Feb 12, 2026 · 1h 19m · lennys-podcast
OpenAI’s head of platform engineering on the next 12-24 months of AI | Sherwin Wu
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
Sherwin Wu, OpenAI's Head of Platform Engineering, joins Lenny Rachitsky to discuss how autonomous agent orchestration is transforming software engineering, the evolution of technical management, and strategic product design in an era of rapidly advancing foundational models.
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 21.4% of the talking time here. How this is scored →
Lenny as informed peer 4.1 Guest teaching 6.1 Guest disagreement 1.4 Lenny pushing back 1.6
speaking balance: gold is Lenny, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 29:31 Pushing back on the internal support limitation constraint
Sherwin rejects the premise that a single-person billion-dollar startup is limited by internal support burdens, arguing that software cost deflation enables outsourcing to hyper-specialized micro-SaaS providers.
Hardest push from Lenny ▶ 28:31 Lenny's skepticism on single-person billion-dollar scaleLenny directly challenges the one-person billion-dollar startup narrative by highlighting the unavoidable bottleneck of scaling human customer support tickets.
Biggest teaching moment ▶ 44:40 Explaining why models eat product scaffoldingSherwin educates the host on the AI bitter lesson, explaining how blindly listening to customer requests for vector stores or agent frameworks locks teams into local maxima before models absorb those layers.
Lenny holds their own ▶ 35:18 Synthesizing the surgeon metaphor into an automated AI unblockerLenny takes Sherwin's management analogy of the surgeon and synthesizes an actionable prompt idea for querying organizational knowledge bases to anticipate engineer blockers.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Lenny as informed peer | Guest teaching | Guest disagreement | Lenny pushing back | Why |
|---|---|---|---|---|---|---|
| AI-Driven Software Engineering at OpenAI | 4 | 6 | 1 | 2 | Sherwin shares internal OpenAI metrics regarding Codex adoption and PR volume. Lenny probes on the exact definition of AI-written code and references previous conversations with Kevin Weil and OpenClaw developer Peter. | |
| Software Engineers as Wizards and Tech Leads | 4 | 6 | 1 | 1 | Sherwin articulates the evolution of engineers from direct coders into managers of agent fleets, invoking SICP and the Sorcerer's Apprentice metaphor. Lenny complements the framing with the genie metaphor. | |
| Overcoming Agent Failure Modes and Context Bottlenecks | 3 | 6 | 1 | 1 | Lenny asks about developer stress when agents fail, leading Sherwin to reveal an internal OpenAI experiment maintaining a 100% Codex-written codebase without manual escape hatches. | |
| Streamlining Code Reviews and CI/CD Pipelines | 4 | 5 | 2 | 3 | Lenny pushes on potential circular risks when Codex writes and reviews its own code and asks if multi-model review is utilized. Sherwin qualifies his claims by explaining human oversight remains at 30% attention. | |
| The Evolution of Engineering Management in AI | 5 | 5 | 1 | 1 | Sherwin discusses how engineering managers can manage wider spans of control and spend majority time unblocking top performers. Lenny reinforces the point with a quote from Marc Andreessen. | |
| Unpriced Impacts: One-Person Billion-Dollar Startups and Micro-SaaS | 6 | 6 | 3 | 5 | Lenny explicitly challenges the concept of a one-person billion-dollar startup due to customer support scaling limits. Sherwin pushes back with an alternative framing around an ecosystem of specialized micro-SaaS services. | |
| Management Lessons: The Surgeon Metaphor and AI Blockers | 5 | 5 | 1 | 1 | Sherwin describes his management philosophy via the Mythical Man-Month surgeon metaphor, prompting Lenny to propose using internal AI agents to predict organizational blockers ahead of time. | |
| Sponsor Segment: Datadog and Eppo Experimentation | 4 | 7 | 2 | 2 | Following the mid-roll sponsor break, Sherwin explains why many corporate AI initiatives suffer negative ROI due to top-down mandates lacking bottom-up technical champion teams. Lenny synthesizes the anti-pattern. | |
| Navigating Fast-Moving AI: The Bitter Lesson and Scaffolding | 5 | 7 | 2 | 1 | Lenny asks about the contrarian take that listening to customers can misguide AI roadmaps. Sherwin explains how rapid model improvements eat scaffolding like vector stores and agent frameworks. | |
| The Roadmap Ahead: Long-Horizon Tasks and Native Multimodal Audio | 3 | 7 | 1 | 1 | Sherwin outlines upcoming technical advances, citing METR benchmarks on multi-hour task execution horizons and native speech-to-speech multimodal capabilities. | |
| The Untapped Frontier of Business Process Automation | 4 | 7 | 2 | 1 | Sherwin explains the untapped value of deterministic business process automation outside tech bubbles. Lenny clarifies the scope and economic impact compared to software engineering. | |
| OpenAI Platform Philosophy, Neutrality, and Global Access | 4 | 6 | 1 | 2 | Sherwin outlines OpenAI's platform neutrality, commitment to not blocking competitors, and the mission to democratize frontier models to hundreds of millions of weekly active users. | |
| The OpenAI Developer Tooling Stack | 3 | 7 | 1 | 1 | Lenny asks for an overview of the developer platform architecture. Sherwin systematically breaks down the layer stack from Responses API up through Agents SDK, Agent Kit, and Evals. | |
| Career Advice for Navigating the Generative AI Era | 4 | 5 | 1 | 1 | Sherwin offers career advice on embracing the next few years of rapid AI progress, advising practitioners to filter out X-driven noise by adopting a couple of core tools hands-on. | |
| Lightning Round: Books, Anime, Ubiquiti, and Real Estate Insights | 3 | 6 | 1 | 1 | In the lightning round, Sherwin shares book picks, anime interests, Ubiquiti hardware setups, and quantitative real estate pricing variables from his Opendoor tenure. |
Statements from this episode (25)
Wu: OpenAI engineering managers write all their code using Codex
“And so I know for myself and some of the other EMs,
Engineering managers at OpenAI.
all of our code is written by Codex at this point.”
Wu: 95% of OpenAI engineers use Codex and 100% of PRs are AI-reviewed
“So, 95% of engineers use Codex.
100% of our PRs are reviewed by Codex daily as well, so basically any code that goes into production that's merged in, Codex kind of has its eyes on and suggests improvements, suggests changes in the PRs.”
Wu: OpenAI engineers using Codex open 70% more PRs
“So they're actually opening 70% more PRs and than the engineers who aren't using Codex as much and the gap is widening.”
Wu: OpenAI engineers juggle up to 20 parallel Codex threads simultaneously
“I know many of the engineers on my team basically have like, 10 to 20 threads kind of being pulled on at the same time. Obviously not active running codex jobs. But just a lot of parallel threads. They're checking in on what they're doing. They're steering the…”
Wu: OpenAI team is maintaining a 100% Codex-written codebase without manual coding
“So there's a team that that's actually doing an experiment right now with an open AI where they are basically maintaining a 100% code expert and code base. So, you know, like, you know some, you know, you'll have the AI write code, but you'll obviously end up …”
Wu: Coding agent failures are usually caused by missing context or underspecification
“A lot of the time when the coding agent is not doing what you want, it's usually a problem with context and just, like, information that you've given it. It's just either underspecified or there's just not enough information around how to do something availabl…”
Wu: Codex cuts OpenAI code review times from 15 to 3 minutes
“And it makes, you know, code reviews go from a, you know, I don't know, 10:15 minute task to sometimes even just like a two to three minute task, because you have a bunch of suggestions already, already baked in.”
Wu: The vast majority of code at OpenAI is authored by AI
“Almost every engineer heavily uses Codex in all of their tasks at this point, and so I, you know, if I were to guesstimate, like, the vast majority of code at this point is, It was probably authored by AI.”
Wu: AI tools widen the productivity spread across engineering teams
“Codex really empowers, like, top performers to get a lot, like, to be a lot more productive, and so it really, like, and I think this may be true for AI more broadly, like, across society, which is, like, the people who really lean in, or, like, the people who…”
Wu: OpenAI managers use internal ChatGPT for employee performance reviews
“We're doing performance reviews right now, and it's actually really easy to use ChatGPT with internal knowledge, hooked up to GitHub and like our Notion docs and Google docs. To give a, get a really good sense of what this person has done over the last 12, 12 …”
Wu: AI will enable engineering managers to manage far more than 6-8 engineers
“My sense is I think managers will be able to manage much larger teams in this world. Kind of like how, you know, like software engineers are managing 20 to 30 codexes. My sense of these tools will allow managers, people manage to be higher leverage and will al…”
Wu: A one-person billion-dollar startup will likely exist eventually
“It's like, yeah, if, you know, if people are so high leverage, at some point there will likely be a one person billion dollar startup.”
Wu: Rise of AI micro-companies could shrink venture-scale return opportunities
“You know, it might, we might end up in in a world where there's just, like, a handful of big players that are offering platforms and supporting all of these startups, but, you know, the types of venture-scale return startups that can really hundred or thousand…”
Rachitsky: Bearish on one-person billion-dollar startups due to support scaling
“It's hard for me to imagine one person, like I'm bearish on this billion dollar startup. I just want to share this thought simply because of the support costs, even if AI is helping you at a billion dollars, just like, unless your ACVs are, you know, very high…”
Wu: Organizational barriers are the main bottleneck to AI-assisted shipping
“If people are just like cranking PR after PR, the main thing bottlenecking progress and, you know, shipping something tends to be organizational or like process oriented.”
Wu: Tech-adjacent non-engineers drive the most enthusiasm for enterprise AI adoption
“The pattern I've seen is it tends to be these like software engineering adjacent, like basically technical people, but are not software engineers. I think those are the ones who get tend to get most excited around this. It's like, you know, maybe the, it's lik…”
Wu: Blindly following customer requests leads to AI product local maxima
“Because the field is changing so much at any point in time, you know, a lot of people are kind of in this local, local maximum. And if you just blindly listen to your customers, they'll, they'll be like, yeah, I want a better vector store. Like I want a better…”
Wu: AI startups should build for capabilities that are 80% viable
“My general advice, and I've been giving this to people for a while, and I think it's still true today, is make sure you're building for where the models are going and not where they are today. You know, the, it's clearly a moving target, and I think a lot of t…”
Wu: AI models could execute multi-hour to day-long tasks in 12-18 months
“If you follow this trend, like, I think, like, in the next 12 to 18 months, we could see models that could do multi-hour long tasks very, very coherently. At some point, it might reach, like, you know, six hours a day long task.”
Wu: Native speech-to-speech multimodal models will improve dramatically in 6-12 months
“I think they're gonna get a lot better at audio over the next six to 12 months, especially the likes, you know, the native multimodal models, the speech-to-speech ones.”
Wu: Enterprise audio AI is hugely underrated compared to text coding tools
“Audio, especially in the enterprise and in a business setting, I think is a hugely underrated domain still. Like, everyone talks about coding. It's all text. But we're talking in audio. A lot of the world's business is done via audio. A lot of services and ope…”
Sherwin Wu: AI business process automation is massively underrated in Silicon Valley
“And so I'm just extremely bullish on this general category of, like, And I think it's underrated because it's so different from what we think about in Silicon Valley, people tend to not think about it. But how can we apply AI and some of the tools and framewor…”
Wu: AI startups fail from lack of customer resonance, not OpenAI competition
“Every startup that I've seen that is kind of fizzled out is not because open AI or, you know, big lab or Google or something has come to squash them. It's because they built something and it like really didn't resonate with the customers.”