Nick Turley (Head of ChatGPT at OpenAI)09 August 2025

Inside ChatGPT: The fastest-growing product in history

6Frameworks

15Insights

Episode overview

Lenny Rachitsky interviews Nick Turley about how ChatGPT emerged from an OpenAI hackathon effort and grew into a product with more than 700 million weekly active users. Turley explains why AI teams must ship early to discover emergent use cases, then improve model behavior, capabilities, safety, and interface quality using real-world evidence. They also discuss GPT-5, ChatGPT’s evolution toward a personalized AI with broader action space, and the importance of keeping users in control as systems become more agentic.

Key ideas

ChatGPT began as a rapidly productized, open-ended research demo intended to reveal real use cases, not as a confidently forecast consumer success.
For AI products, the model and product are inseparable: teams must observe real usage and systematically improve models for writing, coding, advice, recommendations, and other valued tasks.
Turley divides retention gains roughly among model improvements, research-driven capabilities such as search and memory, and conventional product changes such as removing login friction.
“Maximally accelerated” is a forcing question for identifying the true critical path, while high-stakes work such as frontier-model safety requires deliberate red-teaming, system cards, safeguards, and external input.
Natural language will remain central, but Turley believes turn-by-turn chat is limiting and expects AI systems to render richer interfaces and take actions through tools.
ChatGPT is intended to amplify users rather than replace them, with confirmations, visible agent activity, and other design choices preserving user control.
Because AI capabilities and use cases are emergent, shipping should be an intentional early point in the journey toward a polished product, followed by disciplined iteration.
Turley advises early-career people to surround themselves with energizing, capable colleagues and follow genuine curiosity, because asking the right questions becomes more valuable as AI can answer more of them.

Transcript available · source text is retained privately and is not published

Frameworks in this episode

Leadership4 steps

Barrels-and-Ammunition Team Design

Staff each team from the gap, not from a fixed PM/EM/designer template

Read →

Communication4 steps

Evals as Articulating Success

An eval is just a clear spec of ideal behavior — the shared language of AI product work

Read →

Strategy5 steps

Run Toward the Hard Use Cases

Don't disable high-stakes uses to avoid downside — engineer them to be great

Read →

Innovation5 steps

Ship-to-Learn: The Emergent-Product Loop

When product properties are emergent, launching is how you discover them

Read →

Leadership4 steps

The Maximally Accelerated Question

A forcing question that separates critical path from what can wait

Read →

Productivity5 steps

Two-Mode Prioritization for AI Products

Prioritize backward from model magic AND forward from customer needs

Read →

People & resources mentioned

Attributed to the moment in the episode. Timestamps are approximate.

People · 16

Sam AltmanMentionsthere's everybody has uh and had and continues to have like such deep trust in in Sam and Greg and our leadership team
“And then Sam Alman's just like, "Hey, let me tweet about it."”
▶ Watch at ~0:00approx.
Kevin WeilMentionsWe got connected through Kevin Wheel, who is former CPO at OpenAI, now head of science at OpenAI.
“Kevin Wheel, your CPO, said to ask you about this principle of is it maximally accelerated?”
▶ Watch at ~0:30approx.
Nick TurleyMentionshead of ChachiPT, Nick Turley was like almost going to become professional jazz pianist.
“Today my guest is Nick Turley. Nick is head of chatbt at OpenAI.”
▶ Watch at ~1:00approx.
Claire VoMentionsClaire is a longtime chief product officer at color optimizely and currently Chief product officer at launch darkle
“A huge thank you to Kevin Wheel, Claire Vo, George O'Brien, Joanne Jen, and Peter Ding for suggesting topics for this conversation.”
▶ Watch at ~1:30approx.
Peter DengMentionsHe was VP of product at OpenAI where he oversaw product design and engineering for chat GBT
“A huge thank you to Kevin Wheel, Claire Vo, George O'Brien, Joanne Jen, and Peter Ding for suggesting topics for this conversation.”
▶ Watch at ~1:30approx.
Joanne JangMentionsthe person who I recruited who recruited me uh Joanne
“A huge thank you to Kevin Wheel, Claire Vo, George O'Brien, Joanne Jen, and Peter Ding for suggesting topics for this conversation.”
▶ Watch at ~1:30approx.
George O'BrienMentionsmentioned
“A huge thank you to Kevin Wheel, Claire Vo, George O'Brien, Joanne Jen, and Peter Ding for suggesting topics for this conversation.”
▶ Watch at ~1:30approx.
Christina CacioppoMentionsI am very excited to have Christina Copo CEO and co-founder vanta joining me
“Christina Casiopo, CEO and co-founder of Vanta, joining me for this very short conversation.”
▶ Watch at ~3:00approx.
Lenny RachitskyMentionstoday we've got another very special compilation episode something I've been pulling on more and more with the podcast and the newsletter
“Thanks for having me, Lenny.”
▶ Watch at ~4:30approx.
Rahul VohraMentionsRahul from superhuman is kind of well known for his approach to finding product Market fit
“This is the survey that Rahulvore at superhuman popularized in his first round article.”
▶ Watch at ~40:30approx.
Charlie MungerMentionsHis words and documents were assembled into Poor Charlie's Almanack.
“Charlie Munger, maybe.”
▶ Watch at ~58:30approx.
Keith RaboisMentionsKeith or boy who's uh who I think led many of the rounds of ramp he started Founders fun famous investor
“I think Keith Ke Keith Ro calls this like like barrels I think um barrels of ammunition”
▶ Watch at ~1:07:00approx.
Mike KriegerMentionsMike is chief product officer at Anthropic the company behind Claude. He's also the co-founder of Instagram.
“both Kevin Wheel and Mike Kger shared which is that eval are becoming a huge new skill”
▶ Watch at ~1:14:00approx.
Joanne JangMentionsthe person who I recruited who recruited me uh Joanne
“the person who I recruited who recruited me uh Joanne”
▶ Watch at ~1:24:30approx.
John RawlsRecommendsmentioned
“anything by like rolls and nosic like I like the political stuff”
▶ Watch at ~1:30:00approx.
Robert NozickRecommendsmentioned
“anything by like rolls and nosic like I like the political stuff”
▶ Watch at ~1:30:00approx.

Resources · 49

InstacartMentionscompany · Apoorva Mehta, Max Mullen, and Brandon Leonardo
“You were a product leader at Dropbox, then Instacart.”
▶ Watch at ~0:00approx.
DropboxMentionscompany · Drew Houston and Arash Ferdowsi
“You were a product leader at Dropbox, then Instacart.”
▶ Watch at ~0:00approx.
ChatGPT SearchMentionssoftware · OpenAI
“I've been really excited about what we've been doing in search.”
▶ Watch at ~1:00approx.
TwitterMentionswebsite · Twitter
“Chat GPT is now driving more traffic to my newsletter than Twitter.”
▶ Watch at ~1:00approx.
Lenny's NewsletterCoinednewsletter · Lenny Rachitsky
“Chat GPT is now driving more traffic to my newsletter than Twitter.”
▶ Watch at ~1:00approx.
OpenAIMentionscompany
“Nick is head of chatbt at OpenAI.”
▶ Watch at ~1:00approx.
Lenny's PodcastCoinedpodcast · Lenny Rachitsky
“If you enjoy this podcast, don't forget to subscribe and follow it in your favorite podcasting app or YouTube.”
▶ Watch at ~1:30approx.
Lenny's NewsletterRecommendsnewsletter · Lenny Rachitsky
“Check it out at lenny'snewsletter.com and click bundle.”
▶ Watch at ~2:00approx.
ConductorMentionssoftware · Orkes
“the company behind Open-source Conductor, the orchestration platform powering modern enterprise apps and agentic workflows.”
▶ Watch at ~2:00approx.
OrkesMentionscompany
“This episode is brought to you by Orcus, the company behind Open-source Conductor”
▶ Watch at ~2:00approx.
VantaCoinedsoftware · Christina Cacioppo
“we started Vanta in 2018 focused on founders”
▶ Watch at ~3:30approx.
VantaRecommendssoftware · Christina Cacioppo
“They can get $1,000 off Vanta at vanta.com/lenny.”
▶ Watch at ~4:30approx.
GPT-4oUsessoftware · OpenAI
“you've probably been on GPD40 for you know a while”
▶ Watch at ~5:30approx.
GPT-5Usessoftware · OpenAI
“I've been using it for a while.”
▶ Watch at ~6:00approx.
GPT-5Mentionssoftware · OpenAI
“it is also, you know, the smartest um most useful and um fastest Frontier model um that we've ever launched.”
▶ Watch at ~6:00approx.
OpenAIMentionscompany
“that is the thing that I think most people notice”
▶ Watch at ~6:00approx.
o3Mentionssoftware · OpenAI
“especially if they weren't using 03 already.”
▶ Watch at ~7:00approx.
ChatGPT MemoryMentionssoftware · OpenAI
“you saw us launch some of those things, you know, with uh, improved memory earlier this year.”
▶ Watch at ~11:30approx.
ChatGPT Agent ModeMentionssoftware · OpenAI
“we built this way of sort of watching what the AI is doing when it's in agent mode.”
▶ Watch at ~13:00approx.
WaymoMentionscompany · Alphabet
“in the same way that when you're in a Whimo, you you get that screen”
▶ Watch at ~13:00approx.
GPT-4Mentionssoftware · OpenAI
“right around the time that GB4 finished training”
▶ Watch at ~14:00approx.
ChatGPTCoinedsoftware · OpenAI
“we decided to ship something open-ended because we just wanted a real use case distribution”
▶ Watch at ~16:00approx.
ChatGPTUsessoftware · OpenAI
“every question I have, I go to it.”
▶ Watch at ~18:00approx.
InstacartMentionscompany · Apoorva Mehta, Max Mullen, and Brandon Leonardo
“I actually learned that at Instacart when I when I showed up there”
▶ Watch at ~22:30approx.
ChatGPT SearchMentionssoftware · OpenAI
“search is one example of that”
▶ Watch at ~30:30approx.
FigmaUsessoftware · Figma
“Like I love being in Figma.”
▶ Watch at ~35:30approx.
Google DocsUsessoftware · Google
“I love being in, you know, uh, Google Docs.”
▶ Watch at ~35:30approx.
Code InterpreterMentionssoftware · OpenAI
“remember when we shipped code interpreter we learned so much after u we shipped it”
▶ Watch at ~38:00approx.
Data AnalysisMentionssoftware · OpenAI
“now it's known as I think data analysis and chat GBT or something like that”
▶ Watch at ~38:30approx.
DiscordUsessoftware · Discord Inc.
“ship a Google form to Discord with like I think the four questions you're supposed to ask”
▶ Watch at ~39:00approx.
Google FormsUsessoftware · Google
“ship a Google form to Discord with like I think the four questions you're supposed to ask”
▶ Watch at ~39:00approx.
Van Westendorp Price Sensitivity MeterUsesother · Peter van Westendorp
“the survey just to give it the official name the van western drop survey uh is how you guys ended up pricing chap.”
▶ Watch at ~40:30approx.
DisneyMentionscompany · The Walt Disney Company
“someone at Open who I really respect sometimes uses is like we're kind of like Disney”
▶ Watch at ~47:00approx.
TikTokUsessoftware · ByteDance
“there's these crazy Tik Tok posts that go viral and they have like 2,000 use cases in the comments”
▶ Watch at ~48:00approx.
AirtableMentionssoftware · Airtable
“you think about air tableable you think about notion all these companies they took like years to just build and craft”
▶ Watch at ~48:30approx.
NotionMentionssoftware · Notion Labs
“you think about air tableable you think about notion all these companies they took like years to just build and craft”
▶ Watch at ~48:30approx.
MS-DOSMentionssoftware · Microsoft
“I still feel like chat feels a little bit like MS DOS.”
▶ Watch at ~50:00approx.
Microsoft WindowsMentionssoftware · Microsoft
“we haven't built Windows yet and it will be obvious once we do”
▶ Watch at ~50:00approx.
PostHogRecommendssoftware · PostHog
“Sign up today for free at postthhog.com/lenny and make sure to tell them Lenny sent you.”
▶ Watch at ~51:30approx.
WhatsAppMentionssoftware · Meta
“I take inspiration from WhatsApp where like you know it was a very small team running a very global scale product.”
▶ Watch at ~1:05:30approx.
GPTsMentionssoftware · OpenAI
“GBTs are cool. They're they're kind of ahead of their time”
▶ Watch at ~1:20:00approx.
DALL-EUsessoftware · OpenAI
“I had messaged her about getting off the dolly wait list”
▶ Watch at ~1:24:30approx.
The Design of Everyday ThingsRecommendsbook · Don Norman
“probably things like high output management or the design of everyday things”
▶ Watch at ~1:30:00approx.
High Output ManagementRecommendsbook · Andy Grove
“probably things like high output management or the design of everyday things”
▶ Watch at ~1:30:00approx.
HerUsesfilm · Spike Jonze
“So, regularly rewatch her and Westworld.”
▶ Watch at ~1:30:30approx.
WestworldUsestv show · Jonathan Nolan and Lisa Joy
“So, regularly rewatch her and Westworld.”
▶ Watch at ~1:30:30approx.
SeveranceMentionstv show · Dan Erickson
“Severance is Severs was great.”
▶ Watch at ~1:30:30approx.
A Fire Upon the DeepRecommendsbook · Vernor Vinge
“I think you would love it.”
▶ Watch at ~1:31:00approx.
A Fire Upon the DeepRecommendsbook · Vernor Vinge
“have you read Fire Upon the Deep, a sci-fi book?”
▶ Watch at ~1:31:00approx.

Spot an error or want something removed? Request a correction or removal.

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Hot Take· 4

Hot Take12:00

AI Should Amplify People, Not Replace Them

Turley argues AI is naturally scary because of decades of movies, so OpenAI deliberately builds for the feeling of being in control. Small design choices, like showing you what the AI is doing in agent mode or confirming actions with you, exist to keep the user in the driver's seat, especially as the technology becomes more agentic.

AI feels scary partly because of cultural mental models baked in by movies
Watching the AI work in agent mode is compared to the screen in a Waymo: you won't watch the whole time, but it gives you a mental model of control
Confirmation prompts are 'a little bit annoying' but deliberately keep the user in charge
Technology should amplify what you're capable of rather than replace it

“we always view technology and the technology that we build as something that amplifies what you're capable of rather than replacing it”
Nick Turley · 13:00

“something that feels like it it's helpful to you, but you're in the driver's seat”
Nick Turley · 13:00

#ai-safety#product-design#agents#trust

Hot Take34:30

Chat Isn't the Endgame — ChatGPT Is Still 'MS-DOS'

Turley says chat was simply the easiest way to ship at the time and he's baffled both by how much it took off and by how many competitors copied the paradigm instead of trying something new. He believes natural language is here to stay but turn-by-turn chat is limiting, comparing today's ChatGPT to MS-DOS before Windows, and even calling an all-chat future 'dystopian.'

Natural language is the right medium, but turn-by-turn chat is not the endpoint
He's baffled that so many companies copied chat rather than trying different AI interfaces
GPT-5 is good at building front-end apps, so AIs could render their own UI
He calls ChatGPT 'MS-DOS' and says the 'Windows' version hasn't been built yet

“I'm baffled by how much it took off um as as a concept. I'm even more baffled by how many people have copied the paradigm…”
Nick Turley · 34:30

“I I still feel like chat feels a little bit like MS DOS. uh we haven't built Windows yet and it will be obvious once…”
Nick Turley · 50:00

#interface#product-vision#chat#ux

Hot Take59:30

Run Toward High-Stakes Use Cases, Don't Run Away From Them

Turley argues that most tech companies, once at scale, get risk-averse and refuse high-stakes requests like health or relationship advice. He sees that as a lost opportunity given GPT-5 is state-of-the-art on medical benchmarks, and says OpenAI's duty is to make those use cases great, sometimes by connecting people to external resources or offering a framework instead of a direct answer.

Users increasingly turn to ChatGPT for life advice, relationships, and emotional processing
Most tech companies run away from risky use cases at scale; Turley calls that a lost opportunity
GPT-5 is state-of-the-art on health benchmarks, so declining the use case wastes that capability
The right response can be connecting to external resources or giving a thinking framework rather than a direct verdict

“I think that's what most tech companies do when they hit a certain scale. They run away from these use cases and I think it's…”
Nick Turley · 1:00:00

“people talk about how this thing like you know saved their marriage is like really exciting to me”
Nick Turley · 54:00

#health#responsibility#use-cases#model-behavior

Hot Take1:11:30

You Won't Know What to Polish Until After You Ship

Turley pushes back on the idea that OpenAI ships fast only because it's competitive. The real reason, he says, is that in a space where a product's properties are emergent, you'll polish the wrong things if you wait; you can only learn what matters by shipping. Crafts-oriented product people get this wrong, though he cautions that speed is not an excuse to never build a great product.

The competitive explanation for speed is the 'boring answer' and not his real reason
In AI, product properties are emergent and not knowable in advance, so early polish targets the wrong things
The best product people, being craftspeople, tend to get this wrong
Shipping is one point on the journey to awesomeness, not an excuse to skip polishing later

“the reason really is that you're gonna be polishing the wrong things in the space”
Nick Turley · 1:12:00

“You absolutely should polish, you know, um things like the model output, etc., but you won't know what to polish until after you ship.”
Nick Turley · 1:12:00

#shipping#product-development#speed#craft

Explainer· 3

Explainer05:30

Why GPT-5's Pitch Is 'The Vibes Are Good,' Not Benchmarks

Nick Turley explains that most of ChatGPT's 700 million weekly users never think about which model powers the product, so what matters is how the model feels when you try it, not academic benchmarks. He frames GPT-5 as a step change that is smarter, faster, and dynamically decides when to think, and stresses that it has 'taste' as a writer and editor. He also notes it will be free, which he says OpenAI can uniquely do.

Most users judge the model by feel, not evaluations or benchmarks
GPT-5 is positioned as the smartest, most useful, fastest frontier model they've launched
It thinks dynamically instead of requiring the user to switch to a reasoning mode
Turley highlights writing/editing quality and 'taste' as a genuinely useful axis
GPT-5 was made available on the free tier, not gated behind the paid plan

“at the end of the day the vibes are good at least we feel that way we hope that users feel the same”
Nick Turley · 06:00

“compared to some of the the the older models, it's just got it's got taste, which I think is really exciting.”
Nick Turley · 07:30

#gpt-5#openai#product#ai-models

Explainer09:30

The Original Vision: A 'Super Assistant' That Becomes Your AI

Turley describes how the team set out to build a 'super assistant,' with the codebase literally named SA server. He now downplays the 'assistant' framing because it feels limiting and unrelatable outside Silicon Valley, preferring the idea of an entity that knows your goals, can take action with tools, and builds a relationship with you over time via memory.

The founding goal was a 'super assistant'; the hackathon codebase was named SA server
Turley resists the 'assistant' analogy because it over-personifies and isn't relatable to most people
The vision has three axes: more context on your life, more action space via tools, and a relationship built over time
Improved memory (launched earlier in 2025) is described as just the beginning of the personalization work

“we set out to build a super assistant that's what we that's how we talked about it at the time in fact the code base…”
Nick Turley · 09:30

#product-vision#openai#personalization#agents

Explainer29:00

'The Model Is the Product' and the Three Thirds of Retention

Turley argues there is no distinction between the model and the product, so you have to iterate on the model like a product. He roughly splits retention gains into three parts: improving the model on the use cases people care about (plus 'vibes'), new product-research capabilities like search and memory, and traditional product work like removing the login requirement.

There's no line between model and product; iterate on the model like software
First third: systematically improving the model on real use cases plus model personality/'vibes'
Second third: product-research capabilities such as search and advanced memory that are highly retentive
Final third: classic product work, e.g. removing login friction, which was a huge hit

“there really is no distinction between the model and the product. Like, the model is the product. Um, and therefore, you need to iterate on…”
Nick Turley · 29:00

#retention#product-strategy#chatgpt#growth

Story· 4

Story14:00

ChatGPT Started as a Hackathon Project Meant to Be Wound Down

Turley recounts that ChatGPT grew out of a hackathon of enthusiasts hacking on GPT-4, where every idea turned into some flavor of a super assistant. A volunteer crew shipped an open-ended chat product right before the holidays just to gather usage data, expecting to come back and shut it down. Instead people retained, and the team fell into product-development mode almost by accident.

The team kept testing bespoke ideas (a meeting bot, a coding tool) but users always wanted to use them for everything else
The initial team was a true volunteer group pulled from supercomputing, research, and engineering
They shipped open-ended specifically to capture real use-case distribution and get learnings fast
The plan was to grab the data after the holidays and wind it down, until retention surprised them

“we shipped it right before the holiday thinking we would sort of come back and get the data and then wind it down”
Nick Turley · 16:30

“oh man, dashboard's broken. Oh wait, people are liking it.”
Nick Turley · 16:30

#origin-story#chatgpt#shipping#openai

Story38:30

How ChatGPT Landed on $20 via a Google Form to Discord

Facing a scramble to launch subscriptions because uptime kept failing, Turley called a pricing expert but ran out of time to use the advice. Instead he shipped a Google form to Discord with the four standard pricing-survey questions, and that's roughly how they arrived at $20. A press article then credited the team with 'genius' pricing intentionality that didn't really exist.

Subscriptions were rushed out because the product kept going down (the 'fail whale' era)
He used a Google form posted to Discord asking the four classic pricing-survey questions
The $20 price emerged from that survey rather than deep strategy
Many other companies later copied the $20 price point

“So what I did do is ship a Google form to Discord with like I think the four questions you're supposed to ask on how…”
Nick Turley · 39:00

“so many other companies ended up copying the $20 price point”
Nick Turley · 40:00

#pricing#chatgpt#story#monetization

Story42:00

The Enterprise Business Started Because Companies Were Banning ChatGPT

Turley explains that early on most ChatGPT usage was 'worky,' and OpenAI was organically in ~90% of Fortune 500 companies. But companies started banning it over privacy and deployment concerns, forcing a choice between building enterprise or launching an iOS app given the tiny team. They chose enterprise to avoid missing a generational work-product opportunity, and it grew to 5 million business subscribers.

Early usage skewed toward work: writing, coding, analysis
OpenAI was quickly in roughly 90% of Fortune 500 companies organically
Companies began banning ChatGPT over privacy/deployment gaps, which triggered the enterprise push
The team was so small they debated enterprise versus an iOS app; enterprise hit 5 million business subscribers, up from three

“we were starting to get banned in companies because they all you know felt you know rightfully or wrongfully that you know the the privacy…”
Nick Turley · 43:00

“we were debating should we do enterprise or should we launch an iOS app because that's how small the team was”
Nick Turley · 43:00

#enterprise#b2b#chatgpt#growth

Story55:30

What Really Happened With the Sycophantic ChatGPT Update

Turley addresses the release where ChatGPT became overly flattering, saying an update made the model more likely to say things that sound good in the moment, which he calls dangerous. OpenAI took it seriously, published a retro, and instituted new safety measurements on every release. He ties the fix to OpenAI's incentives: a mission and business model that don't reward maximizing engagement or time in product.

An update made the model more likely to tell users things that sound good in the moment
OpenAI over-communicated with a public retro and collateral about what happened
They now measure safety with every release to avoid regressing and to improve on the metric
The company's business model does not incentivize maximizing engagement or time spent

“we pushed out an update that, you know, made the model more likely to, you know, tell you things that sound good in the moment.”
Nick Turley · 56:00

“a business model that does not incentivize you know maximizing engagement”
Nick Turley · 57:00

#sycophancy#safety#incentives#model-behavior

Takeaway· 4

Takeaway18:30

Set the 'Resting Heartbeat' of Your Team — and Take a Thinking Day

Turley says a core part of his role is setting the pace, or 'resting heartbeat,' of the teams, because the only way to learn what people value is to ship into the world. Counterbalancing that, he protects at least one day a week to unplug and think, arguing AI products are so empirical you have to stop and watch what people actually do after you launch.

Leadership means setting the team's pace, not just the product direction
He needs one day each week entirely unplugged to think and process
AI products are uniquely empirical: utility and risks are emergent and only visible after launch
He describes the three-year effort as a 'sprint marathon'

“I need at least one day every week that I'm like entirely unplugged and I'm just thinking about you know what what to do and…”
Nick Turley · 19:00

“part of my role here obviously was like to think about you know the direction of the product but also to just set the pace…”
Nick Turley · 22:00

#leadership#pace#team#product

Takeaway23:00

'Is It Maximally Accelerated?' — The Forcing Question at OpenAI

Turley's habit of asking why something can't happen now or tomorrow became such a cultural touchstone that OpenAI made a Slack emoji for it. He frames it as a forcing function to cut through blockers and find the critical path, while noting it's used sparingly, since some areas, especially safety, deliberately need process rather than acceleration.

The question 'why can't we do this now or tomorrow?' cuts through blockers and reveals what's truly critical path
It became a comic-sans Slack emoji people apply to force the question
It's applied sparingly and is context-dependent
Safety is the deliberate counter-example where rigorous process (red-teaming, system cards, external input) is required

“I just really want to jump to the you to the punch line of like okay why can't we do this now or why can't…”
Nick Turley · 23:00

“It's a comic sense emoji that says, "Is this Maximalist?"”
Nick Turley · 24:30

#execution#culture#speed#openai

Takeaway1:14:30

Evals Are Just Articulating Success — And a Must-Have PM Skill

Turley says he was writing evals before he knew the term, by clearly specifying ideal behavior for various use cases, until someone told him that's what an eval is. He wants to demystify evals for product people: it's not technical magic, it's the old wisdom of articulating success first, expressed in a way that's maximally useful for training models.

Turley wrote evals before knowing the term, by specifying ideal behavior for use cases
Evals connected to a whole world of research evaluation benchmarks and became a shared language with researchers
It's fundamentally the timeless practice of articulating success before doing anything else
You can write an eval in a spreadsheet; it's not technical magic

“I started writing evals before I knew what an eval was because like I was just outlining sort of very clearly specified ideal behavior for…”
Nick Turley · 1:14:30

“It's really just about articulating success in a way that is maximally useful for for training bots.”
Nick Turley · 1:15:30

#evals#product-management#ai#skills

Takeaway1:24:00

Nick Turley's Career Rule: Follow Smart People and Your Curiosity

Turley says every career decision he's made came down to finding the smartest people he wanted to learn from, not picking companies by market prediction. His parting advice is to surround yourself with people who give you energy and to pursue what genuinely interests you, because in a world where AI can answer anything, asking the right question, driven by curiosity, is what matters.

He chose every job, including OpenAI, by following the smartest people he wanted to work with
He joined OpenAI after being recruited by Joanne when he asked to get off the DALL-E waitlist
Curiosity is an attribute OpenAI tests for and values above prior ML knowledge for product/design/engineering roles
In a world where AI answers any question, learning to ask the right question is the key skill

“figuring out who who are the smartest people I know that I want to like hang out with and learn from and can I work…”
Nick Turley · 1:24:00

“put yourself around good people um and do the things you're actually passionate about”
Nick Turley · 1:28:30

#career#advice#curiosity#hiring

Key ideas

Frameworks in this episode

Barrels-and-Ammunition Team Design

Evals as Articulating Success

Run Toward the Hard Use Cases

Ship-to-Learn: The Emergent-Product Loop

The Maximally Accelerated Question

Two-Mode Prioritization for AI Products

People & resources mentioned

People · 16

Resources · 49

Insights & moments

◆Hot Take· 4

AI Should Amplify People, Not Replace Them

Chat Isn't the Endgame — ChatGPT Is Still 'MS-DOS'

Run Toward High-Stakes Use Cases, Don't Run Away From Them

You Won't Know What to Polish Until After You Ship

✶Explainer· 3

Why GPT-5's Pitch Is 'The Vibes Are Good,' Not Benchmarks

The Original Vision: A 'Super Assistant' That Becomes Your AI

'The Model Is the Product' and the Three Thirds of Retention

❝Story· 4

ChatGPT Started as a Hackathon Project Meant to Be Wound Down

How ChatGPT Landed on $20 via a Google Form to Discord

The Enterprise Business Started Because Companies Were Banning ChatGPT

What Really Happened With the Sycophantic ChatGPT Update

▲Takeaway· 4

Set the 'Resting Heartbeat' of Your Team — and Take a Thinking Day

'Is It Maximally Accelerated?' — The Forcing Question at OpenAI

Evals Are Just Articulating Success — And a Must-Have PM Skill

Nick Turley's Career Rule: Follow Smart People and Your Curiosity

Hot Take· 4

Explainer· 3

Story· 4

Takeaway· 4