If AI Eats The Routine, What Human Skills Survive?

The hype cycle is over; the accountability era has arrived. We unpack how gen AI has moved from pilots to proof, with daily use now common among senior leaders and measurable ROI becoming the standard. 

Pulling from the Wharton GBK Collective’s year‑three findings, Google Notebook LMs agents  trace the three waves of adoption and show why “accountable acceleration” is the defining theme for late 2025.

TLDR / At A Glance:

• Gen AI Usage Is Mainstream
  • 82% use Gen AI weekly (+10pp YoY)
  • 46% use it daily (+17pp YoY)
  • ChatGPT (67%), Copilot (58%), Gemini (49%) dominate
  • Interesting note: Gemini grew fastest (+9% YoY)
• 𝗥𝗢𝗜 𝗜s Now Table Stakes for Businesses
  • 72% formally track ROI metrics
  • 74% report positive returns
  • 88% expect budget increases in next 12 months
  • 60% now have Chief AI Officers - strategy has moved to the C-suite
  • 30% of Gen AI technology spending now goes to internal R&D - enterprises are building custom solutions, not just buying off-the-shelf tools

Across functions, the story is clear: practical, repeatable work is getting faster. Data analysis, meeting and document summarisation, and everyday writing see the broadest gains, while legal and operations post surprising leaps in self‑reported expertise as tools slot into structured workflows. 

Google NotebookLMs agents examine the sector split too — tech, telecoms, finance, and professional services are far ahead, while retail and complex physical operations navigate slower integration and tougher data constraints.

ROI is rising, budgets are expanding, and the spend is changing shape. With 88% planning to increase investment and many allocating £5m+ to gen AI, enterprises are shifting 30% of their budgets into internal R&D to build custom capabilities on top of sanctioned platforms like ChatGPT, Copilot, and Gemini. 

That move, from generic efficiency to defensible differentiation, raises the stakes on governance, data, and talent. 

The toughest challenges now centre on people. Leadership is consolidating responsibility with CAIO roles, access is broadening under tighter guardrails, and AI is increasingly used to manage risk. Yet a training paradox persists: lack of training is a top barrier even as training investment softens. 

Google Notebook LMs agents dig into augmentation vs skill atrophy, the scramble for advanced talent, and why many leaders expect to hire more interns for AI‑enabled entry roles. 

Our closing challenge: if automation eats the routine, which human skills will you invest in to drive the next wave of return?

Full study by The Wharton School and GBK Collective here: https://knowledge.wharton.upenn.edu/special-report/2025-ai-adoption-report/

Enjoy the conversation? 

Follow the show, share it, and leave a review to help more people find it.

Chapters

• 0:00 Setting The Aim: Beyond Hype
• 0:55 Three Waves Of Enterprise Adoption
• 1:29 Daily Use And Rising Expertise
• 2:19 Function-Specific Wins And Gaps
• 3:13 Sector Leaders And Laggards
• 3:45 Sanctioned Tools Replace Shadow IT
• 3:57 Pivot To ROI And Accountability
• 5:05 Why Tier Ones Struggle With ROI
• 5:42 Budgets Rise And Custom Builds Grow
• 6:27 Top Risks: Security, Complexity, Accuracy
• 6:45 Leadership, Guardrails, And Training
• 7:27 Skills: Augmentation Vs Atrophy
• 8:06 Talent Shortage And Training Dip
• 8:37 Interns, New Roles, And Morale
• 9:19 Final Takeaways And Human Advantage

Transcript

SPEAKER_01: 0:00

Welcome to the deep dive. Today we're really pushing past the initial hype around generative AI.

SPEAKER_00: 0:05

Aaron Powell Yeah, getting into the real details.

SPEAKER_01: 0:07

Aaron Powell Exactly. We're diving into the hard numbers on enterprise adoption, you know, as we head into late 2025. We want to move beyond just the pilot projects and maybe the vendor stories.

SPEAKER_00: 0:18

The success stories.

SPEAKER_01: 0:19

Aaron Ross Powell Right. And analyze what's actually working or maybe even stalling inside the biggest U.S. corporations.

SPEAKER_00: 0:24

Aaron Powell That's the plan. And for this deep dive, uh we're pulling heavily from the Wharton GBK Collective's year three full report. Okay. And what's key here is that this is a repeated study. They've been tracking senior decision makers in large enterprises think a thousand employees or more over time.

SPEAKER_01: 0:39

Aaron Ross Powell So consistency.

SPEAKER_00: 0:41

Trevor Burrus Consistency, yeah. The real goal is to get past those one-off wins, the anecdotes, and give you, the listener, some solid, accountable benchmarks for how enterprises are really adopting this stuff.

SPEAKER_01: 0:51

Aaron Powell Okay, let's unpack this journey then. The report frames it quite well, I think. We saw Gen AI move from what, Waze 1 in 2023? Yeah. That was the crazy exploration phase.

SPEAKER_00: 1:01

Total exploration, everyone jumping in.

SPEAKER_01: 1:03

Then wave two in 2024, which felt a bit more organized, the experimentation phase.

SPEAKER_00: 1:09

Aaron Powell Right. More structured pilots.

SPEAKER_01: 1:11

And now the report says we're firmly in wave three. They're calling it accountable acceleration.

SPEAKER_00: 1:18

Accountable being the key word there.

SPEAKER_01: 1:20

Definitely. So to make sense of it all, we're going to look at three main themes. Everyday, AI, proving value, you know, ROI, and the human capital side of things. Let's kick off with usage. Because the speed up we've seen just this past year. Well, so it's actually pretty surprising. Gen AI isn't just an idea anymore.

SPEAKER_00: 1:38

No, it's baked into the workday for many leaders.

SPEAKER_01: 1:40

Aaron Powell The numbers say 82% of leaders are using it at least weekly now. That's up 10 points.

SPEAKER_00: 1:46

Which is solid growth.

SPEAKER_01: 1:47

But the really big shift, I think, is the daily usage. That's taken a massive leap. Forty six percent are now using Gen AI every single day.

SPEAKER_00: 1:55

That's a 17-point jump from last year. Daily use. That's significant penetration.

SPEAKER_01: 2:00

It really is. And what's fascinating is how quickly the feeling of competence is growing alongside that usage.

SPEAKER_00: 2:06

Yeah, the expertise piece. 32% of leaders now call themselves experts. That's up eight points from 2024.

SPEAKER_01: 2:13

Okay, so people feel like they're getting good at this.

SPEAKER_00: 2:15

They do. And it's not just the usual suspects like the IT folks. Look at legal their self-identified expertise jumped 23 points.

SPEAKER_01: 2:23

23. In legal.

SPEAKER_00: 2:25

Yeah. And operations saw a 24-point jump.

SPEAKER_01: 2:29

Wow. That is a huge structural shift. Why legal and operations, you often think of them as maybe slower on the uptake for new tech?

SPEAKER_00: 2:38

Well, I think it signals the tools are maturing, right? Right. AI can now handle really specific structured tasks effectively in those areas.

SPEAKER_01: 2:46

Aaron Powell Tasks that maybe needed new human skills before.

SPEAKER_00: 2:49

Exactly. For legal, think document management, reviewing standard terms, even drafting initial contract versions. Okay. And for operations, it's things like supply chain optimization, analyzing huge data sets, the kind of work where AI can provide immediate practical value. The return on learning the tool is fast.

SPEAKER_01: 3:06

Aaron Powell So if they're using it daily, where are they seeing these wins? Like what are the actual tasks?

SPEAKER_00: 3:11

It seems to be really solidifying around those practical, repeatable office tasks, things that boost employee productivity directly. Like what? The top three, both most used and highest rated, are number one, data analysis, 73% adoption there.

SPEAKER_01: 3:28

Okay. Big one.

SPEAKER_00: 3:29

Number two, document and meeting summarization at 70%.

SPEAKER_01: 3:34

The classic use case.

SPEAKER_00: 3:35

And number three, general document editing and writing, 68%. So it confirms teams are, you know, folding Gen AI into their existing workflows. Less about massive transformation right now.

SPEAKER_01: 3:47

Aaron Powell More about saving time on the mundane stuff.

SPEAKER_00: 3:49

Aaron Powell Buying back time, exactly.

SPEAKER_01: 3:50

Aaron Powell But you mentioned specific functions finding more tailored uses to beyond just summaries.

SPEAKER_00: 3:55

Aaron Powell Absolutely. We can actually measure that using an index score, how much more a function uses a tool compared to the average. So for instance, IT indexes really high on code writing and generation. Their score is 123. Trevor Burrus, Jr.

SPEAKER_01: 4:06

Meaning they use it 23% more than average.

SPEAKER_00: 4:08

Aaron Powell Precisely. And legal, building on that expertise wave we talked about, they index at 133 for generating legal contracts.

SPEAKER_01: 4:16

Trevor Burrus, Jr.: 133, that's specialized.

SPEAKER_00: 4:18

Trevor Burrus, Jr.: Yeah, that's a clear sign. It's moving beyond general office help into high-value, function-specific work.

SPEAKER_01: 4:24

Aaron Powell Okay, but here's the maybe the critical point. This adoption isn't happening evenly everywhere, is it? There are still gaps.

SPEAKER_00: 4:32

Oh, definitely. And that divide seems persistent, maybe even widening. Some functional areas like marketing sales and operations have kind of lagged since 2023.

SPEAKER_01: 4:40

Aaron Powell Still. Why do you think that is?

SPEAKER_00: 4:42

It could be data sensitivity in marketing, perhaps, or the complexity of integrating with physical processes and operations. It varies.

SPEAKER_01: 4:49

And industry-wise.

SPEAKER_00: 4:50

We see the digital sectors leading the charge tech telecom, banking finance, professional services, they're all at 90% or more weekly use.

SPEAKER_01: 4:58

The usual leaders.

SPEAKER_00: 4:59

Right. But then you look at sectors like retail, down at 63% weekly use, manufacturing is at 80%.

SPEAKER_01: 5:05

So a real spread.

SPEAKER_00: 5:06

A big spread. And critically, about one in six decision makers are still what the report calls laggards, using it less than weekly.

SPEAKER_01: 5:14

One in six.

SPEAKER_00: 5:15

Yeah. And they often face more restrictions internally, lower trust in the tech. They really do risk getting left behind from a productivity standpoint. So it's clearly not niche anymore. Is the report confirming that companies are moving past that shadow IT phase? You know, are they using sanctioned paid platforms now?

SPEAKER_01: 5:34

Absolutely. Usage for the big three Chat GPT, Copilot, Gemini, it's all up since 2024.

SPEAKER_00: 5:40

Okay.

SPEAKER_01: 5:41

And here's the key indicator: the vast majority of these subscriptions are employer paid now.

SPEAKER_00: 5:46

Uh, so it's official software.

SPEAKER_01: 5:48

Exactly. It's moved from personal tinkering to budgeted, sanctioned enterprise software. And that brings security oversight with it.

SPEAKER_00: 5:56

Okay, let's shift gears. Section two, proving value. The whole market conversation feels different now, doesn't it? We're kind of past the initial FOMO, the excitement phase.

SPEAKER_01: 6:06

Totally. The pivot to accountability that's really the defining feature of this wave three. ROI measurement is becoming standard practice.

SPEAKER_00: 6:13

They're actually measuring it properly now.

SPEAKER_01: 6:14

Seems so. Nearly three-quarters, 72% say they're tracking structured business-linked ROI metrics.

SPEAKER_00: 6:20

Like what kind of metrics?

SPEAKER_01: 6:21

Things tied to the bottom line. Profitability, throughput, workforce productivity. It's moving the question from is it cool to did it actually help us? And are they seeing results?

SPEAKER_00: 6:31

Well, based on these metrics, the outlook seems overwhelmingly positive. 74% report seeing positive ROI already.

SPEAKER_01: 6:39

Already. That's fast.

SPEAKER_00: 6:40

It is. And looking ahead, four out of five expect positive returns within the next two to three years. That shows real conviction, I think.

SPEAKER_01: 6:49

74% positive now. That's a strong number. But we need to dig into that a bit, right? Because the report also notes something interesting about the really big companies.

SPEAKER_00: 6:58

Yes, the nuance. If ROI is so clear, why are those huge Tier 1 enterprises, we're talking over$2 billion in revenue, why are they more likely to say it's still too early to tell on ROI, about 34% of them.

SPEAKER_01: 7:12

That's the key question, isn't it? If the biggest players are struggling to measure the return, is that overall 74% positive figure may be a bit skewed by smaller, nimbler companies?

SPEAKER_00: 7:22

That's likely part of the story. It points to the strategic takeaway. Agility matters. Smaller tier two and tier three firms often report quicker ROI because, well, they're just more agile. They have less legacy tech, fewer bureaucratic layers to cut through. For the giant tier one companies, integrating this across thousands of employees, complex legacy systems, it just takes longer. They see the potential, but the payback timeline stretches out.

SPEAKER_01: 7:48

And does industry play a role here too?

SPEAKER_00: 7:50

Definitely. We see that same pattern again. Returns are strongest in those digital heavy sectors like Tech Telecom, 88% positive ROI there. Wow. But then you look at sectors with complex physical operations like retail, only 54% positive ROI reported. Manufacturing is better at 75%.

SPEAKER_01: 8:08

Aaron Powell So the low-hanging fruit for ROI is really in optimizing digital processes.

SPEAKER_00: 8:13

That way, yes.

SPEAKER_01: 8:14

Okay. So even with that nuance, the early returns look promising overall. Budget confidence must be, well, pretty high.

SPEAKER_00: 8:21

Aaron Powell It's through the roof. 88% expect their Gen AI spending to increase in the next 12 months.

SPEAKER_01: 8:26

88%.

SPEAKER_00: 8:27

Yeah, that's a 16-point jump just from last year. And get this two-thirds of enterprises are now budgeting$5 million, or more specifically for Gen AI tech.

SPEAKER_01: 8:37

$5 million plus. So where's that money going? Is it just paying for more licenses or are things shifting?

SPEAKER_00: 8:42

This is where it gets really interesting, I think. It speaks to building competitive advantage.

SPEAKER_01: 8:46

How so?

SPEAKER_00: 8:47

Well, according to the IT function surveyed, a pretty significant chunk, 30% of the Gen AI tech budget, is now going into internal RD research and development.

SPEAKER_01: 8:58

Aaron Powell 30% on internal RD.

SPEAKER_00: 9:00

Uh-huh.

SPEAKER_01: 9:00

So they're not just using AI, they're building their own stuff.

SPEAKER_00: 9:03

Exactly. It suggests firms are moving beyond just using off-the-shelf tools. They're creating custom, tailored AI capabilities, likely for their own unique internal processes.

SPEAKER_01: 9:14

Aaron Powell That's a big shift from just efficiency gains.

SPEAKER_00: 9:17

It is. It suggests they're prioritizing building unique advantages, things competitors can't easily copy rather than just standardizing on common tools.

SPEAKER_01: 9:25

Aaron Powell Okay. Fascinating. Before we move to the people side, what about the hurdles? The steed bumps, we know the benefits, efficiency, qualities of that. But what are leaders still worried about?

SPEAKER_00: 9:35

Aaron Powell Well, the top barrier hasn't changed much. Security risks are still number one. That's been a constant concern.

SPEAKER_01: 9:41

Aaron Powell Understandable.

SPEAKER_00: 9:42

Followed by operational complexity actually integrating this stuff, and then the accuracy or sometimes inaccuracy of the results.

SPEAKER_01: 9:49

Right, the hallucination problem.

SPEAKER_00: 9:51

But there's a new entry in the top ten concerns this year, which is telling. Lack of training resources.

SPEAKER_01: 9:57

Ah, lack of training. That leads us perfectly into the next section, doesn't it? The human side. Right. Section three, the human capital lever. It feels like the bottleneck has really shifted now. It's less about the tech itself.

SPEAKER_00: 10:10

And much more about the people, the skills, the organizational structure.

SPEAKER_01: 10:14

And on that structure point, leadership seems to be consolidating, getting more formal.

SPEAKER_00: 10:18

That's a very clear trend. It shows accountability coming from the top. Executive leadership focus on Gen AI has surged year over year.

SPEAKER_01: 10:27

How much?

SPEAKER_00: 10:28

Well, 60% of enterprises now have a chief AI officer or a similar dedicated role leading the charge.

SPEAKER_01: 10:34

60% have a CAIO or equivalent. Wow.

SPEAKER_00: 10:37

Yeah, though it's important to note, over half of those are existing execs taking on new AI responsibilities rather than brand new hires, necessarily.

SPEAKER_01: 10:45

Still, it shows focus. And what about access versus control? Are they letting more people use it, but maybe with tighter rules?

SPEAKER_00: 10:52

That's exactly the balance they're striking. Access is definitely broadening. 70% of firms now let all employees use Gen AI in some capacity.

SPEAKER_01: 11:01

70%. Broad access.

SPEAKER_00: 11:02

But the guardrails are tightening significantly alongside that. 64% have adopted formal data security policies specifically for AI. Okay. And 61% are rolling out mandatory employee training and awareness programs.

SPEAKER_01: 11:16

Training is becoming key.

SPEAKER_00: 11:18

And interestingly, leaders are also starting to use AI to manage risk itself, kind of flipping the script.

SPEAKER_01: 11:25

Using AI to manage AI risks.

SPEAKER_00: 11:27

And other risks too. 62% are using AI for risk management, especially in IT security and financial risk assessment.

SPEAKER_01: 11:34

Okay, that's smart. Now this brings us to that central tension, the future of work question. Is AI enhancing skills replacing them or causing them to atrophy?

SPEAKER_00: 11:43

Well, the dominant view is still positive, leaning towards enhancement. 89% agree that Gen AI enhances employees' skills.

SPEAKER_01: 11:50

89%, that's high.

SPEAKER_00: 11:52

Which supports that idea of augmentation, working alongside humans. More people agree with enhancement 89% than agree it replaces employees, 71%.

SPEAKER_01: 12:02

Okay, so augmentation wins out, but there's a but, isn't there?

SPEAKER_00: 12:06

There is a persistent note of caution. Forty-three percent of leaders are now specifically warning about the risk of skill atrophy.

SPEAKER_01: 12:14

Skill atrophy, meaning people forget how to do things manually.

SPEAKER_00: 12:18

Essentially, yeah. A decline in proficiency because the AI is doing it. And interestingly, this concern is actually higher among VPs than mid-managers.

SPEAKER_01: 12:25

Why the difference?

SPEAKER_00: 12:26

VPs might have a better, maybe broader view of day-to-day skill practice across teams. They might be seeing the potential loss of that muscle memory more clearly.

SPEAKER_01: 12:35

Okay, but here's where it feels like a massive contradiction. Leaders are worried about skill loss. They flagged lack of training resources as a top barrier.

SPEAKER_00: 12:44

Right.

SPEAKER_01: 12:44

Yet the report says actual investment in training has softened. It fell eight percentage points.

SPEAKER_00: 12:50

It's a genuine strategic puzzle, isn't it? A misalignment. Companies know they need skilled people. Recruiting folks with advanced gen AI technical skills is a top challenge for nearly half of them, 49%.

SPEAKER_01: 13:01

Almost half need advanced talent.

SPEAKER_00: 13:03

Yet they seem to be pulling back, or at least not increasing, investment in training the people they already have.

SPEAKER_01: 13:09

So what's the strategy then?

SPEAKER_00: 13:11

It suggests organizations are kind of split, maybe unsure whether to bet on massive internal upskilling programs for current staff.

SPEAKER_01: 13:19

Or just try to buy the talent they need through aggressive hiring.

SPEAKER_00: 13:23

Exactly. It seems like they're hedging their bets, or maybe haven't decided on the primary path forward yet.

SPEAKER_01: 13:27

Aaron Powell And speaking of hiring, if we look specifically at the VP level, there's another weird paradox around hiring for junior roles, right?

SPEAKER_00: 13:35

Aaron Powell Yeah. Senior leaders are surprisingly divided on this. You'd think AI might reduce the need for interns, right? Automating basic tasks. Aaron Powell That's the common fear. And 17% do expect to hire fewer interns because of AI. Aaron Powell Okay.

SPEAKER_01: 13:48

Some alignment there.

SPEAKER_00: 13:49

But a much larger group, 49%, almost half actually anticipate hiring more interns.

SPEAKER_01: 13:54

Aaron Powell More interns?

SPEAKER_00: 13:55

Why? The thinking seems to be that while Gen AI automates some basic stuff, it also creates new kinds of entry-level work. Maybe roles focused on managing the AI tools, prompting, checking outputs, integrating AI into workflows.

SPEAKER_01: 14:09

Aaron Powell So new AI-enabled junior roles.

SPEAKER_00: 14:11

Aaron Powell Potentially, yeah. Roles that organizations need to fill quickly to actually leverage the AI effectively. Aaron Powell Okay.

SPEAKER_01: 14:17

So after looking at the tech stack, the tools, the ROI, what's the final verdict? What's the real bottleneck holding back even faster progress?

SPEAKER_00: 14:26

Aaron Powell It's landed exactly where many predicted it would. It's about people, talent, and culture.

SPEAKER_01: 14:31

The human element.

SPEAKER_00: 14:32

Precisely. The toughest challenges leaders report now are recruiting that advanced technical talent, that's 49%, providing effective training for everyone else, 46%. Okay. And maybe the trickiest one for long-term success. Maintaining employee morale, especially in roles being significantly impacted by Gen AI, that's at 43%. Morale.

SPEAKER_01: 14:51

That's huge.

SPEAKER_00: 14:52

It is. The human side of the equation is now clearly the main constraint on how fast this accountable acceleration can really happen. Hashtag tag outro.

SPEAKER_01: 15:00

So wrapping this deep dive up, it seems clear Gen AI adoption in the enterprise is, well, it's fully mainstream now.

SPEAKER_00: 15:07

No longer niche.

SPEAKER_01: 15:08

And crucially, companies are actually measuring and seeing structured ROI today. We've definitely moved from just pilots to focusing on real performance metrics.

SPEAKER_00: 15:19

The accountability phase is real.

SPEAKER_01: 15:20

But it feels like the next wave of scalable performance, getting even more value out of this that hinges almost entirely on getting the people part right. The skills, the processes, the governance.

SPEAKER_00: 15:33

It's not just about having the fanciest tools anymore.

SPEAKER_01: 15:36

Exactly.

SPEAKER_00: 15:36

Which brings us to maybe a final thought for you, the listener, to consider. We flagged that finding. 30% of gen AI tech budgets are going into internal RD.

SPEAKER_01: 15:45

Right, building custom solutions.

SPEAKER_00: 15:46

Custom solutions, likely including things like AI agents designed for automating specific internal processes and workflows. Early findings suggest maybe 58% of enterprises are already using these agents. So if enterprises are investing heavily in building custom automation to replace repetitive tasks, and we know leaders are already worried about skill proficiency declining, 43% are concerned about skill atrophy.

SPEAKER_01: 16:09

Where are you going with this?

SPEAKER_00: 16:11

Well, the provocative question for you is in this next phase of accountable acceleration, with automation taking over more routine work, what are the uniquely human, non-technical skills that you personally need to focus on protecting and developing?

SPEAKER_01: 16:25

Ah, the skills AI can't replicate easily.

SPEAKER_00: 16:27

Exactly. Things like creativity, critical thinking, complex strategic judgment, empathy, collaboration. Because if the routine stuff gets automated, those higher order human skills, they become the real key levers for driving future ROI. They become truly priceless.