysnewsletter.com/p/the-rise-of-cursor-michael-truell • The Pitt on HBO Max: https://www.hbomax.com/shows/pitt-2024/e6e7bad9-d48d-4434-b334-7c651ffc4bdf • The Bear on Hulu: https://www.hulu.com/series/the-bear-05eb6a8e-90ed-4947-8c0b-e6536cbddd5f • Granola: https://www.granola.ai • Wispr Flow: https://wisprflow.ai • 11 products I love, free for a year—the biggest Product Pass expansion in 2 years: https://www.lennysnewsletter.com/p/productpass-summer2026launch • How to debug a team that isn’t working: the Waterline Model: https://www.lennysnewsletter.com/p/how-to-debug-a-team-that-isnt-working • The high-growth handbook: Molly Graham’s frameworks for leading through chaos, change, and scale: https://www.lennysnewsletter.com/p/the-high-growth-handbook-molly-graham — Recommended books: • Emotional Intelligence: Why It Can Matter More Than IQ: https://www.amazon.com/dp/055338371X • High Growth Handbook: Scaling Startups from 10 to 10,000 People: https://www.amazon.com/High-Growth-Handbook-Elad-Gil/dp/1732265100 • Scaling People: Tactics for Management and Company Building: https://press.stripe.com/scaling-people — Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email podcast@lennyrachitsky.com. — Lenny may be an investor in the companies discussed. To hear more, visit www.lennysnewsletter.com

Adam Ward

Head of Talent

CursorCursor
Training Data/May 26, 2026/46m

How Cursor Trained Composer on Fireworks: Distributed Infrastructure for High-Performance RL

Cursor's Federico Cassano and Fireworks' Dmytro Dzhulgakov explain how they collaborated to build Composer as a specialized foundation model. The core insight: models have finite capacity in their weights, and allocating all those bits to the singular task of software engineering in Cursor frees the model to be both better at the task and far more efficient at inference. Rather than start from pre-training and work up, they took an unconventional top-down approach — mid-training and RL on top of an open-source base to get a useful model into users' hands fast, then specializing the model around real Cursor usage. With Fireworks providing distributed infrastructure, Composer delivers frontier-class coding performance with the speed of a much smaller model. Hosted by Sonya Huang, Sequoia Capital

Federico Cassano

Researcher

CursorCursor
Dmytro Dzhulgakov

Co-Founder

Fireworks AIFireworks AI
Latent Space: The AI Engineer Podcast/Dec 30, 2025/45m

[State of RL/Reasoning] IMO/IOI Gold, OpenAI o3/GPT-5, and Cursor Composer — Ashvin Nair, Cursor

From Berkeley robotics and OpenAI’s 2017 Dota-era internship to shipping RL breakthroughs on GPT-4o, o1, and o3, and now leading model development at Cursor, Ashvin Nair has done it all. We caught up with Ashvin at NeurIPS 2025 to dig into the inside story of OpenAI’s reasoning team (spoiler: it went from a dozen people to 300+), why IOI Gold felt reachable in 2022 but somehow didn’t change the world when o1 actually achieved it, how RL doesn’t generalize beyond the training distribution (and why that means you need to bring economically useful tasks into distribution by co-designing products and models), the deeper lessons from the RL research era (2017–2022) and why most of it didn’t pan out because the community overfitted to benchmarks, how Cursor is uniquely positioned to do continual learning at scale with policy updates every two hours and product-model co-design that keeps engineers in the loop instead of context-switching into ADHD hell, and his bet that the next paradigm shift is continual learning with infinite memory—where models experience something once (a bug, a mistake, a user pattern) and never forget it, storing millions of deployment tokens in weights without overloading capacity. We discuss: * Ashvin’s path: Berkeley robotics PhD → OpenAI 2017 intern (Dota era) → o1/o3 reasoning team → Cursor ML lead in three months * Why robotics people are the most grounded at NeurIPS (they work with the real world) and simulation people are the most unhinged (Lex Fridman’s take) * The IOI Gold paradox: “If you told me we’d achieve IOI Gold in 2022, I’d assume we could all go on vacation—AI solved, no point working anymore. But life is still the same.” * The RL research era (2017–2022) and why most of it didn’t pan out: overfitting to benchmarks, too many implicit knobs to tune, and the community rewarding complex ideas over simple ones that generalize * Inside the o1 origin story: a dozen people, conviction from Ilya and Jakob Pachocki that RL would work, small-scale prototypes producing “surprisingly accurate reasoning traces” on math, and first-principles belief that scaled * The reasoning team grew from ~12 to 300+ people as o1 became a product and safety, tooling, and deployment scaled up * Why Cursor is uniquely positioned for continual learning: policy updates every two hours (online RL on tab), product and ML sitting next to each other, and the entire software engineering workflow (code, logs, debugging, DataDog) living in the product * Composer as the start of product-model co-design: smart enough to use, fast enough to stay in the loop, and built by a 20–25 person ML team with high-taste co-founders who code daily * The next paradigm shift: continual learning with infinite memory—models that experience something once (a bug, a user mistake) and store it in weights forever, learning from millions of deployment tokens without overloading capacity (trillions of pretraining tokens = plenty of room) * Why off-policy RL is unstable (Ashvin’s favorite interview question) and why Cursor does two-day work trials instead of whiteboard interviews * The vision: automate software engineering as a process (not just answering prompts), co-design products so the entire workflow (write code, check logs, debug, iterate) is in-distribution for RL, and make models that never make the same mistake twice — Ashvin Nair * Cursor: https://cursor.com * X: https://x.com/ashvinnair_ Full Video Episode Timestamps 00:00:00 Introduction: From Robotics to Cursor via OpenAI00:01:58 The Robotics to LLM Agent Transition: Why Code Won00:09:11 RL Research Winter and Academic Overfitting00:11:45 The Scaling Era and Moving Goalposts: IOI Gold Doesn't Mean AGI00:21:30 OpenAI's Reasoning Journey: From Codex to O100:20:03 The Blip: Thanksgiving 2023 and OpenAI Governance00:22:39 RL for Reasoning: The O-Series Conviction and Scaling00:25:47 O1 to O3: Smooth Internal Progress vs External Hype Cycles00:33:07 Why Cursor: Co-Designing Products and Models for Real Work00:34:14 Composer and the Future: Online Learning Every Two Hours00:35:15 Continual Learning: The Missing Paradigm Shift00:44:00 Hiring at Cursor and Why Off-Policy RL is Unstable This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.latent.space/subscribe

Ashvin Nair

Model development lead

CursorCursor
The a16z Show/Dec 16, 2025/52m

Ryo Lu (Cursor): AI Turns Designers to Developers

Ryo Lu spent years watching his designs die in meetings. Then he discovered the tool that lets designers ship code at the speed of thought: Cursor, the company where Ryo is now Head of Design. In this episode, a16z General Partner Jennifer Li sits down with Ryo to discuss why "taste" is the wrong framework for understanding the future, why purposeful apps are "selfish," how System 7 holds secrets about AI interfaces, and the radical bet that one codebase can serve everyone if you design the concepts right instead of the buttons.   Timecodes: 00:01:45 - Design Becomes Approachable to Everyone 00:02:36 - From Years to Minutes: Product Feedback Loops Collapse 00:07:54 - "Each role used their own tool...their own lingo" 00:13:15 - "If you don't have an opinion, you’ll get AI slop" 00:17:18 - The Lost Art of Being a Complete Builder 00:21:42 - Design Is Not About Aesthetics 00:28:57 - User-Centric vs System-Centric Philosophy 00:34:00 - AI as Universal Interface, Not Chat Box 00:38:42 - "Simplicity is the Biggest Constraint" 00:43:42 - "I Don't Sit in Figma All Day Making Mocks" 00:46:33 - RyoOS: Building A Personal Operating System 00:48:45 - "We've been doing the same thing since 1984"   Resources: Follow Ryo Lu on X: https://x.com/ryolu_ Follow Jennifer Li on X: https://x.com/JenniferHli Follow Erik Torenberg on X: https://x.com/eriktorenberg   Stay Updated: If you enjoyed this episode, be sure to like, subscribe, and share with your friends!   Find a16z on X: https://x.com/a16z Find a16z on LinkedIn: https://www.linkedin.com/company/a16z Listen to the a16z Podcast on Spotify: https://open.spotify.com/show/5bC65RDvs3oxnLyqqvkUYX Listen to the a16z Podcast on Apple Podcasts: https://podcasts.apple.com/us/podcast/a16z-podcast/id842818711 Follow our host: https://x.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures. Stay Updated: Find a16z on YouTube: YouTube Find a16z on X Find a16z on LinkedIn Listen to the a16z Show on Spotify Listen to the a16z Show on Apple Podcasts Follow our host: https://twitter.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

Ryo Lu

Head of Design

CursorCursor
The a16z Show/Nov 10, 2025/28m

Michael Truell: How Cursor Builds at the Speed of AI

When four MIT grads decided to build a code editor while everyone else was building AI agents, they created the fastest-growing developer tool ever built.  Cursor CEO Michael Truell joins a16z’s Martin Casado to discuss the deliberate constraints that led to breakthroughs: why they rejected the "democratization" narrative to focus on power users, how their 2-day work trials test for agency over credentials, and the strategic decision to own the editor when conventional wisdom said it was impossible.   Resources: Follow Michael on X: https://x.com/mntruell   Stay Updated:  If you enjoyed this episode, be sure to like, subscribe, and share with your friends! Find a16z on X: https://x.com/a16z Find a16z on LinkedIn: https://www.linkedin.com/company/a16z Listen to the a16z Podcast on Spotify: https://open.spotify.com/show/5bC65RDvs3oxnLyqqvkUYX Listen to the a16z Podcast on Apple Podcasts: https://podcasts.apple.com/us/podcast/a16z-podcast/id842818711 Follow our host: https://x.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures. Stay Updated: Find a16z on YouTube: YouTube Find a16z on X Find a16z on LinkedIn Listen to the a16z Show on Spotify Listen to the a16z Show on Apple Podcasts Follow our host: https://twitter.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

Michael Truell

Co-Founder

CursorCursorAnysphereAnysphere
The New Stack Podcast/Sep 19, 2025/48m

Why Linear Built an API For Agents

Cursor, the AI code editor, recently integrated with Linear, a project management tool, enabling developers to assign tasks directly to Cursor's background coding agent within Linear. The collaboration felt natural, as Cursor already used Linear internally. Linear's new agent-specific API played a key role in enabling this integration, providing agents like Cursor with context-aware sessions to interact efficiently with the platform. Developers can now offload tasks such as fixing issues, updating documentation, or managing dependencies to the Cursor agent. However, both Linear’s Tom Moor and Cursor’s Andrew Milich emphasized the importance of giving agents clear, thoughtful input. Simply assigning vague tasks like “@cursor, fix this” isn’t effective—developers still need to guide the agent with relevant context, such as links to similar pull requests. Milich and Moor also discussed the growing value and adoption of autonomous agents, and hinted at a future where more companies build agent-specific APIs to support these tools. The full interview is available via podcast or YouTube. Learn more from The New Stack about the latest in AI and development in Cursor AI and Linear:   Install Cursor and Learn Programming With AI Help Using Cursor AI as Part of Your Development Workflow Anti-Agile Project Tracker Linear the Latest to Take on Jira Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

Tom Moor

Head of Engineering

LinearLinear
Andrew Milich
CursorCursor
The Official SaaStr Podcast: SaaS | Founders | Investors/Sep 5, 2025/31m

SaaStr 818: Anthropic, Cursor, Fal & Bessemer: The Realities of Scaling AI

SaaStr 818: Anthropic, Cursor, Fal & Bessemer: The Realities of Scaling AI Join Talia Goldberg (Bessemer Venture Partners), Kelly Loftus (Anthropic), Jacob Jackson (Cursor), and Gorkem Yurtseven (FaL - Feautres and Labels) as they discuss the evolving landscape of AI, business models, metrics, and the future of generative media. 00:00 - Introduction & Panelist Bios 01:19 - How AI Companies Are Changing Metrics 02:20 - The New Economics: Margins, Growth, and Pricing 04:41 - Usage-Based Models & The Cost to Serve 06:31 - The Impact of Expensive Models on Margins 08:00 - Go-To-Market Team Structures at Anthropic 09:36 - Scaling Sales Teams & The Quota Debate 12:01 - Hiring and Team Building Tactics 14:47 - How AI Is Used Internally at These Companies 17:47 - Key Decisions & Pivots in Company Journeys 20:07 - Collaboration vs. Competition: Cursor & Anthropic 22:23 - Measuring Productivity Gains from AI 24:26 - The Metrics That Matter Most 27:16 - Final Thoughts & Audience Q&A --------------------- Fin is the #1 AI Agent for resolving complex queries like refunds, transaction disputes, and technical troubleshooting—all with speed and reliability. See how Fin can deliver the highest resolution rates and highest-quality customer experience at fin.ai/saastr. --------------------- If you're serious about B2B and AI, you need to be in London this December 2nd and 3rd. SaaStr AI London is bringing together more than 2,000 leaders and founders for two days of practical advice on scaling into the new year. We'll have speakers flying in from OpenAI, Wiz, Clay, Intercom, and all your favorite SaaS companies, including yours truly with Harry Stebbings for a live 20VC podcast. It'll be fun, and it's all in the heart of London. Don't miss out: get your tickets with my exclusive discount by going to podcast.saastrlondon.com --------------------- Hey everybody, the biggest B2B + AI event of the year will be back - SaaStr AI in the SF Bay Area, aka the SaaStr Annual, will be back in May 2026. With 68% VP-level and above, 36% CEOs and founders and a growing 25% AI-first professional, this is the very best of the best S-tier attendees and decision makers that come to SaaStr each year. But here's the reality, folks: the longer you wait, the higher ticket prices can get. Early bird tickets are available now, but once they're gone, you'll pay hundreds more so don't wait. Lock in your spot today by going to podcast podcast.saastrannual.com to get my exclusive discount SaaStr AI SF 2026. We'll see you there.

Talia Goldberg

Partner

Bessemer Venture PartnersBessemer Venture Partners
Kelly Loftus

GTM Leader

AnthropicAnthropic
Jacob Jackson
CursorCursor
Gorkem Yurtseven

Founder

fal.aifal.aifalfal
Decoder with Nilay Patel/Aug 4, 2025/56m

Why tech is racing to adopt AI coding

This is Casey Newton, founder and editor of the Platformer newsletter and cohost of the Hard Fork podcast. I’ll be guest hosting the next few episodes of Decoder while Nilay is out on parental leave. For the next three weeks, I’ll be talking to leaders in the productivity space about what they’re building, and how they can help us get things done.  My guest today: Michael Truell, the CEO of Anysphere, the maker of automated programming platform Cursor AI. I sat down with Michael to talk about his product and how it works, why coding with AI has seen such incredible adoption, and what the future of automated programming really looks like.  Read the full transcript on The Verge. Links:  Anysphere, hailed as fastest growing startup ever, raises $900 Million | Bloomberg AI coding assistant Cursor draws a million users without even trying | Bloomberg Anthropic rehires AI leaders from Anysphere | The Information Cursor apologizes for unclear pricing changes that upset users | TechCrunch OpenAI looked at buying Cursor creator before turning to rival Windsurf | CNBC Interview with Anysphere CEO Michael Truell about coding with AI | Stratechery Credits: Decoder is a production of The Verge and part of the Vox Media Podcast Network. Our producers are Kate Cox and Nick Statt. Our editor is Ursa Wright.  The Decoder music is by Breakmaster Cylinder. Learn more about your ad choices. Visit podcastchoices.com/adchoices

Michael Truell

Co-Founder

CursorCursorAnysphereAnysphere
Lenny's Podcast: Product | Career | Growth/May 1, 2025/1h 11m

The rise of Cursor: The $300M ARR AI tool that engineers can’t stop using | Michael Truell (co-founder and CEO)

Michael Truell is the co-founder and CEO of Anysphere, the company behind Cursor—the fastest-growing AI code editor in the world, reaching $300 million in annual recurring revenue just two years after its launch. In this conversation, Michael shares his vision for the future, lessons learned, and advice for preparing for the fast-approaching AI future. What you’ll learn: • Cursor's early pivot from automating CAD to automating code • Michael’s vision for “what comes after code” and how programming will evolve • Why Cursor built their own custom AI models despite not starting there • Key lessons from Cursor’s rapid growth • Why “taste” and logic design will become more valuable engineering skills than technical coding ability • Why the market for AI coding tools is much larger than people realize—and why there will likely be one dominant winner • Michael’s advice for engineers and product teams preparing for the AI future — Brought to you by: Eppo—Run reliable, impactful experiments Vanta—Automate compliance. Simplify security OneSchema—Import CSV data 10x faster — Where to find Michael Truell: • X: https://x.com/mntruell • LinkedIn: https://www.linkedin.com/in/michael-t-5b1bbb122/ • Website: https://mntruell.com/ — In this episode, we cover: (00:00) Introduction to Michael Truell and Cursor (04:20) What comes after code (08:32) The importance of taste (12:39) Cursor’s origin story (18:31) Why they chose to build an IDE (22:39) Will everyone become engineering managers? (24:31) How they decided it was time to ship (26:45) Reflecting on Cursor's success (32:03) Counterintuitive lessons on building AI products (34:02) Inside Cursor's stack (38:42) Defensibility and market dynamics in AI (46:13) Tips for using Cursor (51:25) Hiring and building a strong team (59:10) Staying focused amid rapid AI advancements (01:02:31) Final thoughts and advice for aspiring AI innovators — Referenced: • Cursor: https://www.cursor.com/ • Microsoft Copilot: https://copilot.microsoft.com/ • Scaling laws for neural language models: https://openai.com/index/scaling-laws-for-neural-language-models/ • MIT: https://www.mit.edu/ • Telegram: https://telegram.org/ • Signal: https://signal.org/ • WhatsApp: https://www.whatsapp.com/ • Devin: https://devin.ai/ • Visual Studio Code: https://code.visualstudio.com/ • Chromium: https://chromium.googlesource.com/chromium/src/base/ • Exploring ChatGPT (GPT) Wrappers—What They Are and How They Work: https://learnprompting.org/blog/gpt_wrappers • OpenAI’s CPO on how AI changes must-have skills, moats, coding, startup playbooks, more | Kevin Weil (CPO at OpenAI, ex-Instagram, Twitter): https://www.lennysnewsletter.com/p/kevin-weil-open-ai • Behind the founder: Marc Benioff: https://www.lennysnewsletter.com/p/behind-the-founder-marc-benioff • DALL-E 3: https://openai.com/index/dall-e-3/ • Stable Diffusion 3: https://stability.ai/news/stable-diffusion-3 — Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email podcast@lennyrachitsky.com. — Lenny may be an investor in the companies discussed. This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.lennysnewsletter.com/subscribe

Michael Truell

Co-Founder

CursorCursorAnysphereAnysphere
Gradient Dissent: Conversations on AI/Apr 29, 2025/50m

Inside Cursor: The future of AI coding with Co-founder Sualeh Asif

In this episode of Gradient Dissent, host Lukas Biewald talks with Sualeh Asif, the CPO and co-founder of Cursor, one of the fastest-growing and most loved AI-powered coding platforms. Sualeh shares the story behind Cursor’s creation, the technical and design decisions that set it apart, and how AI models are changing the way we build software. They dive deep into infrastructure challenges, the importance of speed and user experience, and how emerging trends in agents and reasoning models are reshaping the developer workflow. Sualeh also discusses scaling AI inference to support hundreds of millions of requests per day, building trust through product quality, and his vision for how programming will evolve in the next few years. ⏳Timestamps: 00:00 How Cursor got started and why it took off 04:50 Switching from Vim to VS Code and the rise of CoPilot 08:10 Why Cursor won among competitors: product philosophy and execution 10:30 How user data and feedback loops drive Cursor’s improvements 12:20 Iterating on AI agents: what made Cursor hold back and wait 13:30 Competitive coding background: advantage or challenge? 16:30 Making coding fun again: latency, flow, and model choices 19:10 Building Cursor’s infrastructure: from GPUs to indexing billions of files 26:00 How Cursor prioritizes compute allocation for indexing 30:00 Running massive ML infrastructure: surprises and scaling lessons 34:50 Why Cursor chose DeepSeek models early 36:00 Where AI agents are heading next 40:07 Debugging and evaluating complex AI agents 42:00 How coding workflows will change over the next 2–3 years 46:20 Dream future projects: AI for reading codebases and papers 🎙 Get our podcasts on these platforms: Apple Podcasts: https://wandb.me/apple-podcasts Spotify: https://wandb.me/spotify YouTube: https://wandb.me/youtube Follow Weights & Biases: https://x.com/weights_biases https://www.linkedin.com/company/wandb

Sualeh Asif

Co-Founder

CursorCursor
Lex Fridman Podcast/Oct 6, 2024/2h 38m

#447 – Cursor Team: Future of Programming with AI

Aman Sanger, Arvid Lunnemark, Michael Truell, and Sualeh Asif are creators of Cursor, a popular code editor that specializes in AI-assisted programming. Thank you for listening ❤ Check out our sponsors: https://lexfridman.com/sponsors/ep447-sc See below for timestamps, transcript, and to give feedback, submit questions, contact Lex, etc. Transcript: https://lexfridman.com/cursor-team-transcript CONTACT LEX: Feedback – give feedback to Lex: https://lexfridman.com/survey AMA – submit questions, videos or call-in: https://lexfridman.com/ama Hiring – join our team: https://lexfridman.com/hiring Other – other ways to get in touch: https://lexfridman.com/contact EPISODE LINKS: Cursor Website: https://cursor.com Cursor on X: https://x.com/cursor_ai Anysphere Website: https://anysphere.inc/ Aman’s X: https://x.com/amanrsanger Aman’s Website: https://amansanger.com/ Arvid’s X: https://x.com/ArVID220u Arvid’s Website: https://arvid.xyz/ Michael’s Website: https://mntruell.com/ Michael’s LinkedIn: https://bit.ly/3zIDkPN Sualeh’s X: https://x.com/sualehasif996 Sualeh’s Website: https://sualehasif.me/ SPONSORS: To support this podcast, check out our sponsors & get discounts: Encord: AI tooling for annotation & data management. Go to https://encord.com/lex MasterClass: Online classes from world-class experts. Go to https://masterclass.com/lexpod Shopify: Sell stuff online. Go to https://shopify.com/lex NetSuite: Business management software. Go to http://netsuite.com/lex AG1: All-in-one daily nutrition drinks. Go to https://drinkag1.com/lex OUTLINE: (00:00) – Introduction (09:25) – Code editor basics (11:35) – GitHub Copilot (18:53) – Cursor (25:20) – Cursor Tab (31:35) – Code diff (39:46) – ML details (45:20) – GPT vs Claude (51:54) – Prompt engineering (59:20) – AI agents (1:13:18) – Running code in background (1:17:57) – Debugging (1:23:25) – Dangerous code (1:34:35) – Branching file systems (1:37:47) – Scaling challenges (1:51:58) – Context (1:57:05) – OpenAI o1 (2:08:27) – Synthetic data (2:12:14) – RLHF vs RLAIF (2:14:01) – Fields Medal for AI (2:16:43) – Scaling laws (2:25:32) – The future of programming PODCAST LINKS: – Podcast Website: https://lexfridman.com/podcast – Apple Podcasts: https://apple.co/2lwqZIr – Spotify: https://spoti.fi/2nEwCF8 – RSS: https://lexfridman.com/feed/podcast/ – Podcast Playlist: https://www.youtube.com/playlist?list=PLrAXtmErZgOdP_8GztsuKi9nrraNbKKp4 – Clips Channel: https://www.youtube.com/lexclips

Aman Sanger

Co-founder, Cursor (Anysphere)

AnysphereAnysphereCursorCursor
Arvid Lunnemark

Co-Founder

CursorCursor
Michael Truell

Co-Founder

AnysphereAnysphereCursorCursor
Sualeh Asif

Co-Founder

CursorCursor
Latent Space: The AI Engineer Podcast/Jun 10, 2024/4h 29m

ICLR 2024 — Best Papers & Talks (Benchmarks, Reasoning & Agents) — ft. Graham Neubig, Aman Sanger, Moritz Hardt)

Our second wave of speakers for AI Engineer World’s Fair were announced! The conference sold out of Platinum/Gold/Silver sponsors and Early Bird tickets! See our Microsoft episode for more info and buy now with code LATENTSPACE. This episode is straightforwardly a part 2 to our ICLR 2024 Part 1 episode, so without further ado, we’ll just get right on with it! Timestamps [00:03:43] Section A: Code Edits and Sandboxes, OpenDevin, and Academia vs Industry — ft. Graham Neubig and Aman Sanger * [00:07:44] WebArena * [00:18:45] Sotopia * [00:24:00] Performance Improving Code Edits * [00:29:39] OpenDevin * [00:47:40] Industry and Academia [01:05:29] Section B: Benchmarks * [01:05:52] SWEBench * [01:17:05] SWEBench/SWEAgent Interview * [01:27:40] Dataset Contamination Detection * [01:39:20] GAIA Benchmark * [01:49:18] Moritz Hart - Science of Benchmarks [02:36:32] Section C: Reasoning and Post-Training * [02:37:41] Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection * [02:51:00] Let’s Verify Step By Step * [02:57:04] Noam Brown * [03:07:43] Lilian Weng - Towards Safe AGI * [03:36:56] A Real-World WebAgent with Planning, Long Context Understanding, and Program Synthesis * [03:48:43] MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework [04:00:51] Bonus: Notable Related Papers on LLM Capabilities Section A: Code Edits and Sandboxes, OpenDevin, and Academia vs Industry — ft. Graham Neubig and Aman Sanger * Guests * Graham Neubig * Aman Sanger - Previous guest and NeurIPS friend of the pod! * WebArena * * Sotopia (spotlight paper, website) * * Learning Performance-Improving Code Edits * OpenDevin * Junyang Opendevin * Morph Labs, Jesse Han * SWE-Bench * SWE-Agent * Aman tweet on swebench * LiteLLM * Livecodebench * the role of code in reasoning * Language Models of Code are Few-Shot Commonsense Learners * Industry vs academia * the matryoshka embeddings incident * other directions * Unlimiformer Section A timestamps * [00:00:00] Introduction to Guests and the Impromptu Nature of the Podcast * [00:00:45] Graham's Experience in Japan and Transition into Teaching NLP * [00:01:25] Discussion on What Constitutes a Good Experience for Students in NLP Courses * [00:02:22] The Relevance and Teaching of Older NLP Techniques Like Ngram Language Models * [00:03:38] Speculative Decoding and the Comeback of Ngram Models * [00:04:16] Introduction to WebArena and Zotopia Projects * [00:05:19] Deep Dive into the WebArena Project and Benchmarking * [00:08:17] Performance Improvements in WebArena Using GPT-4 * [00:09:39] Human Performance on WebArena Tasks and Challenges in Evaluation * [00:11:04] Follow-up Work from WebArena and Focus on Web Browsing as a Benchmark * [00:12:11] Direct Interaction vs. Using APIs in Web-Based Tasks * [00:13:29] Challenges in Base Models for WebArena and the Potential of Visual Models * [00:15:33] Introduction to Zootopia and Exploring Social Interactions with Language Models * [00:16:29] Different Types of Social Situations Modeled in Zootopia * [00:17:34] Evaluation of Language Models in Social Simulations * [00:20:41] Introduction to Performance-Improving Code Edits Project * [00:26:28] Discussion on DevIn and the Future of Coding Agents * [00:32:01] Planning in Coding Agents and the Development of OpenDevon * [00:38:34] The Changing Role of Academia in the Context of Large Language Models * [00:44:44] The Changing Nature of Industry and Academia Collaboration * [00:54:07] Update on NLP Course Syllabus and Teaching about Large Language Models * [01:00:40] Call to Action: Contributions to OpenDevon and Open Source AI Projects * [01:01:56] Hiring at Cursor for Roles in Code Generation and Assistive Coding * [01:02:12] Promotion of the AI Engineer Conference Section B: Benchmarks * Carlos Jimenez & John Yang (Princeton) et al: SWE-bench: Can Language Models Resolve Real-world Github Issues? (ICLR Oral, Paper, website) * “We introduce SWE-bench, an evaluation framework consisting of 2,294 software engineering problems drawn from real GitHub issues and corresponding pull requests across 12 popular Python repositories. Given a codebase along with a description of an issue to be resolved, a language model is tasked with editing the codebase to address the issue. Resolving issues in SWE-bench frequently requires understanding and coordinating changes across multiple functions, classes, and even files simultaneously, calling for models to interact with execution environments, process extremely long contexts and perform complex reasoning that goes far beyond traditional code generation tasks. Our evaluations show that both state-of-the-art proprietary models and our fine-tuned model SWE-Llama can resolve only the simplest issues. The best-performing model, Claude 2, is able to solve a mere 1.96% of the issues. Advances on SWE-bench represent steps towards LMs that are more practical, intelligent, and autonomous.” * Yonatan Oren et al (Stanford): Proving Test Set Contamination in Black-Box Language Models (ICLR Oral, paper, aman tweet on swebench contamination) * “We show that it is possible to provide provable guarantees of test set contamination in language models without access to pretraining data or model weights. Our approach leverages the fact that when there is no data contamination, all orderings of an exchangeable benchmark should be equally likely. In contrast, the tendency for language models to memorize example order means that a contaminated language model will find certain canonical orderings to be much more likely than others. Our test flags potential contamination whenever the likelihood of a canonically ordered benchmark dataset is significantly higher than the likelihood after shuffling the examples. * We demonstrate that our procedure is sensitive enough to reliably prove test set contamination in challenging situations, including models as small as 1.4 billion parameters, on small test sets of only 1000 examples, and datasets that appear only a few times in the pretraining corpus.” * Outstanding Paper mention: “A simple yet elegant method to test whether a supervised-learning dataset has been included in LLM training.” * Thomas Scialom (Meta AI-FAIR w/ Yann LeCun): GAIA: A Benchmark for General AI Assistants (paper) * “We introduce GAIA, a benchmark for General AI Assistants that, if solved, would represent a milestone in AI research. GAIA proposes real-world questions that require a set of fundamental abilities such as reasoning, multi-modality handling, web browsing, and generally tool-use proficiency. * GAIA questions are conceptually simple for humans yet challenging for most advanced AIs: we show that human respondents obtain 92% vs. 15% for GPT-4 equipped with plugins. * GAIA's philosophy departs from the current trend in AI benchmarks suggesting to target tasks that are ever more difficult for humans. We posit that the advent of Artificial General Intelligence (AGI) hinges on a system's capability to exhibit similar robustness as the average human does on such questions. Using GAIA's methodology, we devise 466 questions and their answer. * * Mortiz Hardt (Max Planck Institute): The emerging science of benchmarks (ICLR stream) * “Benchmarks are the keystone that hold the machine learning community together. Growing as a research paradigm since the 1980s, there’s much we’ve done with them, but little we know about them. In this talk, I will trace the rudiments of an emerging science of benchmarks through selected empirical and theoretical observations. Specifically, we’ll discuss the role of annotator errors, external validity of model rankings, and the promise of multi-task benchmarks. The results in each case challenge conventional wisdom and underscore the benefits of developing a science of benchmarks.” Section C: Reasoning and Post-Training * Akari Asai (UW) et al: Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection (ICLR oral, website) * (Bad RAG implementations) indiscriminately retrieving and incorporating a fixed number of retrieved passages, regardless of whether retrieval is necessary, or passages are relevant, diminishes LM versatility or can lead to unhelpful response generation. * We introduce a new framework called Self-Reflective Retrieval-Augmented Generation (Self-RAG) that enhances an LM's quality and factuality through retrieval and self-reflection. * Our framework trains a single arbitrary LM that adaptively retrieves passages on-demand, and generates and reflects on retrieved passages and its generations using special tokens, called reflection tokens. Generating reflection tokens makes the LM controllable during the inference phase, enabling it to tailor its behavior to diverse task requirements. * Self-RAG (7B and 13B parameters) outperforms ChatGPT and retrieval-augmented Llama2-chat on Open-domain QA, reasoning, and fact verification tasks, and it shows significant gains in improving factuality and citation accuracy for long-form generations relative to these models. * Hunter Lightman (OpenAI): Let’s Verify Step By Step (paper) * “Even state-of-the-art models still regularly produce logical mistakes. To train more reliable models, we can turn either to outcome supervision, which provides feedback for a final result, or process supervision, which provides feedback for each intermediate reasoning step. * We conduct our own investigation, finding that process supervision significantly outperforms outcome supervision for training models to solve problems from the challenging MATH dataset. Our process-supervised model solves 78% of problems from a representative subset of the MATH test set. Additionally, we show that active learning significantly improves the efficacy of process supervision. * To support related research, we also release PRM800K, the complete dataset of 800,000 step-level human feedback labels used to train our best reward model. * * Noam Brown - workshop on Generative Models for Decision Making * Solving Quantitative Reasoning Problems with Language Models (Minerva paper) * Describes some charts taken directly from the Let’s Verify Step By Step paper listed/screenshotted above . * Lilian Weng (OpenAI) - Towards Safe AGI (ICLR talk) * OpenAI Model Spec * OpenAI Instruction Hierarchy: The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions Section D: Agent Systems * Izzeddin Gur (Google DeepMind): A Real-World WebAgent with Planning, Long Context Understanding, and Program Synthesis (ICLR oral, paper) * [Agent] performance on real-world websites has still suffered from (1) open domainness, (2) limited context length, and (3) lack of inductive bias on HTML. * We introduce WebAgent, an LLM-driven agent that learns from self-experience to complete tasks on real websites following natural language instructions. * WebAgent plans ahead by decomposing instructions into canonical sub-instructions, summarizes long HTML documents into task-relevant snippets, and acts on websites via Python programs generated from those. * We design WebAgent with Flan-U-PaLM, for grounded code generation, and HTML-T5, new pre-trained LLMs for long HTML documents using local and global attention mechanisms and a mixture of long-span denoising objectives, for planning and summarization. * We empirically demonstrate that our modular recipe improves the success on real websites by over 50%, and that HTML-T5 is the best model to solve various HTML understanding tasks; achieving 18.7% higher success rate than the prior method on MiniWoB web automation benchmark, and SoTA performance on Mind2Web, an offline task planning evaluation. * Sirui Hong (DeepWisdom): MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework (ICLR Oral, Paper) * We introduce MetaGPT, an innovative meta-programming framework incorporating efficient human workflows into LLM-based multi-agent collaborations. MetaGPT encodes Standardized Operating Procedures (SOPs) into prompt sequences for more streamlined workflows, thus allowing agents with human-like domain expertise to verify intermediate results and reduce errors. MetaGPT utilizes an assembly line paradigm to assign diverse roles to various agents, efficiently breaking down complex tasks into subtasks involving many agents working together. Bonus: Notable Related Papers on LLM Capabilities This includes a bunch of papers we wanted to feature above but could not. * Lukas Berglund (Vanderbilt) et al: The Reversal Curse: LLMs trained on “A is B” fail to learn “B is A” (ICLR poster, paper, Github) * We expose a surprising failure of generalization in auto-regressive large language models (LLMs). If a model is trained on a sentence of the form ''A is B'', it will not automatically generalize to the reverse direction ''B is A''. This is the Reversal Curse. * The Reversal Curse is robust across model sizes and model families and is not alleviated by data augmentation. We also evaluate ChatGPT (GPT-3.5 and GPT-4) on questions about real-world celebrities, such as ''Who is Tom Cruise's mother? [A: Mary Lee Pfeiffer]'' and the reverse ''Who is Mary Lee Pfeiffer's son?''. GPT-4 correctly answers questions like the former 79\% of the time, compared to 33\% for the latter. * * Omar Khattab (Stanford): DSPy: Compiling Declarative Language Model Calls into State-of-the-Art Pipelines (ICLR Spotlight Poster, GitHub) * presented by Krista Opsahl-Ong * “Existing LM pipelines are typically implemented using hard-coded “prompt templates”, i.e. lengthy strings discovered via trial and error. Toward a more systematic approach for developing and optimizing LM pipelines, we introduce DSPy, a programming model that abstracts LM pipelines as text transformation graphs, or imperative computational graphs where LMs are invoked through declarative modules. * DSPy modules are parameterized, meaning they can learn how to apply compositions of prompting, finetuning, augmentation, and reasoning techniques. * We design a compiler that will optimize any DSPy pipeline to maximize a given metric, by creating and collecting demonstrations. * We conduct two case studies, showing that succinct DSPy programs can express and optimize pipelines that reason about math word problems, tackle multi-hop retrieval, answer complex questions, and control agent loops. * Within minutes of compiling, DSPy can automatically produce pipelines that outperform out-of-the-box few-shot prompting as well as expert-created demonstrations for GPT-3.5 and Llama2-13b-chat. On top of that, DSPy programs compiled for relatively small LMs like 770M parameter T5 and Llama2-13b-chat are competitive with many approaches that rely on large and proprietary LMs like GPT-3.5 and on expert-written prompt chains. * * MuSR: Testing the Limits of Chain-of-thought with Multistep Soft Reasoning * Scaling Laws for Associative Memories * DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models * Efficient Streaming Language Models with Attention Sinks This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.latent.space/subscribe

Graham Neubig

Associate Professor, Carnegie Mellon University

Carnegie Mellon UniversityCarnegie Mellon UniversityAll Hands AIAll Hands AI
Aman Sanger

Co-founder, Cursor (Anysphere)

AnysphereAnysphereCursorCursor
Moritz Hardt

Director, Max Planck Institute for Intelligent Systems

Max Planck Institute for Intelligent SystemsMax Planck Institute for Intelligent Systems

Sign in to see 2 more episodes

Free, one click with Google. Browse every guest and company across the catalog.