AI at Work, Spring 2026
Millions of entrepreneurs building with ChatGPT
Hi
Welcome (back) to The Prompt. AI isn’t well understood, but we learn a lot in our work that can help:
One month, 4 million+ ChatGPT-supported entrepreneurs
Comparing job risk with worker concern
People want AI in more hands, not fewer
Department of Labor’s text-based AI literacy program
If you find this helpful, make sure you’re signed up for the next issue. Also be sure to join the energy-projects experts from NVIDIA, Oracle and OpenAI this Thursday for a discussion about the strategic importance of compute. More below.
[News] Millions of entrepreneurs building with ChatGPT
More than 3 years in, we can now see clear indicators that ChatGPT is driving productivity gains, yes – but it’s also fueling entrepreneurship, giving people the tools they need to develop and launch a wide range of new businesses.
We see this in how small companies are using our tools. OpenAI Economic Research conservatively estimates that at least 4 million Americans who were starting, running, or planning a business in March 2026 did so with help from ChatGPT. Use spanned sectors from retail to professional services, trades, hospitality, and health. Much of the activity was concentrated in service-based businesses, where founders often start lean and do everything themselves.
Separately, we also see a large cohort of business customers that have fewer than 10 employees – another strong signal that our tools aren’t only serving established companies, but helping new ones take shape, launch, and compete.
With AI, small businesses increasingly have access to capabilities that used to be the preserve of large, well-capitalized companies: market research, customer support, legal and compliance help, product development, data analysis, design, and operations support. An entrepreneur can now test an idea, build a brand, understand customers, and run day-to-day workflows with a level of precision and insight that would have been hard to imagine a decade ago.
A few examples show what this looks like in practice:
Matt Rosenberg used ChatGPT to test whether his Thai restaurant could work before signing a lease, then kept using it for permits, pricing, reviews, social posts, margins, and daily operations after Bangkok Rush opened.
Kaija Pack uses ChatGPT as an extra operator for Break Life Houston, the world’s largest rage room (yes, check it out), helping with competitor research, outreach, customer communication, pricing explanations, and ideas to bring people in on slower weekdays.
Alison Leeds used ChatGPT to turn thousands of post-surgery recovery stories into weekly art-and-color cards, building a recovery-support design business for breast-cancer patients and caregivers.
Rob Cahill’s 10-person biotech company Junevity gives ChatGPT to every employee and uses OpenAI models for literature review, genomics, target selection, and operational planning.
With our tools already lowering the costs of starting, experimenting, and operating a business across sectors, a wide range of entrepreneurs are increasingly able to build sustainable businesses on their own terms. – Adam Cohen, Head of Economic Policy
Last November, OpenAI partnered with DoorDash, SCORE, and 16 regional business-support organizations to host over 500 small business owners in a multi-city hackathon. You can read about it here.
[Data] Comparing job risk with worker concern
Workers with jobs at higher near-term risk of automation are no more concerned about losing their jobs than most other workers, according to our latest research. Where their concern about their heightened risk shows up is in greater demand for controls and agency over how AI is deployed at their place of work.
We paired OpenAI Economic Research’s new framework for how different jobs will transition in the age of AI with a 5,060-sample survey via TrueDot of how people facing various degrees of AI-driven job risk feel about their position in the workforce. We organized respondents by the 900+ occupations that were classified in the framework, then cut the results based on the framework’s four identified risk categories of: 1) jobs facing less immediate change due to AI, 2) jobs that will be reorganized because of AI, 3) jobs that will see more demand because of AI, and 4) jobs at higher risk of automation by AI.
The results offer a look at how actual risk of job automation and concern about job automation align, and don’t.
Percentage of adults selecting each response, by job-risk category:
Workers in roles at higher risk of automation are in line with those in less at-risk roles in seeing their position as likely to still require people to do the work over the next two to five years, and are no more concerned than other groups about there being “fewer hours, fewer shifts, fewer job openings, or fewer workers needed overall.”
But because they are aware of how their roles are at risk, they are much more likely to want to see controls and have agency over how AI is used at their workplace. For example, they are significantly more likely to say that the best AI tools should be put into the hands of “people like me, so more people can benefit directly,” rather than given first to big companies that may be more ready to deploy AI.
It’s actually people in jobs that will see more demand with AI who are most concerned that AI may reduce the need for people in their roles (51% concerned). We think there are two reasons for this, despite the research suggesting they’re in a stronger position. First, this group is the most exposed to AI at work and at home (72% are weekly AI users, 42% use AI multiple times a day), and likely are more familiar with advancements in AI capabilities since the introduction of ChatGPT. Second, one in three of these workers are in technology-related fields, a sector that has seen a drumbeat of layoff announcements in the last couple of years, often vaguely attributed to AI.
Those in jobs that will see more demand also show much greater support for “workers getting a share of the gains if AI makes their workplace more productive”: 25%, compared to 16% among gen pop. These are people who are experiencing how AI can improve productivity and grow companies, and they want to participate in that success.
People in roles classified as less likely to see near-term change report lower concern than other groups.
[Insight] People prefer democratized AI
Perhaps unsurprising but still worth noting. The same 5,000-person survey finds strong majorities believing that AI should be free and accessible to all, and that free and accessible AI is safer than tightly controlled AI (an idea explained in OpenAI’s recent policy blueprint to democratize cyber defense).
[Resource] AI literacy, one text at a time
As AI reshapes work, government needs to prepare and respond. That’s the thinking behind the Department of Labor’s AI 101 course, a free text-based program designed to help more people build AI literacy and confidence. Taylor Stockton, the department’s Chief Innovation Officer, explains the thinking behind the program. This conversation has been edited for space and clarity.
How do people sign up?
They can text READY to 20202.
Why launch this now?
A lot of Americans are still on the sidelines. Some are unsure about AI, some are skeptical, and some are fearful. We wanted to create an easy first step that helps people feel more comfortable and start seeing how these tools could be useful at work and at home.
Are you hearing that uncertainty in your own life too?
Absolutely. In my own family, some people are excited, some are skeptical, and my mom wasn’t using AI at all.
Why text messages?
Because it keeps the friction as low as possible. You can sign up with a single text, and most of the experience happens over text too, which makes it more accessible for people who may have limited broadband or internet access.
What happens after that?
It’s a free, one-week course delivered by text: about 10 minutes a day of bite-sized learning and challenges.
What do people actually learn?
Each day focuses on a different part of the Department of Labor’s AI literacy framework. There are short lessons on how AI works, how to prompt effectively, how to evaluate outputs, and how to use AI responsibly. There are multiple-choice questions, small challenges, examples from both work and home life, and links for anyone who wants to go deeper.
Is it interactive?
Yes. It’s designed as a back-and-forth experience. The goal is not just to give people information, but to help them reflect on how AI fits into their own job or daily responsibilities.
What would success look like?
We had 10,000 Americans sign up in the first 24 hours, and would like to reach more than 1 million Americans. But it’s not just about volume. We also want to see whether people become more confident using AI and better able to connect it to their work and daily lives.
And finally: your mom. Where does she stand now?
She’s halfway through the course. She may be the toughest critic of all, because her skepticism was high to start. I’ll be interested to see if we’ve moved her a few notches by the end.
[Event] Compute, competitiveness, and country
Compute powers every layer of AI: frontier research, model performance, products, deployment, and revenue. When we announced our Stargate AI infrastructure project in January 2025, we set out to secure 10 gigawatts of AI infrastructure in the United States by 2029, and we’ve already identified over 8GW of that. Our available compute grew roughly 3x year-over-year from 0.2GW in 2023, to 0.6GW in 2024, to ~1.9GW in 2025.
Given compute’s importance to US competitiveness on AI, and to our own ability to continue to make AI tools freely available so that hundreds of millions of our regular users can create opportunity for themselves and others, the OpenAI Forum is putting compute center-stage on Thursday.
Join infrastructure experts Nick Edwards of OpenAI, Dion Harris of NVIDIA, and Drew Bryck of Oracle for a discussion about how investments in computing capacity and related infrastructure can keep the US the global leader in the next era of technology, and what it takes to build this future responsibly in partnership with local communities.
For those interested, we’ll have a live audience in the room in DC and a livestream for the public; as always, Forum members will have the chance to submit questions.
4:00 PM - 6:00 PM EST on May 7
[Disclosure]
Graphics created by Base Three using ChatGPT.
You don’t work