Andrew Ambrosino

1. Key Themes

Implementation Is No Longer the Bottleneck — Curation Is

The fundamental inversion of the product process is the core thesis of this episode. Andrew argues that the entire design process — research, docs, prototypes — was predicated on implementation being expensive. That assumption is now broken, and the scarce resource has shifted entirely.

"It's backwards. The implementation is actually not the expensive part anymore. It's, dare I say taste. 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?" 00:00:49

The Medium You Choose to Express an Idea Now Carries Enormous Signal

Because prototypes and documents are both now trivially easy to produce, the act of choosing which medium to use is itself a high-stakes decision. Picking the wrong medium anchors people to the wrong stage of the process.

"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 people's hands to try out and to stress test an interaction pattern, it's a prototype... it's really important to pick the medium." 00:08:05

"There's a mismatch between... you see this fully polished prototype that looks like it's ready to go out the door and enough people at a company see that and they're like, can we release this now? But the appropriate... we're actually in the early design process stage and nobody's just saying that." 00:19:22

AI Is Structurally Lagging at Design — and the Reasons Are Non-Obvious

Andrew offers a layered explanation for why AI underperforms at design that goes beyond "it hasn't been trained enough." He identifies three compounding problems: grading difficulty, misaligned research incentives, and a missing abstraction layer between visual and code semantics.

"Design's a little bit harder to grade, than software... creating a loop where you can train the model and like what's good design and what's bad design is just a little bit more tedious and onerous than, you know, does the code compile?" 00:00:35

"The labs historically invest in making their models good at things that accelerate AI research and... the model being able to write correct code would accelerate research, right? In a way that you can't really make the same case for design." 00:13:04

Role Collapse Is Real But Dangerous If Misunderstood

Andrew warns against the fashionable move of eliminating product roles entirely. He distinguishes between healthy blurring of boundaries versus the irresponsible abandonment of accumulated disciplinary knowledge.

"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. And everybody's just going to be like a builder. And then what happens is this whole discipline of product that's been built up and has like real best practices, real things that have been tried and failed... just gets abandoned because people are like, oh, I wrote some code." 00:01:12

"Your role is the average of what you spend your time on." 00:23:29

The Codex App Is Quietly Becoming a General Knowledge Work Platform — Not Just a Dev Tool

One of the most operationally significant revelations in the episode is that non-engineers at OpenAI refused to leave Codex for apps designed specifically for them, forcing the team to reconsider the entire product strategy.

"We had 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... nobody would leave the Codex app for the apps that were allegedly for these other personas." 00:53:13

Product-Market Fit for AI Features Depends on Model Timing — Not Just Shape

Andrew articulates a non-obvious planning insight: the same product can fail or succeed based purely on which model powers it, with no changes to the product itself. This reframes the concept of "failed product" entirely.

"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." 00:33:23

"The whole premise of whether features were good or not were based on whether they were smart enough, not the shape of them." 00:33:23

Build Features Before Models Are Ready — Then Shelve Them and Wait

Rather than planning linear roadmaps, Andrew's framework is to prototype broadly, park things that are model-limited, and continuously re-test them as model capability jumps.

"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 there's like a new leap in models, let's try that thing again with it swapped out." 00:32:56

The Codex App's Direction: A Home Base That Reaches Into Other Apps

The vision is not a super app that replaces everything but rather an orchestration layer — an AI home base that delegates out to existing specialized tools rather than trying to rebuild them.

"It's not just about, hey, we're drawing a rectangle on the screen and everything needs to happen in that rectangle. It's this thing should be a home for you where you start work, you end work, you automate work, and it uses whatever you need to do." 00:56:48

"Codex and ChatGPT can use that video editor, right? It can interact, offload stuff to it... And then there is Dan Shipper's thing, which is, hey, I have these web apps that you can click around and use, but I want to be able to open these in Codex and have Codex do extra stuff with it." 00:58:33

Dog Fooding as Product Discovery, Not Convenience

The Codex team uses Codex for everything — deliberately, even when it's not the best tool — because the friction of doing so is the product discovery mechanism.

"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... we often don't improve our process so that we can make the product better to do it, which is a deeply like uncomfortable place to be in." 00:23:01

2. Contrarian Perspectives

PRDs Are Not Dead — And Neither Is the Design Process

Against the popular narrative, Andrew argues that PRDs and the design process still have value. The real problem is confusing the medium with the stage of work.

"PRDs are dead and prototypes are in. And I actually don't believe this at all... for engineers, it's really tempting to write a lot of documents, a lot of documents that are not worth reading... if implementation is abundant, then it's really important to pick the right format for the point you're trying to make." 00:07:05

"To say that the design process is dead, I feel like it's both true and false... to throw the process out completely or throw like the overlay of the process... that is still more important than ever." 00:20:41

Being Too AGI-Pilled Can Kill a Product

The original Codex web product failed not because the idea was wrong but because it assumed more model capability than existed. Claude Code, with a more conservative interaction model, won. The lesson is that matching product ambition to actual model capability is more important than projecting where models will be.

"We were too AGI pilled for the moment... Claude Code comes out, totally local, doesn't pretend to be as — it's not as AGI pilled, right? It's like, can I ask you questions? It's going to sit there. You can't just delegate your life to it. That worked way better, right? Cause that's the point that the models were at." 00:37:27

AI Models Are Systematically Bad at Deleting Code — and Nobody Is Fixing It

Andrew raises what he calls a core problem with putting development on autopilot that gets almost no public attention: models increase code complexity rather than reduce it, because they have no incentive to delete.

"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." 00:40:35

Design Novelty Is Actually More Important Than Design Correctness — Unlike Engineering

Most people assume AI will master design the same way it mastered coding, just with more training. Andrew argues design has a structural difference: novelty is a feature, not a bug, whereas engineering rewards pattern replication.

"There's an amount of like novelty that is more important in design than it actually isn't software engineering. Software engineering, you almost want it to over-index on known patterns. Whereas design, it's like, no, there's an element of randomness here and novelty." 00:14:19

Don't Get Married to Your Process — Get Married to the Outcomes You Deliver

This sounds like generic advice but the implication is radical: your process is now expected to change continuously and clinging to tool mastery is a career liability.

"Do not get married to your exact process. Get married to like the outcomes that you were uniquely able to deliver and then do things like change your process to try things... I'm the best at understanding Figma auto layout. Like, right. What are you doing?" [00:08:23 — post-recording section at 01:07:55]

3. Companies Identified

OpenAI

AI research and product company. The context for the entire episode — how Codex grew from CLI to the flagship desktop app, with 5 million weekly active users growing 6x since January, and nearly 100% internal employee usage.

"Internally at OpenAI, nearly 100% of their employees use Codex weekly. And that is not just the engineers." 00:01:44

Codex (OpenAI)

OpenAI's desktop coding and knowledge-work agent app, now expanding beyond developers.

"Since this January, Codex usage has grown 6x. They currently have over 5 million weekly active users." 00:01:44

Anthropic / Claude Code

Mentioned as a direct competitive reference. Claude Code's more conservative, non-AGI-pilled interaction model is credited with outperforming the original Codex web product in the market.

"Claude Code comes out, totally local, doesn't pretend to be as — it's not as AGI pilled, right?... That worked way better." 00:36:58

Linear

Design and project management SaaS tool. Cited twice: once as the gold standard of software product design that models simply replicate, and once as Andrew's favorite software product before Codex.

"Every new website that came out was just a copy of Linear's website. Linear's website, great design, great taste. If a model did that, I'd be like, wow, this is an incredible leap here. Right. If I have a model that outputs Linear's website every time, that's not the challenge here." 00:13:54 "Linear until this, Linear was like my favorite, at least software product." 01:04:34

Figma

Design tool mentioned as part of the historical design process evolution — pulling prototyping earlier into the workflow.

"What we saw with Figma and Origami and all of these tools is that you can fast forward some of the insights by pulling interactive prototypes earlier into the process." 00:18:25

WorkOS

B2B enterprise authentication platform. Sponsor, described by Lenny as "essentially Stripe for enterprise features."

"Literally every startup that I'm an investor in that starts to expand upmarket ends up working with WorkOS." 00:05:52

Mercury

Business banking fintech. Sponsor, praised for building banking "by product people, not by bankers," with API, CLI, MCP server, and a new conversational AI interface called Command.

"Does your bank have an API, a terminal native CLI, or an AI ready MCP server? I don't think so." 00:34:27

Adobe Premiere Pro

Video editing software. Codex autonomously built an extension for Premiere Pro to enable video editing through natural language — cited as a breakthrough example of the app's extensibility model.

"What Codex did was built itself an extension that could be installed into Premiere Pro that it could then talk to and say, hey, Premiere Pro extension, can you please change this marker inside of the Premiere Pro app? That was pretty nuts when we saw that happening." 00:58:08

Microsoft Excel

Mentioned as a specific integration example where Codex talks directly to Excel's add-in rather than replacing it.

"The app talks directly to the add-in in Microsoft Excel on your desktop. And when it's done, you can close Excel." 00:56:48

Every (Dan Shipper's company)

AI-focused media and product company. Dan Shipper's prediction that people will run SaaS apps inside Codex is described as a persistent, active product thesis that OpenAI is engaging with.

"Dan Shipper was on the podcast, he had this prediction that we're going to start using Codex to run our SaaS apps inside of it. So instead of going in Chrome. I know he slacks me about this every day, asking for stuff." 00:49:06

Replit

Listed among companies powered by WorkOS in the sponsor read. 00:05:22

Vercel

Listed among companies powered by WorkOS. 00:05:22

Sierra

Listed among companies powered by WorkOS. 00:05:22

Clay

Listed among companies powered by WorkOS. 00:05:22

Cursor

Listed among companies powered by WorkOS. 00:05:22

Netflix

Mentioned as the platform for the revived Magic School Bus series. 01:04:05

Notion

Mentioned as a tool Andrew used to track the Codex launch, and as a common pattern people set up for memory/knowledge bases.

"I had a, you know, Notion doc somewhere with everything that needed to happen. And I was like automating, like going out to gather updates from pull requests from Slack channels and like updating the status tracker." 00:43:34

Obsidian

Mentioned as a common personal knowledge base tool people set up with AI — cited as something that should become a native product primitive instead.

"A lot of people... are like, well, I set up an Obsidian base or a Notion area and I tell it how to basically build my mind palace." 00:00:28

4. People Identified

Product and engineering lead for the Codex desktop app at OpenAI. Designer-turned-engineer-turned-PM, former founder. His background across all three disciplines and his role building the product he uses to build the product gives him an unusually credible lens on role collapse and the future of product work.

"I spent so long feeling like I should not be a software engineer because I didn't care about like assembly language or memorizing TypeScript syntax. And it's like, there have always been parts to these roles that are that sort of gatekeeping." 00:26:27

Dan Shipper

Co-founder of Every. Mentioned for his persistent, specific prediction that Codex will become the browser through which people run their SaaS apps.

"Dan Shipper was on the podcast, he had this prediction that we're going to start using Codex to run our SaaS apps inside of it... I know he slacks me about this every day, asking for stuff." 00:49:06

Jenny (Head of Design, Claude Code / Anthropic)

Referenced by first name as having appeared on a recent episode of Lenny's podcast. Her thesis — that the design process is dead and design now steers things in motion rather than upfront — is a direct foil for Andrew's more nuanced view.

"Jenny, who is the head of design for Claude Code... she had this whole kind of thesis that the design process is dead. There's no time for design. Things are moving too fast. Just build now and design is kind of steering things as things move along." 00:16:47

Alexander (PM, Codex team at OpenAI)

Senior product person on the Codex team. Cited specifically for having a master's degree in CS, representing the kind of technically deep PM that is thriving in this new environment.

"Alexander has a master's degree in computer science, which is — I do not have a master's degree in computer science." 00:22:05 "I have had a lot of conversations with Alexander about this zone defense analogy." 00:29:00

Brent (In-house DX Videographer, OpenAI)

Named specifically as the first person to use Codex for video editing — and for the app independently building a Premiere Pro extension to enable it. A vivid proof point of the app's extensibility ambitions.

"Our in-house DX videographer, Brent, was tasked with editing all these videos. Right. And he edited all the videos with Codex... Codex is not a video editor per se... it was able to understand that he used Premiere Pro... but it couldn't do everything. So naturally what Codex did was built itself an extension that could be installed into Premiere Pro." 00:57:11

Paul Graham

Y Combinator co-founder. Cited as an example of a person with demonstrably great taste who defies the aesthetic stereotype of the word — used to argue that taste is primarily a systems-thinking and judgment concept, not an aesthetic one.

"Paul Graham clearly has great taste and wears cargo shorts, right? Like, we gotta tease out what taste means a little bit." 00:10:44

5. Operating Insights

Use a "Baby Version" of Your Product as a Design Prototyping Environment

The Codex team maintains a dramatically simplified replica of the production codebase specifically for rapid design exploration — a low-risk sandbox that approximates all real interactions but can be vibe-coded over without consequence.

"Many companies right now have this idea of like a baby version of the product, like baby Cursor. We have baby Codex, 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? Cause you can be like, well, what if the sidebar worked like this? Or what if a pane came in and had like a group chat here?" 00:20:14

Use "Zone Defense" to Distribute Product People Across Gaps, Not Clusters

Rather than having PMs work closely together on the same areas, Andrew deliberately spaces them out to maximize coverage of the problem space — like a force-directed graph pushing nodes apart to fill gaps.

"If two product people are working too closely, that's often not a good signal... you kind of want, as a product org, to do this force-directed activity where you're like, where are the gaps... The tastemakers to guide things from inception to what the product should be. And that means you basically want company coverage." 00:29:00

Set Up a Daily Automated Brief From Your Own Communication Channels

Andrew describes a concrete personal operating system: a scheduled Codex task that reads his 3,000 Slack channels, filters for what matters, and generates a prioritized morning brief that he can coach over time.

"I will get up in the morning. I will see the daily brief that I have from like everything from the 3,000 Slack channels that I'm in, like which things need my attention. I can kind of message back and be like, all right, give me five questions and I'll answer them." 00:44:33 "Next time this runs, like, can you please worry about this instead? Or can you deemphasize this work stream... I can kind of coach it along the way. It'll update the way that it notifies me." 00:45:17

Treat "Failed" Features as Versioned Artifacts to Test Against Future Models

Rather than killing features that underperform, Andrew advocates parking them and re-testing with each major model upgrade — since the failure was often model-capability-limited, not product-shape-limited.

"Push people not to be stubborn about, no, this isn't working, so it's a bad feature. Like, no, it might not be ready yet... The re-releasing of it with different intelligence totally changes the outcome here." 00:36:28

Add Precision Inversely to Horizon — Hazy Long-Range Plans Are Honest Plans

Any precision added to a nine-month plan is described as false precision that wastes organizational time. The operating rule is: the further out the plan, the hazier it must stay.

"The shorter term something is, the more detail it needs... 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 precision. And like, you're just going to waste time." 00:32:02

6. Overlooked Insights

The Abstraction Layer Between Visual Design and Code Is the Real Unsolved Problem in AI Design

Andrew briefly describes something that nobody in the conversation pauses to fully examine: there is a layer of design intelligence that is not about aesthetics at all but about shared code abstractions between visually different components. This is not a training data problem or a taste problem — it is a structural reasoning problem that current models cannot handle, and it is what actually blocks AI from taking over design end-to-end.

"There's sort of an abstraction layer that is an interplay between the software design and the code that's being written. Like this thing over here in this corner should share X, Y, and Z in the code base with this thing down here... if tomorrow our company did a rebrand, the shallow version of this is that we have to update 263 components one by one. The deep version is like the semantics between these two things that look different... they're both in a list that have the style that conveys this interaction pattern to the user. And I think like that is still feeling a little out of reach with the current technology." 00:14:48

The investment implication: any startup that solves semantic design abstraction reasoning — not pixel generation, not component suggestion, but the structural reasoning layer between visual intent and code architecture — would be filling a gap that OpenAI itself acknowledges as unsolved.

The Codex Video Editor Story Reveals a Hidden Extensibility Business Model

The Premiere Pro extension story was mentioned as a cool anecdote but its strategic significance was dropped immediately. Codex, when given a task it cannot complete natively, autonomously writes and installs its own extensions into third-party software. This is not a product feature — it is a potential platform strategy that bypasses the need for formal integration partnerships entirely. If generalized, it means Codex could extend itself into any desktop application without the target company's cooperation.

"What Codex did was built itself an extension that could be installed into Premiere Pro that it could then talk to and say, hey, Premiere Pro extension, can you please change this marker inside of the Premiere Pro app? That was pretty nuts when we saw that happening." 00:58:08

The implication for investors and operators: this autonomous extension-writing capability, if productized deliberately, could make Codex a de facto integration layer across the entire desktop software ecosystem — competing with Zapier, MCP servers, and enterprise integration middleware without any of the standard partnership overhead.