Skip to content

Podcast

Why Enterprise AI Has a Leadership Problem

The AI Daily Brief: Artificial Intelligence News and Analysis

Source ↗ ← All highlights
  • Shift From Tech Project To Operating Model
    • Treat AI as an operating model shift, not a tech project.
    • Nathaniel cites KPMG embedding AI enterprise-wide to raise human capability rather than running it as a standalone tech initiative. Transcript: Nathaniel Whittemore All right, folks, quick pause. Here’s the uncomfortable truth. If your enterprise AI strategy is we bought some tools, you don’t actually have a strategy. KPMG took the harder route and became their own client zero. They embedded AI and agents across the enterprise, how work gets done, how teams collaborate, how decisions move, not as a tech initiative, but as a total operating model shift. And here’s the real unlock. That shift raised the ceiling on what people could do. Humans stayed firmly at the center while AI reduced friction, surfaced insight, and accelerated momentum. The outcome was a more capable, more empowered workforce. If you want to understand what that actually looks like in the real world, go to www.kpmg slash AI. (Time 0:09:35)
  • Agent Deployment Is Now Core Not Experimental
    • Agent deployment crossed 50% and is now central to enterprise workflows rather than edge experiments.
    • KPMG Q1 shows 54% of orgs have agents live, with 40% scaling, 6% building multi-agent systems, and 9% orchestrating. Transcript: Nathaniel Whittemore We have discussed enterprise AI implicitly or by extension quite a bit recently without necessarily going super deep on what recent numbers are telling us. I shared the AI maturity maps framework last week, which is a way of looking at AI readiness and AI adoption across six different dimensions, including deployment depth, systems integration, And governance, and shared a bit about what our research had told us about where organizations are right now and why we think in many cases it’s behind where they need to be. But of course, that’s different than digging into the actual numbers themselves. And recently, we’ve gotten a bunch of different studies, all with direct sourcing from inside companies, that are telling some similar and some different stories about enterprise AI. So what we’re going to do today is talk through what those studies are telling us, where they agree, where they disagree, and what I think the sum total is, and why even all of this still Might be missing something. First up, we have some research from A16Z. Now, where this data comes from is the aggregation of private data from a number of leading enterprise AI startups who live and work inside many of these big corporations. Here’s a couple of the highlight numbers. A16Z found that about 19% of the Global 2000 are live-paying customers of a leading AI startup, with that number rising to 29% of the Fortune 500. That means the enterprises have signed a top-down contract with an AI startup, successfully converted a pilot, and gone live with the product in their organization. Now, 29% might seem low, but as you heard, that does not include pilot efforts, nor, my guess, is it comprehensive across every tool that companies might be using. Their next exploration is what is actually working. And here’s their methodology. A16Z writes, We find that the most indicative way to assess the work the models are inherently better at doing is to overlay revenue momentum across use cases against the theoretical Capabilities of models as defined by GDPVal. They write that to them, these two factors encapsulate both how good models could be as well as how much they’re proving to deliver today. When it comes to use cases and functions, enterprise AI adoption is dominated by coding, support, and search with coding being the absolute biggest by an order of magnitude. The tech, legal, and healthcare sectors they found have been the industry’s most eager to adopt AI. Now, we have talked so extensively about coding being the dominant use case for AI that I don’t think we need to get into it here. But their discussion of support, I think, is interesting for reinforcing why it’s a comparatively good place for organizations to start when it comes to AI. First of all, they point out that a lot of the type of work that AI is doing was already outsourced in some way because, as they put it, companies deemed it too tedious and complicated to Manage themselves. Second, they argued that its discreteness really matters, i.e. The nature of most support interactions is time-bound with a constrained intent that outputs into a well-defined problem for an agent tackle. It’s got an easy ROI profile because support operates on quantifiable metrics like number of tickets answered, satisfaction scores of customers, and resolution rates. And, and I think importantly, they point out that support doesn’t require 100% accuracy to be useful since it has natural off-ramps to a human, e.g. The I’m escalating you to a manager. Now, when it comes to the industries, again, technology is not a surprise, but legal, they write, was primed for adoption of AI because it had actually been left behind by traditional Enterprise software. They write, static workflow tools didn’t accelerate the unstructured nuanced work that lawyers typically did, but AI has made the value prop of technology to lawyers much clearer. AI is excellent at parsing dense text, reasoning over large amounts of text, and summarizing and drafting responses, all work that lawyers regularly do. Healthcare, they argue, is another market that’s responding to AI in a way that it didn’t for traditional software. They write, healthcare was historically a slower market to adopt software because one, the highly skilled and complex work mapped poorly to the problems traditional workflow software Could solve, and two, the dominance of the systems of record EHRs like Epic squeezed net new software vendors. With AI, however, they write, companies have been able to take on discrete human labor work that circumvents the system of record by either replacing administrative work, e.g. Medical scribes, or augmenting higher value work doctors were doing. The work is distinct enough then not to require a rip-and of the EHR. Now, moving out to a more longitudinal view, we have KPMG’s most recent quarterly pulse survey. This data comes from a set of executives from companies primarily with more than a billion dollars in revenue and is their recurring tracker so we get some amount of quarter-over data. A couple of big things stand out. Despite ROI still being hard to quantify in really clear ways, the average anticipated spend on AI just continues to go up. In Q1 of last year, organizations reported to KPMG that they anticipated spending about $114 million on average over the next 12 months. That has now jumped to $207 million. And part of the reason might be that agents are now, to put it bluntly, very real. In Q1 of 2025, only 11% of organizations had agents in deployment. Many more were experimenting or piloting, but that was the number where agents were in full production. In Q2, that jumped meaningfully to 33%. And yet in Q1 of this year, that number is now over 50% for the first time at 54%. Within that 54%, 40% are scaling or deploying, 6% are developing multi-agent systems, and 9% are orchestrating. Piloting is down to 30%, and experimenting is down to 14%. (Time 0:13:22)
  • Coding Dominates Enterprise AI Use Cases
    • Coding, support, and search dominate enterprise AI adoption with coding far ahead.
    • A16Z analysis shows coding leads by an order of magnitude and tech, legal, and healthcare are the most eager adopters. Transcript: Nathaniel Whittemore A16Z found that about 19% of the Global 2000 are live-paying customers of a leading AI startup, with that number rising to 29% of the Fortune 500. That means the enterprises have signed a top-down contract with an AI startup, successfully converted a pilot, and gone live with the product in their organization. Now, 29% might seem low, but as you heard, that does not include pilot efforts, nor, my guess, is it comprehensive across every tool that companies might be using. Their next exploration is what is actually working. And here’s their methodology. A16Z writes, We find that the most indicative way to assess the work the models are inherently better at doing is to overlay revenue momentum across use cases against the theoretical Capabilities of models as defined by GDPVal. They write that to them, these two factors encapsulate both how good models could be as well as how much they’re proving to deliver today. When it comes to use cases and functions, enterprise AI adoption is dominated by coding, support, and search with coding being the absolute biggest by an order of magnitude. The tech, legal, and healthcare sectors they found have been the industry’s most eager to adopt AI. (Time 0:14:28)
  • Support Offers Clear ROI For Early AI
    • Support is a high-ROI AI entry point because interactions are discrete, measurable, and have human off-ramps.
    • A16Z notes support metrics like tickets and satisfaction make ROI clear and imperfect accuracy is acceptable. Transcript: Nathaniel Whittemore First of all, they point out that a lot of the type of work that AI is doing was already outsourced in some way because, as they put it, companies deemed it too tedious and complicated to Manage themselves. Second, they argued that its discreteness really matters, i.e. The nature of most support interactions is time-bound with a constrained intent that outputs into a well-defined problem for an agent tackle. It’s got an easy ROI profile because support operates on quantifiable metrics like number of tickets answered, satisfaction scores of customers, and resolution rates. And, and I think importantly, they point out that support doesn’t require 100% accuracy to be useful since it has natural off-ramps to a human, e.g. The I’m escalating you to a manager. (Time 0:15:46)
  • AI Slips Into Healthcare And Legal Around Legacy Systems
    • Healthcare and legal adopted AI because models handle dense unstructured text without ripping out EHRs or workflows.
    • AI enables medical scribes and legal drafting that bypass heavy systems of record like Epic. Transcript: Nathaniel Whittemore AI is excellent at parsing dense text, reasoning over large amounts of text, and summarizing and drafting responses, all work that lawyers regularly do. Healthcare, they argue, is another market that’s responding to AI in a way that it didn’t for traditional software. They write, healthcare was historically a slower market to adopt software because one, the highly skilled and complex work mapped poorly to the problems traditional workflow software Could solve, and two, the dominance of the systems of record EHRs like Epic squeezed net new software vendors. With AI, however, they write, companies have been able to take on discrete human labor work that circumvents the system of record by either replacing administrative work, e.g. Medical scribes, or augmenting higher value work doctors were doing. The work is distinct enough then not to require a rip-and of the EHR. (Time 0:16:44)
  • Spending Rises Despite Unclear ROI
    • Enterprise AI budgets and agent deployment are rising even as ROI remains hard to quantify.
    • Average planned AI spend rose from $114M to $207M and agents in production jumped from 11% to 54% year-over snapshots. Transcript: Nathaniel Whittemore Despite ROI still being hard to quantify in really clear ways, the average anticipated spend on AI just continues to go up. In Q1 of last year, organizations reported to KPMG that they anticipated spending about $114 million on average over the next 12 months. That has now jumped to $207 million. And part of the reason might be that agents are now, to put it bluntly, very real. In Q1 of 2025, only 11% of organizations had agents in deployment. Many more were experimenting or piloting, but that was the number where agents were in full production. In Q2, that jumped meaningfully to 33%. And yet in Q1 of this year, that number is now over 50% for the first time at 54%. Within that 54%, 40% are scaling or deploying, 6% are developing multi-agent systems, and 9% are orchestrating. Piloting is down to 30%, and experimenting is down to 14%. And in many ways, this agentic adoption kind of defines everything else on the survey. A lot of the considerations are around how to manage new risk from agents. Cyber and employee misuse is up from 32% to 44% when asked about the most difficult society-wide challenge with AI between now and 2030. It’s also coloring challenges around employees. (Time 0:17:44)
  • Risk And Skills Gaps Are Primary Scaling Barriers
    • Risk and skills gaps drive enterprise concerns as agents scale, raising cyber and misuse worries.
    • KPMG reports increases in cyber/employee misuse concerns and 62% cite skills gaps blocking scaling. Transcript: Nathaniel Whittemore A lot of the considerations are around how to manage new risk from agents. Cyber and employee misuse is up from 32% to 44% when asked about the most difficult society-wide challenge with AI between now and 2030. It’s also coloring challenges around employees. While 55% of organizations are seeing slight or significant employee adoption of agents, i.e. Employees beginning to accept and integrate agents into their work, they’re also finding resistance. Interestingly, the resistance appears to be more about skills gaps than concerns about job security, although both rate very highly at 76% and 71% respectively. Agents are also shaping what companies expect from their talent. 57% said that they expect humans to primarily manage and direct AI agents in the next two to three years. 64% said agents had already changed their approach to entry-level hiring, which is interestingly lower than the percentage who said that agents had changed their approach to experienced Hires, which was at 71%. And for the carrot in this equation, 45% of leaders said that they’re willing to pay 11 to 15% more for strong AI skills. When it comes to how they get these skills, most leaders are looking internally first. 87 percent said that they are focused on upskilling or reskilling their current workforce. 68 percent said that they’re hiring for new roles like AI architects. 55 percent said that they’re redesigning existing roles, all of which is much higher than the percentage that are turning to managed services at 39 percent or aqua hires at 17% to get The AI skills they need. Interestingly, when it comes to what leaders value in their talent, while 71% said technical or programming abilities, and this is specifically for skills related to entry-level Employees that need to work with AI agents, 83% said that it’s about adaptability and continuous learning. Now, there are still tons of challenges and barriers to demonstrating ROI. 58% point to risk considerations such as data privacy and cyber. 59% said that they have difficulty quantifying indirect or long-term benefit. 62% see skills gaps. (Time 0:18:44)
  • Enterprises Want Humans To Oversee Agents
    • Firms expect humans to manage agents and are changing hiring and pay for AI skills.
    • 57% expect humans to direct agents, 64% changed entry-level hiring, and 45% will pay 11–15% more for AI skills. Transcript: Nathaniel Whittemore And for the carrot in this equation, 45% of leaders said that they’re willing to pay 11 to 15% more for strong AI skills. When it comes to how they get these skills, most leaders are looking internally first. 87 percent said that they are focused on upskilling or reskilling their current workforce. 68 percent said that they’re hiring for new roles like AI architects. 55 percent said that they’re redesigning existing roles, all of which is much higher than the percentage that are turning to managed services at 39 percent or aqua hires at 17% to get The AI skills they need. Interestingly, when it comes to what leaders value in their talent, while 71% said technical or programming abilities, and this is specifically for skills related to entry-level Employees that need to work with AI agents, 83% said that it’s about adaptability and continuous learning. Now, there are still tons of challenges and barriers to demonstrating ROI. 58% point to risk considerations such as data privacy and cyber. 59% said that they have difficulty quantifying indirect or long-term benefit. 62% see skills gaps. (Time 0:19:38)
  • Agentic AI Exposes Structural Organizational Faults
    • Agentic AI reveals structural organizational gaps like misaligned incentives and siloed teams.
    • Ryder and Workplace Intelligence find agentic adoption is embedding into mission-critical workflows, exposing outdated operating models. Transcript: Nathaniel Whittemore Writer CEO May Habib sums up the difference between last year’s version of the study and this year’s like this. Ownership was murky, IT and the C-suite were locked in a constant tug-of and frustration grew as that massive investments hit a wall. Only 12 months later, the tension has evolved into something much more consequential. It’s now cultural, organizational, and deeply structural. Now, importantly, Ryder is actually dealing with the new reality of agentic. May continues, the shift towards agentic AI has moved at a pace that’s hard to overstate. AI isn’t rolling out at the edges anymore. Instead, organizations are embedding agents directly into their mission-critical workflows, where they make autonomous decisions and fundamentally change how work gets done. On the one hand, you can feel the ambition. There’s an entire cohort of AI-native leaders and employees who are compounding their advantage in real time. They’re working faster, more independently, and more creatively than we could have imagined just a year ago. But all of that enthusiasm is running headlong into chaos. Agendic AI is exposing a deep structural gap that most enterprises just aren’t prepared for. It’s showing up in misaligned incentives, siloed teams, and outdated operating models that are reaching a breaking point. (Time 0:20:48)
  • Leadership Is Stressed And Strategies Are Often Performative
    • Leadership feels stressed and many AI strategies are performative rather than practical.
    • Ryder finds 73% of CEOs report stress, 39% lack a formal revenue strategy, and 75% say strategies are more for show than guidance. Transcript: Nathaniel Whittemore A couple of the highlights that I thought were interesting. One is that on the one hand, leaders are in many ways out ahead of their employees when it comes to AI adoption, with 64% of those surveyed spending at least two hours a day using these tools. 75% of executives believed that AI agents would be part of their company’s C-suite within the next five years. And yet, a full 73% of CEOs said that their company’s AI strategy was causing them stress or anxiety, with 38% reporting a high or crippling amount of stress. 61% of executives fear they could lose their job if they fail to lead their organization through the AI transition. And when you dig into the numbers, it’s not hard to see why the strategy is successful. Basically, the strategy is kind of not clear. 39% don’t have a formal strategy in place to drive revenue from AI, and a full 75% said that their company’s AI strategy was more for show than actual internal guidance, which is obviously Just a recipe for disaster. 56 percent said that AI had created power struggles and disruption at their organization, which is a big jump from 42 percent last year. On the flip side, there’s continuing employee sabotage of the efforts. 29 percent of employees, including 44 percent of Gen Z, admit to sabotaging their company’s AI strategy. And 76% of the C-suite said employee sabotage poses a serious threat to their company’s future. 35% of employees said that they’d entered proprietary, confidential, or sensitive information into a public AI tool. (Time 0:22:16)
  • Low Manager Buy-In Drives Employee Sabotage And Mistrust
    • Employee resistance and sabotage are real, fueled by mistrust and poor manager adoption.
    • Ryder reports 29% of employees admit sabotaging AI strategy and only 35% say their manager is an AI champion; 75% trust AI more than their manager for some tasks. Transcript: Nathaniel Whittemore And 76% of the C-suite said employee sabotage poses a serious threat to their company’s future. 35% of employees said that they’d entered proprietary, confidential, or sensitive information into a public AI tool. And two-thirds of executives said that they believed their company had already suffered a data leak or security breach because of an employee using an unapproved AI tool. And if you want to get a sense of why, I think you have to point to a gap in leadership. Just 35% of employees said that their manager is an AI champion. 75% said that they trust AI more than their manager for certain work tasks. That is just an incredibly damning statistic that is showing up downstream, I think, in everything else. And increasingly what you’re getting then is effectively two tiers in the workplace. 92% of the C-suite said that they’re actively cultivating a new class of AI elite employees, with 60% planning to lay off employees who can’t or won’t use AI. AI super users are about 3x more likely to have gotten both a promotion and a pay raise in 2025 compared to those who aren’t using AI. (Time 0:23:26)
  • Spending Skews Toward Tools Not People
    • Investment heavily favors tools over people, creating a mismatch that hinders adoption.
    • Aggregated data shows ~93% of AI spend goes to infrastructure/models/tools while only 7% is invested in the humans using them. Transcript: Nathaniel Whittemore This reflects something that we found when we were aggregating surveys for the maturity maps, that something like 93% of all AI spending goes to infrastructure and models and compute And tools, compared to just 7% invested in the humans using those things. That is a recipe for disaster, and the disaster is showing up in the data. Which brings us back to a theme which is going to come up in a different way on Monday’s episode about harness engineering. The quintessential lesson of the last year of AI adoption in the enterprise is that picking the tools and getting access to the models is not enough. The companies that are seeing results and getting value out of AI are designing systems and structures that support its use and support the people using it. I call this episode the excited anxiety of enterprise AI because the people who have gone all in using Claude Code or OpenClaw or Cowork or Codex or any of these other tools genuinely Go to sleep and wake up feeling like they have superpowers. And yet for everyone else, they feel increasingly adrift, at risk of obsolescence. (Time 0:25:09)
  • Invest In Systems That Support AI Users
    • Design systems and structures to support people using AI, not just buy tools.
    • Nathaniel emphasizes companies seeing value build frameworks and support around users to translate model access into actual work improvements. Transcript: Nathaniel Whittemore The quintessential lesson of the last year of AI adoption in the enterprise is that picking the tools and getting access to the models is not enough. The companies that are seeing results and getting value out of AI are designing systems and structures that support its use and support the people using it. I call this episode the excited anxiety of enterprise AI because the people who have gone all in using Claude Code or OpenClaw or Cowork or Codex or any of these other tools genuinely Go to sleep and wake up feeling like they have superpowers. And yet for everyone else, they feel increasingly adrift, at risk of obsolescence. (Time 0:25:32)