Hayden Brown, president and chief executive of Upwork, tells Bob Safian what her own marketplace shows about a labor market the monthly jobs report reads as sluggish: AI skills booming inside an otherwise flat hiring picture, full-time roles giving way to fractional work, and clients who moved work to agents now bringing people back. They also get into her May layoff, the agents now posting job ads on her platform, cheap open-source models out of China, and how she plans a business when a one-year roadmap no longer survives eleven months.
👤 Guest: Hayden Brown, president and CEO of Upwork, the global tech platform for freelance workers
🎙️ Host: Bob Safian, who hosts Rapid Response and was editor-in-chief of Fast Company
📰 Published: 31 August 2026 on Pioneers of AI
🔴 YouTube | 🟣 Apple Podcasts | 🔗 Episode page | ⏱️ 30 min | ✅ Time saved: 6 min
Key Takeaways
The government numbers and her platform agree on the direction and disagree on where the money is
"Yeah, there are a lot of competing dynamics. It certainly doesn't tell the whole story."
Small business customers have "pulled back hiring across all channels", while the AI category on the platform grew "over 22% year-over-year"
Work is moving from full-time headcount to fractional engagements, and fast
"Our most recent survey showed a 10-point increase in freelancing in the US within knowledge workers. It's at 38% of the population today. It was at 28% just a year ago."
This is the same creation-and-destruction cycle as before, run harder in both directions
"It's like more destructive on the negatives, but also more creative on the positives."
"We do not see mass job destruction as the future."
She cut a quarter of Upwork's corporate staff and refused to hand AI the credit for it
The redesign question was what each function would look like built from the ground up today, not which workers a model could replace
"But that is not the biggest reason that we made the changes nor the biggest benefit that we were able to get from and kind of driving these changes."
Clients who moved work to AI are moving it back, and agents rarely finish a task on their own
"They almost never succeed in delivering even fairly simple tasks from clients end to end."
Add an expert and a few exchanges and "suddenly the success rate goes up more than 70%"
Machines are now hiring humans, and Upwork made them authenticate first
The MCP server is "a capability to let agents authenticate with Upwork", so good actors and bad actors can be told apart before they post a job
It is also a rescue channel for a client stuck inside Claude or ChatGPT with a deliverable that missed
Verified work history is the asset AI cannot fake
"So this is the most verified work history data really in the world and it's at scale across 18 million freelancers that were active in the last year."
"Not noise that was created by an AI agent applying for them."
Open-source models are a gateway decision, not a loyalty test
"It's not a binary of use Claude or good luck to you and there's nothing you can do."
Rising token costs from the big US providers have opened up cases worth "10x savings or more"
Freelancing is both a choice and a transfer of risk onto the individual, and she says so plainly
"Absolutely both. 58% of current full-time employees are actually thinking about becoming freelancers for a lot of the reasons you mentioned around choice and control and deciding when, where, and for who they work."
"Gen Z is the biggest one: 42% of Gen Z are freelancing."
AI fluency now carries a measurable wage premium on the platform
"Freelancers that have AI skills are earning 34% more per hour than folks who don't have AI skills"
Companies hire those freelancers partly to upskill their own teams by proximity
Nobody becomes capable with AI in a seminar, which is why a new generalist is emerging
"I was actually talking to one of our clients this morning and he said, you can't learn to ride a bike in a seminar."
"And that's one reason we're seeing the rise of a new generalist called the AI orchestrator."
The one-year roadmap is the first casualty of a world where the shocks come from outside
"Now in product, if you have a one-year roadmap and really you're executing the same thing in month 11 that you thought you would be when you started, something's probably wrong"
"It's kind of like let's take the first step and then re-evaluate what is that second step."
The humans-versus-machines argument is itself the obstacle
"The binary debate that's happening around will humans win or machines win, I think that is distracting us from, in some cases, putting our collective societal energy towards building a future where everyone wins"
A Black Swan Every Quarter, and Nobody Knows Until It Has Passed
Safian opened by placing Brown's last appearance on the show: they had talked geopolitics, Russia and Ukraine, and it was before the Great Resignation, before the return-to-office fights, before Gaza and Iran. AI anxiety now dominates workforce conversations, he said, and it feels like decades of change piled together. He asked how she was doing.
Running a company now means budgeting for a shock every quarter. "I have a kind of running joke with some friends who run other companies."
"We all say, what's going to be the Black Swan event of this quarter", she said, because so much is happening in the ecosystem and things are moving faster and faster
She called it a challenging time to be a chief executive and, in the same breath, very exciting
Safian pushed the joke one step further: "Like, you don't even know if it's a black swan until it's already moved on." Brown agreed
The Jobs Report Is Not Wrong; It Just Misses Where the Money Is Moving
Safian said he wonders a lot whether traditional assessments capture the reality of the US labor market in 2026, and put the latest monthly print to her: "It's like the most recent monthly jobs report showed 23,000 jobs lost but at the same time unemployment numbers went down also." He asked how she makes sense of that, how much she focuses on it, and whether it matches what she sees on Upwork.
Her answer was that both halves are true and neither is the whole picture. "Yeah, there are a lot of competing dynamics. It certainly doesn't tell the whole story."
Consistent with the jobs report, she said, the labor market is very subdued, has been for several quarters and continues to be sluggish
The slowdown is loudest among the smallest customers. "You know, those are the guys who are out on the front lines of the economy without a lot of resourcing, without deep pockets and financial reserves."
"And what they tell us is they have pulled back hiring across all channels."
Inside a Flat Market, AI Skills and Fractional Work Are the Only Places With Energy
Safian asked whether Upwork's recently released future of work index captures this, and what else in it the business headlines are missing.
The market is slow in aggregate and hot in patches. "You know, while the picture overall is of this kind of slow-moving market, there are pockets within it that are incredibly vibrant."
"We really see the growth in AI-related skills, AI-related jobs. That is the place where energy and money is flowing."
On the platform itself: "And we see that in our platform with our own AI category growing over 22% year-over-year."
Demand for people who can help a business navigate the new tools runs from small companies right up through large enterprises, she said
The second shift is in the shape of the engagement rather than the skill. Work is moving from full-time toward fractional, which she called a really fascinating trend
"Our most recent survey showed a 10-point increase in freelancing in the US within knowledge workers. It's at 38% of the population today. It was at 28% just a year ago."
Workers moved into freelancing for a variety of reasons, she said, and businesses moved toward contingent talent because they no longer need the same number of full-time employees and because some of these people cannot be reached through full-time hiring channels at all
Jobs Have Been Destroyed and Created Before; This Cycle Runs Harder in Both Directions
Safian laid out the contradiction in the public argument: some layoffs are attributed to AI, while other chief executives say AI lets them add staff because each AI-enabled person is more productive. Both seem to fit what she sees, he said, and it seems early to know which wins.
Her historical anchor: "You know, I think it's 60% of jobs that exist today did not exist in 1940."
Technology and other forces have destroyed and created jobs again and again through many cycles, she said, and this is a version of that happening bigger and faster
The asymmetry cuts both ways at once. "It's like more destructive on the negatives, but also more creative on the positives."
She rejected both camps. One says AI takes all the jobs and anything human is doomed; the other says the opposite. "The reality has to be somewhere in the middle."
"We do not see mass job destruction as the future."
What she does expect is a lot of new jobs, roles and businesses born because of the technology
She Cut a Quarter of Upwork's Corporate Staff and Would Not Let AI Take the Credit
Safian raised the layoff directly: "You announced in May that you had to cut a quarter of Upwork's corporate staff yourselves." She published a frank statement owning the decision, he said, and talked about AI without blaming it. He asked why the cut was necessary.
She framed it as a profitability decision taken under changed market conditions. "I made that decision earlier this year because we really did see that with the market conditions that were changing, this was the right time for us to expand the profitability of the business, which was already part of our long-term goals."
The method was not swapping workers for models. "Instead, we looked across the entire business and asked ourselves in every team, and every department, what would this look like if we really were doing this function from the ground up today using the best technology and tools that are available?"
Alongside that they flattened teams, widened managers' spans of control and cut layers — changes she called evergreen opportunities to refine how the company operates
Safian named the incentive to lie about it. Some chief executives do AI washing, he said: they ascribe changes to AI rather than owning that the strategy was wrong, and the stock price goes up for it. He asked whether she had been tempted to lean into that messaging because it helps the story
She said the messaging had to be true, and that AI was not the main cause. "You know, it was so important to me that our messaging around this be really true and authentic because frankly, of course, AI is helping us and it is a contributor to our efficiency and our productivity."
"But that is not the biggest reason that we made the changes nor the biggest benefit that we were able to get from and kind of driving these changes."
Clients Who Moved Work to AI Are Bringing Humans Back, and Agents Rarely Finish a Task Alone
The more interesting number, she said, is the traffic going the other way. AI washing has been happening, but a lot of companies are reconsidering, because the humans they let go turn out to still be needed and the technology cannot do everything they hoped.
"A recent survey we did showed that 23% of our own clients have actually already moved work back to humans from AI or are about to do so."
Upwork has tested agents doing the work end to end, and the results are poor. "When we've actually done testing of clients trying to use agents directly to deliver work products through our platform, the failure rate of agents is incredibly high."
"They almost never succeed in delivering even fairly simple tasks from clients end to end."
"But when you just add a little bit of human expertise and a few back and forths with an expert, suddenly the success rate goes up more than 70%."
She expects the swing to continue rather than settle. She called it a pendulum that will keep swinging, and said the skill people are developing is knowing when to rely on the machine: "We have to sharpen our intuition around when to trust AI."
Increasingly, she said, people are deciding not to take an AI output at face value and to pair it with a human expert
Machines Are Hiring Humans Now, and Upwork Built the Front Door
Safian pointed out that AI is changing the platform itself: "There's been a rise in AI agents posting job ads machines essentially trying to hire humans." Upwork even sells a way for clients' agents to reach its freelancer database, he said, and asked whether that felt spooky at first.
She treated it as arrival rather than novelty. "I mean, it really feels like the future of work is arriving and it's arriving on our shores right now, right here, whether we want it to or not."
The product answer was identity. "And that was one of the reasons we launched this what's called the MCP server, but really is a capability to let agents authenticate with Upwork."
"So we know who they are. We know if they're good actors or bad actors and then based on that authentication they can go forward and help post jobs, make hires, manage projects, etc."
She was explicit that this is a response, not an invention: the world was already going there
The same connection works as a rescue channel when a chatbot stalls. "And at the same time, we're seeing a lot of clients who are working inside these AI tools, Claude or ChatGPT, and they're getting stuck and they're having a deliverable that didn't meet expectations or they need a human expert to finish it."
With the MCP connection they can share that deliverable with a freelancer immediately and get help in real time, which she called a huge advantage now that AI has reset what people expect of response times
AI Slop Is Flooding Both Sides of the Hiring Process, and Verified Work History Is the Counterweight
Safian turned to Upwork's own agent: "Upwork has its own agent UMA that assists users writing job descriptions and prepping interviews and things like that for a platform conceived to give humans more freedom." There is irony in using AI to prepare an application so another AI can review it, he said, and asked whether the best applicant is now simply whoever prompts the hiring machine best — or whether this is more like a dating app, where the algorithms match better than the people do.
She accepted the premise and named the problem. "I mean these are the big questions and they're the right ones because we are seeing spam job proposals and applications."
The stated intent is amplification plus filtering. "We really just want to put people's own interests and needs on steroids and then have that expressed through the hiring process and the recruitment process that we're supporting them with, including filtering out all of the AI kind of slop and noise that is creeping in now that these tools are so widespread."
Safian described the mechanics from the applicant's side: he can now generate many versions of an application for many jobs that may or may not suit him, without even knowing what he is doing, and the same volume problem runs in reverse on the posting side
Her answer to the volume is evidence. "And I think that's one of the main values that we see being so enduring even in the AI era for Upwork specifically and our customers is the problem we've always solved for them is giving them real tangible accurate data around who can do what."
It is data from the platform observing people doing work and delivering for customers with dollars attached: "So this is the most verified work history data really in the world and it's at scale across 18 million freelancers that were active in the last year."
"Not noise that was created by an AI agent applying for them." — and not, she added, empty promises or claims about expertise
The Big AI Tools Are a New Origination Point for Work, Not Only a Threat
Safian asked whether she has a philosophy for engaging with the big AI players.
"You know, we're at a moment where everything is in flux and we have this growth of these new channels where customers are suddenly doing work inside of ChatGPT and Claude and all of these third-party tools" — which she said is now a new origination point for a business like hers to be referenced, to advertise and to show up
The scale of the audience is the argument. "We also know, with 900 plus million users on ChatGPT, Anthropic growing so massively quarter by quarter, I think we can see that these are the new destinations for work."
So the company is building for it despite the uncertainty. The momentum is building and snowballing, she said, and Upwork has taken the approach of building for these tools and this future even though there will be ongoing volatility as things mature
Open-Source Models Are a Gateway Decision, Not a Loyalty Test
Safian asked whether she differentiates between OpenAI and Anthropic and the open models coming out of China — whether agents built on those models are allowed on Upwork, and whether they have to be treated differently.
The authentication layer is where the judgment sits. "You know, we look at everything in our authentication layer and do make some judgments about who is allowed and not allowed to operate in our ecosystem", and she said those rules will keep changing as new information about safety and security arrives
The point of the framework is what customers asked for: "They want to benefit on both sides, but they need to do so in a place that's not the wild wild west."
Safian said some chief executives have told him recently that the Chinese models are much cheaper and, while not quite as capable, are capable enough
She agreed, with a condition. "And I think a lot of those open source models can be used safely and securely with the right controls."
"It's not a binary of use Claude or good luck to you and there's nothing you can do."
Upwork runs its own models through a gateway and switches on price. "We personally have in our business have taken a gateway approach where we are constantly testing the performance of a variety of different models including open source models and now that token costs have been going up so significantly from the major providers here in the US there are definitely use cases where we know we can get 10x savings or more by using open source models and doing it safely inside of our own environments."
She sees the same choice becoming a marketplace product. Upwork has always sold choice across a large supplier base, and she expects to show customers the pros, the cons and the costs of different combinations of models and agents, with or without human supervision
The trade-off customers already make: "Do I want to pay more for speed and quality or am I happy to go a little bit slower and maybe have some savings?"
Freelancing Is a Choice and a Transfer of Risk at the Same Time
Safian put the romance of freelance and fractional work under pressure: the promise of freedom and making your own schedule is appealing, but "How much of that is a choice versus a necessity, a sort of a symptom in some ways of companies shifting risk onto individuals?"
She refused to pick. "Absolutely both. 58% of current full-time employees are actually thinking about becoming freelancers for a lot of the reasons you mentioned around choice and control and deciding when, where, and for who they work."
Freelancers also tell her the economics have become much more appealing over the last few years
Her origin story for the shift is the last recession, not this one. "Before the 2008-2009 great recession, people at that time, I think, felt they were very secure in a full-time job, and then that recession happened and they realized, wait a minute, I'm actually more secure if I have a portfolio of three to five clients on Upwork."
Those clients keep coming back for repeat work, and the portfolio often includes a podcast, a book or a creator-economy income on the side
The youngest cohort has moved furthest. "Gen Z is the biggest one: 42% of Gen Z are freelancing."
In her account of their reasoning: "Look, I feel more comfortable that I'm not exposed and that I can really both learn and use and build the skills I want and also have income that is predictable and in my control versus at the whim of an employer."
Safian read it as generational: "It's just a different conception of what a career is for younger folks." Brown agreed, and added that freelancing is also how a worker proves AI skills — real work, real deliverables, and a verified history to market afterward
AI Fluency Carries a Wage Premium, and the Burden of Getting It Sits on the Individual
The gap she reports is a pay gap, not a hiring gap. "Freelancers that have AI skills are earning 34% more per hour than folks who don't have AI skills", and she said that gap explains the urgency among talent to move somewhere they can use the tools when their employer will not teach them
Safian named the cost of that: "There's a lot of burden on the individual to keep upgrading those skills as you go." When software learns faster than people do, he said, the pressure to keep refining your own skills rises whether you freelance or not
She did not soften it. "That is the reality. The urgency and the need for all of us, I mean you, me, all of us to become fluent in using AI for our jobs."
"I mean, that is something none of us can escape."
Her claim is that pay is what makes the learning happen. Because it puts food on the table, freelancers move first: "They do it faster. They learn faster. They lean in harder."
"We see this huge advantage actually with the freelance talent pool that's well ahead of full-time employee counterparts in terms of their adoption and facility with AI."
That gap is now a reason to hire freelancers at all. Companies look at their own teams, see the skills missing and count the time and money to climb the curve
"Maybe I can short circuit that by bringing in freelancers who can both add value and do work, but also upskill my team at the same time as a byproduct of sitting side by side with them."
Safian summarized it back: "So it becomes a backdoor way to train my own team at the same time."
"And not have to wait for those fruits to materialize before you start to actually see some real business benefit because a freelancer can be adding value in hours."
Nobody Becomes Capable With AI in a Seminar, Which Is Why the AI Orchestrator Is Rising
Safian asked the question he said everyone has: how long does it take, and what is the best way to train people in AI skills.
She answered with a client's line from that morning. "I was actually talking to one of our clients this morning and he said, you can't learn to ride a bike in a seminar."
"And I think the same is true for AI."
"You know, you could watch a YouTube tutorial and get more familiar, but at the end of the day, until you're using these tools day in and day out" — daily use, she said, is where capability actually comes from
The hiring consequence is a new kind of generalist. "And that's one reason we're seeing the rise of a new generalist called the AI orchestrator."
The role is becoming extremely popular on the platform, she said, because clients want someone who can orchestrate the tools rather than someone very deep in a single domain
"Instead, they have the facility to evolve, adopt new tools, figure them out very quickly because we know the technology is moving so quickly."
Her Own Use of AI Is a Natural-Language Query Layer Over the Company's Data
Safian asked how she uses AI in her own job.
She started with what she takes to be AI's core strengths — gathering data, generating and synthesizing — and described an internal tool built on them: "We call it crystal ball. And it has access to all of our internal data."
The value is removing herself from the reporting queue. "And this is wonderful for me because instead of having to go ask one of my team, like, hey, send me the most recent report on this, or like what happened on this day with this customer type, I can actually just directly query our database and just through natural language get tons of insights around real-time business information, which then we are using in real time to make different decisions."
When the Shocks Come From Outside, the One-Year Roadmap Is the First Casualty
Safian set up the period she has led through: the pandemic, then the invasion of Ukraine, "which prompted you to shut down operations in Russia and Belarus", then a backlash against globalization that threatens Upwork's own premise. He asked how a cadence of change driven from outside the company changes the way she decides, plans and picks what to focus on.
Her first move is subtraction. "We've worked to try to like remove noise in the organization so that it's very clear the small number of things that we're working on, and then if we need to change one of them, that we can communicate that really clearly once that decision is made."
Competing priorities and people moving in different directions make an unstable environment harder to navigate, she said, and the more unstable things are outside the company the more locked down the focus inside it has to be
The planning horizon itself has been cut. She spent many years as a product manager before becoming chief executive, and a one-year roadmap then felt reasonable even knowing it would change
"Now in product, if you have a one-year roadmap and really you're executing the same thing in month 11 that you thought you would be when you started, something's probably wrong, because you will have learned so much more and the technology will have become more capable and all of these things."
"So we've shortened some of our planning cycles, still setting kind of big ambitious goals out in the future but not being as committal to the steps required to get there."
"It's kind of like let's take the first step and then re-evaluate what is that second step." — rather than presuming that today's vantage point knows what that step is
Ranked Lists, and a Portfolio Split Between the Known Road and the Speculative One
Safian asked what tools she uses to focus and prioritize, and offered one: "I remember talking to Daniel Ek at Spotify at one point, and he would say each quarter he'd like create a list of 10 things, but they were ranked."
"Like you couldn't go to number two until you satisfied number one."
She uses the same device. "We use ranked lists all the time like Daniel."
The second tool is a resourcing split. She looks at how much resource sits on things the company has high confidence it will deliver in the near to medium term, because it is a road already traveled, against how much sits on the speculative — future states the company can see coming but that have not arrived
She said internal alignment on the ratio between those two has become more important in this environment
And when something speculative arrives sooner than planned, it moves into the other half of the portfolio, which forces the question of where the next horizon's investment goes
The Humans-Versus-Machines Debate Is What Is Blocking a Future Where Everyone Wins
Safian's last question was what is at stake for the future of work right now.
"The binary debate that's happening around will humans win or machines win, I think that is distracting us from, in some cases, putting our collective societal energy towards building a future where everyone wins, where humans win with AI and empowered by AI, but not at the loss of jobs, opportunities and purpose that people have and will, I think, always need from doing meaningful work."
Safian's Close: The Data Is Guesswork, and Only Humans Can Say What Humans Value
Safian closed alone, over what he had heard.
He said the conversation was rich on how to lead in changing environments, on the good and bad of AI agents, on how to build skills today and on "Gen Z's changing perspective on work and careers"
The part he kept returning to was the meritocracy argument — that Upwork's system quantifies and captures what specific human talent actually gets done, and the value delivered
He then undercut the whole genre, including hers. "When it comes to how AI will change the future of work, so much of the data is essentially guesswork."
Gazing into the future has always involved guesswork, he said, and "Whatever crystal ball Upwork has built internally, it only reflects the past."
"But what we do know is that what's valuable to humans can only be determined by humans."
Brown's bottom line is that AI is destroying and creating work at the same time and faster than previous cycles, and that the thing holding its value through the churn is evidence — a verified record of what a person has actually delivered — which is why she treats the argument over whether humans or machines win as the wrong argument to be having.
Products, Companies & Tools Mentioned
Upwork (Her own company, and the vantage point for the whole conversation: the platform's AI category grew "over 22% year-over-year" while small-business hiring pulled back across all channels)
Upwork's MCP server (The authentication layer that lets an agent identify itself and then post jobs, make hires and manage projects — and lets a client stuck inside a chatbot hand the work to a human immediately)
UMA (Upwork's own agent, which Safian said "assists users writing job descriptions and prepping interviews" — the source of his question about whether AI now writes applications for other AI to read)
Claude and ChatGPT (The third-party tools where she says customers are now originating work, and where the MCP connection plugs a human expert into a stalled deliverable)
Anthropic and OpenAI (The US providers whose rising token costs pushed Upwork toward open-source models for some workloads; she cited "900 plus million users on ChatGPT" and Anthropic "growing so massively quarter by quarter")
Upwork's "crystal ball" (The internal tool she queries in natural language against all of the company's data, instead of asking a team member for a report)
Spotify (Where Safian's prioritization example comes from — Daniel Ek's ranked quarterly list of 10 things, which Brown said Upwork copies)
Books & Resources Mentioned
Upwork's Future Workforce Index 2026 (The research Safian asked about and the source of the freelancing-share, wage-premium and full-time-employee figures she quotes through the interview)
A message from Hayden Brown, Upwork CEO, to employees on 7 May 2026 (The frank statement Safian credits her with publishing when she announced the layoff)
Redesigning work in a crisis, with Hayden Brown (Her previous appearance with Safian, the geopolitics conversation he opens by recalling, which covers the exit from Russia and Belarus)
How to build trust fast, with Daniel Ek (Reid Hoffman's Masters of Scale interview with the Spotify founder Safian names, linked from this episode's own transcript page)
Get the latest market chatter as it happens:
If this saved you the listen, send it to one person who would rather have the time back.