How OpenAI's Head of Business Products Uses ChatGPT to Save Time at Work | Nate Gonzalez
Inside OpenAI's product team: how PMs use ChatGPT, their approach to building products, hiring criteria for new PMs, and overcoming AI adoption barriers
Dear subscribers,
Today, I want to share a new episode with Nate Gonzalez.
Nate leads ChatGPT for Work which is now used by 92% of Fortune 500 companies. In our chat, he reveals how OpenAI runs with <30 PMs, what they look for in new hires, how he personally uses ChatGPT to save time at work, and more.
Watch now on YouTube, Apple, and Spotify.
Nate and I talked about:
(00:00) 92% of Fortune 500 companies already use ChatGPT
(02:06) OpenAI's latest features for ChatGPT at work
(14:52) Why OpenAI has less than 30 PMs for 5,000 employees
(15:58) What traits OpenAI looks for when hiring PMs
(18:31) The 10-minute AI hack that changed how Nate works
(25:56) The most surprising thing about working at OpenAI
(29:48) The biggest barriers to AI adoption and how to overcome them
(38:36) Using ChatGPT roleplay to prep for important meetings
(41:04) ChatGPT's future: From assistant to trusted coworker
(43:06) The specific skill that will keep your job safe in the AI era
This free post is brought to you by... Product Faculty
Product Faculty has partnered with Miqdad Jaffer (product lead at OpenAI) to teach the top AI PM certification on Maven. You can get 25% off Miqdad’s AI PM course with my link below. You’ll learn how to solve real problems with AI, dive into technical concepts, and build real AI prototypes.
Why OpenAI has <30 PMs for 2,000+ people
So I think OpenAI has <30 PMs for 2,000+ people. Why is the PM team so lean?
We want to be the model of what it looks like to build a company on top of AI.
That means using AI to extend every employee. Our PM team is lean, our engineering team is also relatively lean for the size and scope of the business. We're leaning into AI every single day — what can we do better by working with the models directly?
What traits do you look for when hiring PMs at OpenAI?
We look for several things:
Agency. You need high grit and a willingness to work on hard problems.
Product sense. You need to balance user fit with business considerations and be able to creatively brainstorm and justify solutions.
Mission alignment. You need to tie your product back to impact on our mission.
Execution. You need to have a high sense of urgency.
Curiosity. ML experience is great, but we also look for signals that you’re an AI builder and know how the ecosystem is evolving.
Connectors: Extending ChatGPT to support internal knowledge sources
I’d love to get an inside look at how you build products at OpenAI. You just launched Connectors to let companies connect their internal sources like Google Drive, Sharepoint, and Box. How did you build this from ideation to launch?
Models get smarter with more context, and internal context is critical for businesses.
We were initially thinking in the GPT-4 paradigm — quick call-response where you need to sync and index repositories for low latency.
But reasoning models changed everything. The latency constraint relaxed because the model has multiple turns to get the right answer. It could form hypotheses, look at variants, and pull them together. This shift allowed us to scale connectors more quickly and accurately.
How do you decide which internal documents to load into the model’s context? Some of them might be outdated or low quality.
Great question.
We did extensive post-training on recency and "seniority of authorship" — basically a social graph to surface the most relevant content.
Think about someone new onboarding at OpenAI. There might be 100 documents written about a specific subject. We spent tons of time making sure the most relevant documents get pulled forward.
Record mode: How OpenAI approached AI meeting notes differently
The most relevant information internally is often said in meetings. Tell me about how you approached Record Mode.
Sure. Record mode rounds out the knowledge picture. Beyond pre-trained knowledge, web search, and internal documents — there's all the information spoken in meetings that never gets properly recorded.
We created a generalized recording capability to model that information like any other knowledge source. I can ask "What did Peter and I talk about two weeks ago?" and get a timestamped summary where each action item links to the transcript.
So one of the most important skills for PMs is to learn how to write AI evals. How did you evaluate whether Record Mode was good enough to launch?
First, you need to align on the North Star metric, measure it, get a baseline, and hill-climb quality against that baseline.
For Record Mode, we look at transcription accuracy and collect qualitative feedback directly from users on whether summaries are good or bad. We also track signals around follow-up questions — are we being clear and returning accurate information?
We don't aim for perfect before launching. That's not our ethos. We set a high quality bar but prioritize getting to user signal quickly — that's where evals really matter.
An example of OpenAI’s bottoms-up culture
Can you share an example of how ideas can come from the bottom up at OpenAI?
Sure, Canvas is a perfect example. An IC researcher pitched the Canvas idea in her 1st month at the company — I think around July 4th break.
Her manager agreed immediately and staffed 5-6 engineers. The team formed organically — "Hey, there's a really interesting idea, we think it's super high leverage, who wants to work on this?" People gravitated to it saying "I want to work on that problem."
Canvas became our first major UI update to ChatGPT. We went from a basic chat interface to a much richer experience, all from a single individual whose idea wasn't part of any specific product roadmap.
The best idea wins. It doesn't really matter where it comes from within the company.
The biggest misconception of how OpenAI builds products
How do you balance this bottom-up culture with planning?
We run a quarterly planning process. The reality is as soon as you wrap the plan, it's out of date and you're really using it as a trade-off framework.
We try to minimize the process around roadmapping because it's such a fluid, constant iteration. There are so many things we could do that we have high conviction in. How do we focus on the ones with highest impact?
This necessitates talking to customers. I probably have 4-5 customer conversations directly each week and very specific themes emerge. We also work closely with our go-to-market team who are sitting with customers every day, with multiple feedback loops to get that information back to product.
What’s the biggest misconception of how OpenAI builds products?
One big misconception is that moving quickly = cutting corners, particularly around safety.
Our team is deeply mission-oriented. When we think about what makes a successful PM at OpenAI, they're 100% focused on what impact they're driving. This pushes urgency to ship quickly but also to ship responsibly.
There's a whole bunch more that we could just ship, but we are actually holding that bar very deliberately to make sure it's of the necessary quality. We'll hold if there's safety evaluation work needed before something goes out.
You might think of urgency and safety as opposing constructs, but in reality they work together through our culture of urgency and how we drive impact.
How Nate uses ChatGPT to save time at work
How do you personally use ChatGPT to save time at work?
It saves me so much time in these three tasks:
Internal research. Getting up to speed on projects in our research org or technical implementations. I can onboard context much faster without endless meetings pulling time from other teams.
External research. We serve 92% of the Fortune 500 across every industry. I use AI to understand their company and context so we can map our products to their needs more quickly.
Roleplaying with voice. This is huge. Before talking to a candidate I'm keen on or a critical customer interaction, I roleplay with ChatGPT. I'll have it assume their personality so I can hone my craft and get crisp on messaging.
I love the roleplaying example. Do you upload context to make it act like specific people?
Yes, I upload context from connectors and record mode meeting summaries into a project. Then voice mode has background context to play the role or be a thought partner for me.
I also often use ChatGPT to critique my work: "Here's my initial draft. What am I missing? Where could this be stronger? What are the weakest parts of this argument?" It helps with both drafting and actual quality of output.
The biggest barriers to employee AI adoption and how to overcome them
What are the barriers to getting employees to use AI in companies?
Companies often ask "who do I give AI to? My most technical employees? My early adopters?" But that creates cynicism — those with the intuition and those without. Access is key. You need everyone building intuition about what AI is good at now, what it still struggles with, where it's headed. That's how OpenAI moves so quickly — we use our tools every day, so reaching for ChatGPT becomes second nature.
The big trend driving success is internal AI champions.
These champions aren't necessarily CEOs. They're heads of AI, heads of product divisions, often CIOs who want transformational change. We work with them directly to identify the highest value use cases.
What advice would you give someone trying to drive AI adoption at their company?
Two things:
Broad deployment. Get this in every employee's hands so they become fluent. You want them familiar with these tools because that drives bottom-up culture.
Focus on specific use cases. Find one or two bets that will drive outsized value for workflows in your company. These aren't just tools for tools' sake — identify the highest leverage product opportunity and push forward.
Moderna is a great example. They have thousands of GPTs deployed internally. People create GPTs and share them with colleagues, so everyone gets the collective benefit of that knowledge work.
The future: ChatGPT as your trusted coworker
We're moving from AI as copilot to delegating work to AI agents. How will this change the PM role?
Our goal is to extend your productivity by making ChatGPT your virtual coworker.
Imagine waking up, sitting down with ChatGPT in a more personalized interface with a list of tasks that have come in. You delegate some to ChatGPT — "bring these back when done" — while you tackle the top priorities.
Behind the scenes, that could mean orchestration with different agents. But focusing the interaction through ChatGPT reduces cognitive load.
So is it like having a bunch of AI interns to work for you?
Well I think AI agents will be more than interns — after all, they can do PhD-level math and deeply understand any code.
In 2022-2023, it felt like an intern. But with reasoning models and better UI paradigms, it'll feel much more like a coworker that you trust to get work done.
Closing advice for people who want to level up their AI skills
What's your advice for people who want to level up their AI skills?
It's not just "go try the tools." It's how do you make these tools an extension of the way you work.
Don't just learn to write emails faster. Learn to write better emails. Use AI to improve the quality of your thinking and process.
Find quality improvement loops, not just productivity loops.
Ask it to critique your work, point out weak arguments, identify logical fallacies. That's how you actually improve.
If I want to use ChatGPT officially at work, where do I start?
Get started quickly with ChatGPT Team. We just enabled SSO so it's easy for businesses to self-serve.
As you need more advanced compliance and want to work with our go-to-market teams on use cases, consider ChatGPT Enterprise. Both products have the same privacy guarantees — we never train on your data.
Thank you Nate! If you enjoyed this interview, follow Nate on LinkedIn and check out ChatGPT Team or Enterprise for your company.
The criteria for hiring PMs at OpenAI reminded me of a research piece by Linda Hill from HBS on "leading innovation." She argues that successful innovators create a space for collective genius rather than controlling it. What you’re describing — agency, urgency, product sense, curiosity — sounds like you’re hiring for that exact space. That alignment between individual drive and collective direction is what most orgs struggle to operationalize.