Speakers

Scott Belsky — Chief Product Officer, Adobe; Partner, Benchmark; Founder, Behance. Scott Belsky is one of Silicon Valley's most respected product leaders and early-stage investors. As CPO of Adobe, he led the company's AI creative tools strategy and managed M&A. As an investor and partner at Benchmark, his portfolio includes Uber, Pinterest, Warby Parker, and Valence. He is also the founder of Behance, the world's largest platform for creative portfolios, and the author of multiple books on creativity, organization, and the future of work.

Parker Mitchell — Co-Founder and CEO, Valence. Parker leads Valence, an enterprise AI coaching company whose platform Nadia is deployed across global Fortune 500 organizations. He moderated this fireside chat, drawing on his experience building AI coaching infrastructure for enterprise talent and leadership development.

Key Takeaways

• Organizational debt is as dangerous as technical debt — and AI will force a reckoning: Just as software accumulates technical debt from deferred decisions and architectural shortcuts, organizations accumulate organizational debt from decisions that should have been made but weren't. Scott Belsky argues that AI is now the most powerful organizational debt collector in history, collapsing outdated processes faster than any prior technology. The leaders who thrive will be those who proactively identify and eliminate organizational debt before AI exposes it.
• The law of displacement speed is creating two inevitable outcomes — commoditization and vertical operating systems: When technology changes fast enough to displace incumbents rapidly, Scott Belsky argues two things always follow: commoditization of the underlying capability (in this case, AI tokens), and the rise of vertical operating systems that embed the capability at the layer where it creates the most value. Valence, in Scott's view, is building the vertical operating system for the people function — the layer where AI coaching creates compounding organizational value.
• Novelty must precede utility — and enterprise leaders need to protect space for play: Scott Belsky's principle that 'novelty precedes utility' runs headlong into the enterprise reality that urgent always outweighs important. His prescription: protect dedicated space for play and pilots, and redesign incentives so that teams are rewarded for learning — not penalized for failed experiments. The measure of a pilot should never be utilization; it should always be learning.
• AI is restoring the hyper-personalized life humans evolved for — but on our own terms: For most of human history, people lived in small communities where they were deeply known — their strengths, preferences, and relationships recognized by those around them. The Industrial Revolution made everyone anonymous. Scott Belsky believes AI is restoring that personalization, but in a new and better form: one where individuals define their own preferences, share their own data on their own terms, and are known by AI systems they trust.
• AI coaching enables the kind of feedback humans cannot easily give each other: Scott Belsky argues that feedback between manager and employee is inherently compromised — by defensiveness, perceived bias, thin-slice judgment, and the social dynamics of hierarchy. AI coaching changes this because it is built from the residue of a person's own actions and is genuinely oriented toward their development, not a manager's convenience. It is much harder to be defensive with feedback that feels like an extension of yourself.
• Taste and agency are the ultimate human advantages in an AI world: As AI handles more process and execution, Scott Belsky argues the two distinctly human capabilities that will define individual and organizational success are taste — the filters and discernment that determine which of the infinite AI outputs are actually good — and agency — the willingness to believe something is possible when it seems unreasonable, and to pursue it. These are not technical skills. They are capacities that organizations must actively develop in their people.

‍

Full Transcript

Organizational Debt: The Hidden Drag on Enterprise Performance

[00:00:00]

Parker: If you're in the startup world and you come across Scott Belsky, you either know him as the legendary investor — Uber, Pinterest, Warby Parker, a whole bunch of early seed investments —

[00:00:15.036]

Scott: Valence.

Parker: Valence as well. And the Chief Product Officer of Adobe, formerly the founder of Behance. But I want to go even further back. You're something of an organizational design and management nerd at the starting point. Can you tell us about your early days at Goldman Sachs and how that interest came about?

[00:00:38.703]

Scott: The most important career experience I ever had was after a year and a half or two years doing the traditional finance thing at Goldman. I realized this was not for me. I went to my manager, a woman named Catherine, and said, 'I think I need to go do something else.' She said, 'What would your dream job be if you stayed at Goldman?' And I said, 'It would be really cool to understand how the organization works.' I was interested in management — which, of course, as a 23-year-old kid sounds a bit naive.

I ended up getting a role as an analyst on a team called Pine Street, in the executive office, that was all focused on organizational improvement and leadership development for the most senior population of the firm. For three years, I was learning from practitioners — people who would come in to do executive coaching with leaders, but also the hedge funds that would get all this capital and have no management experience. They were total chaotic messes. To learn about the importance of leadership development, accelerating career paths, and everything else at such a young age was an incredible education.

[00:01:54.900]

Parker: And you got a chance to apply that in practice after your startup was acquired by Adobe. A 7,000-person organization. Was anything surprising as you made that transition into a large company?

[00:02:11.360]

Scott: I've now done startup, small company, large company, and back to a smaller team again. There's always this balance — when you have a problem, you throw process at it. And when you're a really great manager, you kill process. It's this incredible tension between process creation and process destruction.

At Adobe, I was most effective when I found a way to collapse the talent stack — get teams more truncated and more direct with one another. And whenever I was destroying a past-due process, I also became highly convinced that organizations don't only suffer from technical debt — the accumulation of bugs and bad technical decisions that plague a product. Organizations also suffer what I call organizational debt, which is the accumulation of decisions that should have been made but weren't.

At Adobe, my favorite thing was to prompt decisiveness. People were always uncomfortable with decisiveness because the easiest decision is to not make a decision yet. I was once inspired by Ken Chenault — who ascended American Express to the very top — when he was asked what most helped him rise through a huge corporate bureaucracy and be known as an innovator. He thought for a moment and said, 'At the end of the day, I would make my bosses make decisions.' That notion of clearing organizational debt, prompting decisions or at least a deadline for decisions, keeping that ship moving inch by inch — that may be one of the greatest things you can do in a bureaucracy.

▶ What Is Organizational Debt and Why Does It Cripple Large Companies?

Organizational debt is the accumulation of decisions that should have been made but weren't — the enterprise equivalent of technical debt. Scott Belsky, former Chief Product Officer at Adobe and partner at Benchmark, introduced this concept based on his experience leading product and M&A at a 25,000-person technology company. Organizational debt compounds silently: every deferred decision becomes a process, every process becomes a constraint, and every constraint slows the organization's ability to respond to change. AI, Belsky argues, is now the most powerful organizational debt collector in history.

The Law of Displacement Speed: What Rapid AI Change Actually Means

[00:04:11.419]

Parker: Ethan Mollick noted that organizations are not designed — they evolve through countless small decisions that plant more seeds than they weed. AI, I think, accelerates that. But I want to introduce your story on the edge of AI and watching the power of the models grow. What was your first experience with large language models, and what was your emotional reaction?

[00:04:50.519]

Scott: I'm always playing with the latest tools and trying to figure out where all this is going. When early LLMs emerged, and when some of the early imaging models emerged, these were toy things — silly and unreliable and full of hallucinations. I remember the days we would ask a simple math problem to a large language model and it would give the wrong answer. Then you'd say, 'No, it's not.' And it would say, 'Oh, you are right. I am wrong.' And you could even trick it back to being wrong again.

[00:05:26.120]

Parker: And then it would make the same mistake again.

Scott: Exactly. What I find fascinating about this moment is what I call the law of displacement speed — when you have such a rapid sense of displacement. A year and a half ago, everyone was saying Google totally missed AI, Anthropic was probably going to run out of money, and OpenAI was going to rule the world. Here we are today: Google is winning, Anthropic is winning enterprise, OpenAI may face pressure. And three months from now, I bet it's different again. Everyone is outpacing one another.

When the law of displacement speed kicks in, my belief is two things happen. First, rapid commoditization — they keep making it better and more competitive, which seems like a path toward commoditization of tokens. Second, it goes more central at what I call the operating system level. The operating systems of our personal lives are Android and iOS. The operating systems of the enterprise are going to be vertical operating systems — AI-driven, built for specific functions. That's why I was excited about what Valence is building. The people function is going to have a vertical operating system. And I think that's where the future is going.

▶ The Law of Displacement Speed: Why Rapid AI Change Leads to Commoditization and Vertical Operating Systems

Scott Belsky, partner at Benchmark and former Adobe CPO, describes the law of displacement speed as the economic and competitive dynamic that emerges when technology changes fast enough to displace incumbents within months rather than years. He argues two outcomes are inevitable when this law kicks in: first, rapid commoditization of the underlying technology (in AI's case, token processing); second, the rise of vertical operating systems — AI-native platforms built for specific organizational functions. Belsky sees enterprise AI coaching as an early example of a vertical operating system for the people function.

Novelty Before Utility: How to Create Space for AI Adoption

[00:06:52.420]

Parker: One of the ideas you've talked a lot about is that novelty precedes utility. For most people's experience of AI, that's come true. How have you seen that play out in the enterprise, which doesn't exactly welcome novelty as a step before utility?

[00:07:09.699]

Scott: The enterprise is tricky, because urgent always outweighs important. The gravitational force of operations is so strong that you never have a free hour to just experiment with a new AI tool. We all just go through our day. And of course, if you don't make time for the longer-term important things, they never get considered. It is absolutely critical to our organizations and our own careers that we refactor and reimagine how we work. How do you make time for that?

Play is part of it. You have to allow yourself — and, more importantly, your team — to try things. And you have to protect them. If they decide to use a tool in a new way, give them different KPIs than success. The KPIs need to be learning. Did we try this new review process with this new tool? Don't penalize the team if it didn't go well. Reward them if they learned something. The constructs of play and protection — always having pilots in an organization — are really important principles for getting socialized with new technology.

Parker: And in the pilot, the measure of success shouldn't be utilization. It has to be learning — how did this new approach change how I do my work?

Scott: Exactly. Change starts with socialization. When I led M&A at Adobe, I learned quickly that you can't just come in with a big, bold proposal, get people to nod, and execute it. It's a long process of getting people comfortable with an idea until it becomes obvious. But when it started, it was anything but obvious. Similarly with new technology — you have to socialize it. Play, pilots, and different incentives are part of that.

▶ Why Play and Protection Are the Foundations of Enterprise AI Adoption

Scott Belsky, former Chief Product Officer at Adobe and partner at Benchmark, argues that enterprise AI adoption fails when the first metrics applied are utilization and ROI rather than learning. His prescription: create protected space for play — deliberately shielded from standard KPIs — and reward teams for what they learned from an AI experiment, not whether it succeeded by traditional measures. Belsky draws the parallel to M&A socialization: change doesn't happen through bold mandates; it happens through sustained exposure until a new approach becomes obvious.

AI Personalization: Being Known on Your Own Terms

[00:09:11.860]

Parker: One of the words you've talked a lot about in the AI era is personalization. You believe AI is going to lead to a new world of personalization in both consumer and enterprise. What will drive that, and why is it so exciting?

[00:09:28.000]

Scott: One of my cardinal beliefs in all of technology and investing is that we, as humans, are longing for the way things once were — but with more scale and efficiency. For the other 300,000 years of human history that we're hardwired for, we lived in small towns and villages. We were known by people. They knew our strengths and weaknesses. They knew our children's names. They knew our likes and dislikes. We were living a hyper-personalized life.

Then, with the Industrial Revolution and everything that followed, we all became anonymous. Everything became generalized. When I go to Nike, it says, 'What gender are you?' When I go to a restaurant, they don't know I'm a vegetarian. But I believe we all want to be known — to feel special in our workplace, in our life, in our communities.

The first response of technology to this was the bad version: knowing people without letting them know how they're being known. That was ad tech. There is a better way. In this AI-enabled world, the next generation of personalization is being known on our own terms — defining our preferences, sharing and syncing the data we choose to share with AI, having private instances where AI can help us. The notion of personalized living — and personalized working — is a huge investment thesis for me.

▶ How AI Is Restoring Personalization to Human Life and Work

Scott Belsky, partner at Benchmark and former Adobe CPO, argues that AI personalization is not a new idea — it is a restoration. For most of human history, people lived in small communities where they were deeply known. The Industrial Revolution made everyone anonymous. AI now makes it possible to be known at scale again, but on different terms: individuals define their own preferences, share their own data, and engage with AI systems they trust. In the enterprise, this means tools like Nadia can know employees' development needs, working styles, and growth priorities — not to surveil, but to genuinely support.

Why AI Coaching Unlocks Feedback Humans Cannot Give Each Other

[00:11:21.080]

Parker: What's fascinating about this new world of technology is we can be in relationship with it, and we can feel seen by something that is not another human being — but the experience is feeling seen. How will that change how we interact with a new set of intelligences we can be in relationship with?

[00:11:45.279]

Scott: When you get feedback from your manager, you are likely to be defensive. You probably think your manager has too thin a slice of judgment to make that call — or to suggest that's your development area. It's a process riddled with bias, or at least perceived bias, and defensiveness.

Now cut to a world where we have a personal relationship with something that is genuinely there to make us better. It's not taking out a bad day on us. It's not making sweeping generalizations about our weaknesses based on one observation. It is much harder to be defensive and unreceptive when you're getting feedback from an extension of yourself — something made from the residue of your daily actions. It's personal in a good way. And it can be a private forum where you start to understand rather than defend.

Parker: The relationship of trust is key. You have to believe that system has your best interests at heart — and that it's independent, not just a sycophant.

Scott: Exactly. Not too affirming. The basic system prompts of large language models are biased toward the user feeling accomplished — inherently the opposite of productive pushback. What's interesting about what Valence and others are exploring is internal debate being choreographed under the hood, optimized toward genuine self-improvement in the vertical of people development.

Parker: Let me share a funny example. The team created a version that would tell someone which Harry Potter house they'd be sorted into.

Scott: The Sorting Hat.

Parker: Right. Someone received Hufflepuff. Their response: 'No, I don't think I should be in Hufflepuff. I think I should be in Gryffindor.' And Nadia's response was: 'Would you like to discuss the difference between how you see yourself and how others might perceive you?'

Scott: 'No, thank you.'

Parker: But you have to have that independence — because we grow in relationship to independent points of view that force us to confront things we might not see. And I think there's a profound opportunity there.

▶ Why AI Coaching Enables More Honest Development Feedback Than Human Managers

Scott Belsky, former Adobe CPO and Benchmark partner, argues that manager-to-employee feedback is structurally compromised by defensiveness, perceived bias, and thin-slice judgment. AI coaching changes the dynamic because the feedback comes from something genuinely oriented toward the employee's development — built from the residue of their own actions, not filtered through a manager's perspective. Belsky notes it is much harder to be defensive with feedback that feels like an extension of yourself. The key, he emphasizes, is that the AI must be independent — not affirming, and willing to offer real pushback.

AI as Organizational Collapser: Preparing for Process Transformation

[00:14:33.285]

Parker: Zooming out — organizations have processes that grew up because of decisions not taken or technology choices made long ago. AI is the ultimate collapser. It is going to collapse a series of those processes faster than any prior technology change. How can organizations prepare?

[00:15:16.799]

Scott: One of the greatest change agents in organizations is celebrating what works — a specific example. Positive examples are like viruses. They run rampant once you showcase them. In my experience, it's in tight, cross-functional teams — what I call collapsed stack teams, where instead of having a designer and an engineer and a product leader and so on, you have people playing dual roles with a tight conduit in their heads — that are the early adopters of new tools. They realize, 'Oh, we don't have to meet Tuesdays just because it's Tuesday,' or, 'We've debuted this new way of managing change orders.' And then, if showcased properly and celebrated, it becomes a best practice across the org.

It's hard to come in from the outside and say, 'You should change this.' But once little pockets of change happen, the job is to amplify and spread them effectively.

▶ How Collapsed Stack Teams Drive AI Adoption Across Large Organizations

Scott Belsky, partner at Benchmark and former Adobe CPO, identifies the best early adopters of AI process change as 'collapsed stack teams' — small, cross-functional groups where individuals play dual roles and share a tight cognitive conduit. These teams move fast enough to discover genuinely new ways of working, and their wins — when celebrated visibly — spread as positive viruses across the organization. Belsky's prescription for enterprise AI transformation: stop trying to change from the top down, and instead find and amplify the pockets of change already happening organically.

Taste and Agency: The Two Human Advantages That AI Cannot Replace

[00:16:40.637]

Parker: You've talked about how the change at organizational scale is going to be difficult. The leaders who make it successful are going to have to put tremendous energy into that system. What advice would you give them?

[00:17:05.019]

Scott: There's a Fortune 500 company whose leader couldn't believe the soundbite that there would be no more junior hires. They are actually hiring 1,000 brand-new, almost intern-level people — and they shifted their hiring intensely toward the very junior end. When I asked why, they said: 'Because they know AI. They just came out of college. They are native to this. I want to swarm my organization with people who live and breathe this technology because they'll be in meetings and look at things being done and say, Why are you doing it that way?'

It's a form of knowledge arbitrage. Junior people may not understand our businesses or processes, but they have something we don't — native experience with the technology. Is that a tactic?

Parker: The core concept is: we don't know the path. We have to co-discover it together, building feedback loops, understanding where to lean in and celebrate what works. Looking to 2027 — what's the bold organizational thing a CEO might try that turns out to be unexpected?

[00:19:03.440]

Scott: We have to start thinking about what humans can uniquely do and how to accentuate development around those things — and what we should genuinely offload to compute. Process will be offloaded. Processes will become recursive and self-improving — they'll examine themselves, start to truncate themselves, and improve on their own. I believe executives are going to apply really stringent forcing functions: 'No more headcount for you. Your forcing function is to find a better way. I'm very worried about you throwing people at this problem.'

But then the question is: with these humans, how do we elevate and tap what they are uniquely capable of? To me, the future of humanity comes down to two things: taste and agency.

Taste is the inputs we get that help us make decisions — the filters we apply to the noise constantly bombarding us. The algorithms fooling us. The cacophony making us believe certain things. The filters. And then discernment: based on the inputs and filters, what decisions do we make? How do we elevate taste in our organizations so people exercise it better?

Then agency. How do we get our people to believe things are possible that others would dismiss? Every industry disrupted by an incumbent coming back, every startup that changed an industry — it always starts with a few humans who believed something others said was wrong. Airbnb defied the entire hotel industry. Everyone thought they were foolish. Every example in history — including examples within our own companies — involves people who took a disproportionate amount of agency. As AI frees up energy from other work, we need our people to channel that energy into more agency.

▶ Why Taste and Agency Are the Last Human Advantages in an AI-Powered World

Scott Belsky, partner at Benchmark and former Adobe CPO, argues that as AI handles more process and execution, the two distinctly human capabilities that will define individual and organizational success are taste and agency. Taste is the capacity to filter noise, apply judgment, and discern which of the infinite outputs AI generates are actually good — the human editorial function. Agency is the willingness to believe something is possible when conventional wisdom says it isn't, and to pursue it. Belsky argues that every meaningful disruption in history began with a few people exercising disproportionate agency, and that AI — by freeing humans from routine work — should increase both.

Diversity and Agent Networks: Why Different Points of View Drive Innovation

[00:21:51.160]

Parker: To wrap it all together — you and I talked about the importance of diversity. Because that agency, if everyone says the same thing, won't reach the conclusion it would with a range of different points of view. How important is diversity, especially in this human-plus-AI era?

[00:22:06.599]

Scott: When you have an extraordinarily different group of extraordinary people sitting around one table who respect one another, that's where the magic happens. Innovation is essentially the edge that will someday become the center. Innovation happens at the edge of reason. If we all went to the same school, same cohort, same age — the same things will be reasonable and unreasonable to us. But if we're all different, and I say something you initially see as completely unreasonable, because it's at the edge that may someday become the center — that's the process of innovation. Through subsequent conversations, it becomes socialized. Before you know it, we're all in agreement about something that started at the complete edge. That's innovation executed.

Now think about agents. In just the last seven days, there's this new phenomenon of agent social networks — agents made from different language models, with different system prompts, coming together and debating a task you ask them to perform. This happens under the hood of products like Valence's as well, to some degree. And as a result, you get a better solution. Duh — it's the same thing as diversity all over again. Multiple points of view argued and synthesized, and feedback as the process of improvement.

Parker: And that's why you're such a great investor — you spot things at those edges. They're a little unreasonable, but you find the kernel of reasonableness and see which ones are going to matter.

Scott: And I debate them with the contrarians I know. The ones who tell me I'm completely wrong. Either that makes me realize I was wrong — or it allows me to gain confidence from being doubted. When we feel confident in the face of doubt, that's when we're really on to something.

Parker: Sharpens the thinking.

Scott: Right.

Parker: We really appreciate you making the time to come join us today.

Scott: Of course. Thanks, everyone.

▶ Why Diversity Is the Structural Foundation of Innovation — and What Agent Networks Are Teaching Us

Scott Belsky, partner at Benchmark and former Adobe CPO, frames diversity as an innovation mechanism: when people from genuinely different backgrounds sit together, ideas that seem unreasonable to one person are reasonable to another — and that tension is precisely where innovation begins. He draws a direct parallel to AI agent social networks, where agents built on different models and system prompts debate the same task and produce better outputs through the collision of perspectives. In both cases, the structural principle is the same: more distinct points of view, properly synthesized, produce better decisions.