Jun 28, 2026 · 1h 9m · lennys-podcast
Why OpenAI is merging Codex and ChatGPT and the future of knowledge work | Andrew Ambrosino
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
Lenny Rachitsky interviews OpenAI's Andrew Ambrosino to discuss how AI inverts software product development, why Codex is transforming into a universal desktop platform for knowledge work, and how product teams must cultivate taste and role fluidity.
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 25.2% of the talking time here. How this is scored →
Lenny as informed peer 4.4 Guest teaching 5.3 Guest disagreement 1.1 Lenny pushing back 0.8
speaking balance: gold is Lenny, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 25:09 Andrew rejects eliminating PM roles
Andrew strongly pushes back against tech industry hype, calling the idea of companies getting rid of PMs a 'terrible idea' and dismissing the notion that roles are just people vibing without real craft skills.
Hardest push from Lenny ▶ 8:28 Lenny challenges immediate prototyping with the primal markLenny challenges the trend of skipping PRDs to jump straight into prototypes, introducing the 'primal mark' concept to argue that early visual artifacts anchor teams to narrow solutions.
Biggest teaching moment ▶ 12:41 Andrew on why AI models struggle with design feedback loopsAndrew breaks down the foundational differences between code generation and visual design, explaining that design lacks objective grading loops like compilers and requires cultural novelty and deep semantic code abstractions.
Lenny holds their own ▶ 46:53 Lenny demonstrates building an agentic spam filter workflowLenny demonstrates his technical fluency and hands-on builder credibility by detailing how he built an automated email classifier in Codex and used autonomous computer use to configure GCP Pub/Sub triggers.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Lenny as informed peer | Guest teaching | Guest disagreement | Lenny pushing back | Why |
|---|---|---|---|---|---|---|
| The Inversion of the Product Development Process | 4 | 5 | 1 | 1 | Lenny opens by asking how AI is reshaping product teams. Andrew explains the inversion of the development process where implementation is cheap and curation and taste become the expensive bottleneck. | |
| Sponsor Message: WorkOS | 4 | 5 | 2 | 1 | Following the sponsor read, Lenny asks about the shift from written PRDs to dozens of prototypes. Andrew pushes back on the claim that 'PRDs are dead', explaining that implementation abundance requires choosing the right medium for the specific clarity needed. | |
| The 'Primal Mark' and Decoupling Visual Polish from Product Readiness | 6 | 5 | 1 | 1 | Lenny introduces the concept of the 'primal mark' to explain why jumping straight to prototypes can anchor teams prematurely. Andrew agrees and explains how visual polish is now divorced from actual product de-risking, redefining what taste means in practice. | |
| Why AI Models Struggle with Design and Abstraction | 4 | 6 | 1 | 1 | Lenny asks why frontier models remain poor at visual design. Andrew provides a detailed technical breakdown, citing the difficulty of automated grading compared to code compilation, AI labs' focus on research-accelerating tasks, and the need for novelty and semantic code abstractions. | |
| Human Novelty and New Interaction Paradigms | 5 | 5 | 2 | 2 | Lenny brings up a previous guest's thesis that traditional design processes are dead in fast-moving AI cycles. Andrew offers a nuanced critique, agreeing that rigid case study rituals are obsolete while emphasizing that foundational problem-space exploration remains critical. | |
| Role Fluidity and Cross-Functional Overlap on the Codex Team | 4 | 5 | 1 | 1 | Lenny inquires into the internal organization and role boundaries of the Codex team. Andrew explains that team members operate with extensive overlap, defining roles not by rigid fences but by the mathematical average of their day-to-day contributions. | |
| The Future of Specialized Roles vs. General 'Builders' | 5 | 6 | 3 | 2 | Lenny asks if functional boundaries are collapsing into generic 'builders'. Andrew forcefully rejects the extreme idea of eliminating product managers, emphasizing that product and engineering remain distinct disciplines with specialized best practices. | |
| Team Structure and High-Agency Hiring at OpenAI | 4 | 5 | 1 | 1 | Lenny asks about hiring profiles and team size on Codex. Andrew outlines their 'zone defense' model where high-agency product leaders create broad organizational coverage rather than clustering on single features. | |
| Product Planning and Building for Future Model Capabilities | 4 | 6 | 1 | 1 | Lenny asks how roadmaps are formulated when model capabilities change rapidly. Andrew explains that high precision on long-term roadmaps is false precision, noting that Codex succeeded in February with the exact same architecture that would have failed in November. | |
| Sponsor Message: Mercury | 5 | 6 | 2 | 1 | Lenny asks about balancing extreme ambition with building ahead of model readiness. Andrew reviews past iterations like Operator and Atlas, warning against becoming too 'AGI pilled' when users require pragmatic stepping stones. | |
| Beyond Coding Loops: Autonomous Development and Code Complexity | 4 | 6 | 2 | 1 | Lenny asks about the modern frontier of AI-native engineering loops. Andrew playfully dismisses basic loops as outdated and points out key technical bottlenecks, such as models increasing codebase complexity and failing to delete code. | |
| How Andrew Uses Codex to Manage Product and Daily Briefs | 4 | 5 | 0 | 0 | Lenny asks how Andrew uses Codex to run his own workflows. Andrew walks through using the app for automated Slack morning briefs, triage across 3,000 channels, and continuous coaching of agent instructions. | |
| Connectors, Computer Use, and Productizing Workflows | 6 | 5 | 0 | 0 | Lenny demonstrates his own hands-on expertise by detailing how he built an automated email spam classifier in Codex and let computer use configure GCP Pub/Sub. Andrew details the product boundary between API connectors, in-app browsers, and raw GUI computer use. | |
| Integrating Browsers, Desktop Tools, and SaaS Apps | 5 | 5 | 1 | 1 | Lenny references Dan Shipper's prediction about running SaaS apps entirely inside Codex. Andrew outlines the technical complexities of embedding browser stacks, multi-tab enterprise security, and resolving keyboard shortcut collisions. | |
| Expanding Codex Beyond Engineers to General Knowledge Work | 4 | 6 | 0 | 0 | Lenny asks about the overarching vision for Codex. Andrew explains how internal OpenAI employees in finance, legal, and marketing adopted Codex despite its developer-centric UI, leading to its expansion into general knowledge work. | |
| The Vision for Codex as a Desktop Home Base and the Premiere Pro Story | 5 | 6 | 1 | 1 | Lenny discusses the convergence of ChatGPT and Codex into a unified desktop hub. Andrew shares an anecdote where an OpenAI videographer used Codex to autonomously write its own Premiere Pro plugin to execute video edits. | |
| Fail Corner: Startups, Slogs, and Internal OpenAI Feedback Loops | 4 | 5 | 0 | 0 | Lenny introduces Fail Corner to discuss past setbacks. Andrew reflects on selling his previous startup for parts and describes intense internal OpenAI feedback loops with 2,000-message Slack threads tearing down product proposals. | |
| Lightning Round: Books, Pop Culture, and the Triad of PM/Design/Eng | 4 | 4 | 1 | 1 | Lenny runs through lightning round questions covering children's books, television habits, and role hierarchies. Andrew playfully dodges declaring a toughest role among PM, design, and engineering while embracing fluid collaboration. | |
| Podcast Outro | 3 | 4 | 0 | 0 | During the outro and post-recording chat, the room producer discusses using Codex for editing video pauses. Andrew and Lenny discuss the necessity of adaptability and avoiding rigid attachment to specific toolchains. |
Statements from this episode (27)
Ambrosino: AI makes taste and curation the expensive bottlenecks in product development
“The implementation is actually not the expensive part anymore. It's dare I say taste. But it's the curation process. It's like of those 90 attempts, like what's good about these, what should we fold into other aspects of this, right? How should we frame this? …”
Ambrosino: PRDs are not dead despite claims from product leaders
“Like you've seen many product leaders say PRDs are dead. And I actually don't believe this at all.”
Ambrosino: Use documents for product clarity and prototypes for interaction testing
“If implementation is abundant. Then it's really important to pick the right format for the point you're trying to make. If that point is product clarity around a vague area, then it might actually be a document. If what you're trying to do is get something in …”
Ambrosino: High-fidelity prototypes no longer signal validated assumptions in AI era
“The medium implied it had baked in a lot of signal around where in the process something was. So if you're seeing something that feels like the app in production, That means that it's late in the process, that assumptions have been de-risped, that, you know, d…”
Ambrosino: The core of product taste is deciding what to build
“But there's like the, like, what should this be? Like, if we can build anything, like what's, what's the goal here and how do we get there? That I think is like actually the real taste question here.”
Ambrosino: AI labs prioritized coding models because code directly accelerates AI research
“I also think that the labs historically invest in making their models good at things that accelerates AI research. And that in the era, the early era of coding models is very clear that the model being able to write correct code. Would accelerate research. In …”
Ambrosino: Design requires novelty whereas software engineering relies on known patterns
“There's an amount of like novelty that is more important in design than it actually is in software engineering. Like software engineering, you almost, you must want it to over index unknown patterns, right? Whereas design, it's like, no, there's an element of …”
Ambrosino: OpenAI's Codex team started in November and now uses the app for everything
“We started the Codex app in November and we weren't using it full time. Now we use it for everything.”
Ambrosino: Traditional design assumes implementation is too expensive to rebuild
“That process is sort of predicated on the assumption that implementation is expensive and that you can really only afford to build once. And so you need to fully like exhaustively go through the problem space and the solution space before implementing.”
Ambrosino: OpenAI maintains a simplified 'baby codex' codebase to prototype faster
“Like we have baby codex, right? A dramatically simplified code base that approximates all of the interactions of the production app. And therefore is a lot quicker to vibe code over, right?”
Ambrosino: OpenAI sees role overlap rather than total existential role collapse
“There's been a lot written about role collapse, existential role collapse. There are no roles anymore. We haven't seen that. We've, we have seen more role collapse in the Codex org than I think other parts of the company and other parts of the economy.”
Ambrosino: Codex Team Deliberately Dogfoods Suboptimal Workflows to Fix the App
“There is a desire among all of us to try to do as much as possible in the app, even when it's not the best tool so that it can become the best tool. And so a lot of design we all work on by using the app and say, okay, what's broken about this? This is a whole…”
Ambrosino: Eliminating the product manager role is a terrible idea
“I I've heard a lot of companies be like, we're getting rid of the product role, which I think is by the way, a terrible idea.”
Ambrosino: OpenAI Codex core team has double-digit engineers and few PMs
“So we got a team double digits of engineers. Probably half that on the design side. You know, a few product people”
Ambrosino: OpenAI runs large team sizes composed mostly of ICs
“At OpenAI, we let teams get very large. So we haven't said, Hey, there's no management, but like the teams are quite large, right? It's mostly ICs.”
Ambrosino: Two PMs working too closely is often a bad sign
“If two product people are working too closely, that's often not a good signal.”
Ambrosino: Every IC is now also a manager of AI agents
“It's not that management is going away. It's not that everyone's an IC, but like everyone's kind of both now, right? If you're an IC, you're not typing code out character by character, right? Like you are managing something. You're managing agents. You're mana…”
Ambrosino: Any precision in a 9-month AI roadmap is false precision
“The basic just is the shorter term something is the more detail it needs. And then it's not that we don't plan for nine months out. It's that that just has to stay very hazy because any amount of precision that you add to a nine month plan right now is false p…”
Ambrosino: AI teams must prototype future features and wait for model breakthroughs
“It basically had to be like, let's list out all of the things that we think we are interested in doing for the next year or two. Let's prototype all of them, decide which things are ready now, and then just let the others sit and bake. And then every time ther…”
Ambrosino: OpenAI Codex would have failed if released three months earlier
“I like, I am very confident that the codex app that we released in February, if that had been ready. In November, it would have absolutely failed in the market. And the only difference was the models between November and February.”
Ambrosino: 100% of current OpenAI product code is written by AI
“Cause if you're using the goalposts from last year, it's like, well, a hundred percent of our product right now is AI written code.”
Ambrosino: Current AI models increase codebase complexity rather than simplifying it
“One thing that I think all models suffer with right now is just, they usually increase complexity. If research is listening at any company, please make the models better at deleting code.”
Ambrosino: OpenAI Codex added computer use and artifact creation in May
“There was a release or a series of releases in May-ish that introduced the in-app browser computer use and artifact creation to the codex app.”
Ambrosino: AI apps must build native memory instead of user mind palaces
“And I think memory is sort of in the shape where we've had a lot of people and a lot of people at other companies too are like, well, I set up an obsidian base or a notion area and I tell it how to basically build my mind palace and how to put, it's like, eh, …”
Ambrosino: OpenAI non-engineers adopted Codex despite its developer-hostile interface
“We have people from marketing, from comms, from finance, from legal, from basically every discipline who are using this codex app, even though it is actively hostile to these people, right?”
Ambrosino: OpenAI Codex independently wrote a Premiere Pro extension to edit videos
“Codex is not a video editor per se, right? It doesn't have any of that UI in it. But it was able to understand that he used Premiere Pro. It could do some edits by editing the files that were backing what was on screen in Premiere Pro, but it couldn't do every…”
Ambrosino: OpenAI integrates Codex with existing tools instead of replacing them
“We're trying to do two things at once with Codex and with now with ChatGPT. One is how can we seamlessly interact with these tools that you're already using and say, like, we don't need to build a better video editor for you. Right. But like Codex and ChatGPT …”