We Moved Fast On AI; Now We Need Brakes

The adrenaline rush is gone and the lights are on. Google Notebook LMs agents dig into Deloitte’s latest State of AI in the Enterprise and confront a tough truth: access exploded, but value is uneven and the governance gap is widening. 

Instead of more shiny pilots, 2026 demands systems thinking, economic rigour, and clear decision rights as AI moves from chat to action.

At a Glance / TLDR:

• access rising but daily usage lagging
• pilot success versus production economics
• three tiers from surface gains to deep transformation
• revenue gap between savings and new income
• job redesign, broken ladder, and pod-based teams
• sovereign AI, local models, and data control
• agentic AI, tool use, and governance deficits
• physical AI growth in APAC and safety needs
• 2026 as a friction year demanding brakes

The podcast starts with the usage gap - why sanctioned tools sit idle - and trace the roadblocks that turn successful sandboxes into expensive production failures. From latency and cost blowouts to brittle data pipelines, we unpack what it takes to move beyond proof-of-concept purgatory. Then we map the three tiers of adoption: surface-level productivity, process redesign, and deep transformation. A standout case turns mining equipment into connected platforms, shifting from digging to predictable, data-driven extraction. That’s the leap from automation to imagination, and it’s where new revenue lives.

The conversation gets candid on jobs. When models make the call, humans can’t be left as rubber stamps. We explore role redesign, escalation rules, explainability, and the “broken ladder” problem created by automating entry-level tasks. A promising answer is pod-based teams - small cross-functional units orchestrating fleets of AI agents - where learning shifts from manual repetition to supervision and exception handling. We zoom out to sovereign AI and the rise of compact local models that run under domestic rules, balancing control, privacy, and latency with the realities of global operations.

Agentic AI is the tipping point: systems that plan, act, transact, and iterate toward goals. The value compounds, but so does the blast radius of mistakes. With 74 percent planning agents soon and only 21 percent ready on governance, we lay out practical brakes: scoped permissions, human-in-the-loop gates, immutable logs, simulator testing, budget limits, and kill-switches. We also scan physical AI - robots and drones scaling fastest in APAC - where safety and uptime meet AI reliability.

If you’re leading AI adoption, ask three things:

1. Are we transforming what we sell, not just how we work? 
2. Do we know who overrules the model and when? 
3. And have we built the brakes for autonomy before hitting the gas? 

Subscribe, share with a teammate who owns the roadmap, and tell us: what’s the first brake you’ll install?

Transcript

From Hype To Hard Work

Google Agent 1 0:00

Welcome back to the deep dive. It's uh January 25th, 2026.

Google Agent 2 0:04

Already. Wow.

Google Agent 1 0:06

I know, right? And I have to say, looking at that date, it feels like the air has changed a bit, doesn't it?

Google Agent 2 0:12

It definitely feels different, less um hype.

Google Agent 1 0:16

That's it. I mean, think back to 2023, maybe 2024. The whole vibe was just pure adrenaline. It was all magic tricks.

Google Agent 2 0:24

Look, this bot can write us on it.

Google Agent 1 0:25

Exactly. Look, this can make a picture of a cat in a spacesuit. It was, you know, it was a party.

Google Agent 2 0:29

It was the honeymoon phase, all the excitement, none of the responsibility. Trevor Burrus, Jr.

Google Agent 1 0:33

Right. But we're in 2026 now. The party's over, the lights are on, and someone's sweeping up the confetti, and all these businesses are standing there with a very, very expensive bill asking.

Google Agent 2 0:43

Okay, that was fun. Now what?

Google Agent 1 0:45

How do we actually make money with this?

Google Agent 2 0:47

That's the perfect way to frame it. I mean, we've moved from the wow era to the work era. And frankly, the work era is just a lot messier.

Google Agent 1 0:53

Aaron Powell, which brings us to today's focus. We are breaking down Deloitte's brand new state of AI in the enterprise report. And this is not a small sample.

Google Agent 2 1:01

Aaron Powell No, this is a big one. Over 3,200 leaders. 24 countries.

Google Agent 1 1:06

Aaron Powell So it's a real global pulse check. And they've titled this year's theme, The Untapped Edge, which I have to be honest, sounds a little optimistic to me.

Google Agent 2 1:15

Aaron Powell It is optimistic, but it's also, I think, a little frustrated. Untapped means the potential just sitting right there, staring us in the face.

Google Agent 1 1:23

Aaron Powell, but we haven't figured out how to grab it yet.

Google Agent 2 1:25

Exactly. And let's talk about why, because the headline numbers, they look great at first glance. Trevor Burrus, Jr.

Google Agent 1 1:31

Right. Access to AI tools jumped by 50% in a single year. That sounds like a massive success story.

Google Agent 2 1:37

On paper, yeah. About 60% of workers now have sanctioned official AI tools. Yeah. IT departments, they did their job, they got the licenses, installed the software. The access problem is, you know, pretty much solved. Aaron Powell Okay.

Google Agent 1 1:50

I'm sensing a big butt coming. Right. Just because I have a gym membership doesn't mean I'm getting fit.

Google Agent 2 1:54

That's it exactly. The report points to this massive usage gap. So despite having access, fewer than 60% of those workers are actually using the tools in their daily workflow.

Google Agent 1 2:05

Wow. So that's a huge disconnect. Companies are spending millions, and the employees are what? Just ignoring them.

Google Agent 2 2:11

Aaron Powell They're falling back on old habits. It's the difference between, well, let's say activation, just swishing the lights on, and transformation, actually changing how you move around the room.

Google Agent 1 2:23

Aaron Powell And that's what we need to get into today. This whole idea of why the pilots aren't turning into production, why the usage isn't turning into value, and you know what this is all doing to the actual people caught in the middle.

Deloitte’s Global AI Pulse

Google Agent 2 2:34

Let's start there with that failure to launch idea.

Google Agent 1 2:37

Yeah. Their report uses this phrase that honestly sent a shiver down my spine the proof of concept track.

Google Agent 2 2:44

It sounds ominous, but it's a very real problem. Organizations have been running hundreds of these little experiments for the last couple of years.

Google Agent 1 2:51

Little pilots. Let's see if AI can summarize this legal brief, that kind of thing.

Google Agent 2 2:56

Right. And in isolation in a lab, those pilots almost always succeed.

Google Agent 1 3:00

Of course they do, because it's a sandbox. Nothing can really go wrong in a sandbox.

Google Agent 2 3:05

It's a vacuum. You're using clean static data. You've got a small team that's hand holding the model. But the second you try to take that and move it into production, hit a wall. A huge wall.

Google Agent 1 3:16

Yeah.

Google Agent 2 3:16

Because the real world is messy. In a pilot, you don't really have to worry about data drift or latency.

Google Agent 1 3:23

Latency, you mean speed.

Google Agent 2 3:24

Speed. In a pilot, if the AI takes 15 seconds to give you an answer, you say, wow, amazing. But in a live customer service chat, 15 seconds is an eternity.

Google Agent 1 3:34

The customer's gone, they've hung up.

Google Agent 2 3:36

They're gone. So to make it fast enough, you suddenly need 10 times the computing power, which means 10 times the cost. Right. And the CFO looks at the bill and says, hold on. The cost to run this thing is higher than the value it's supposed to be giving us. And that's the trap. It works technically, but it fails economically.

Google Agent 1 3:53

That makes so much sense. It's like building a prototype car in your garage versus building a whole factory. Trevor Burrus, Jr.

Google Agent 2 3:59

And the data, it backs this up. Only 25% of companies have moved 40% or more of their experiments into production to date.

Google Agent 1 4:06

Only 25%? That's painfully low. It feels like we're stuck in what you could call pilot purgatory.

Google Agent 2 4:12

We are. However, and this is where things get interesting 54% expect to hit that same mark within the next three to six months.

Google Agent 1 4:20

Whoa. So we are basically sitting on a coiled spring.

Google Agent 2 4:24

That's my read on it, yeah. 2026 is the year of scaling up. The pilots are done, the learnings are in. The companies that succeed this year will be the ones that stop chasing the shiny object and start doing the boring hard work of building real infrastructure.

Google Agent 1 4:39

It's not as sexy, but I guess plumbing is never sexy until the toilet breaks. Okay, so speaking of the bottom line, let's talk about money. The report breaks down AI usage into three tiers, which I found fascinating. It really separates the tourists from the locals.

Google Agent 2 4:53

This is that whole productivity versus reimagination debate. So tier one is what they call surface level. That's about 37% of companies.

Google Agent 1 5:00

Surface level sounds like I'm using AI to write my emails, but my actual job is exactly the same?

Google Agent 2 5:06

Pretty much. You're just adding a layer of AI frosting on top of the same old cake. Then you have group two, about 30%, doing process redesign.

Google Agent 1 5:15

So that's more like changing the recipe, but it's still the same cake?

The Usage Gap And Pilot Purgatory

Google Agent 2 5:19

Yes. Redesigning key processes, but the business model stays the same. But then you have the third group, the 34%, and this is the holy grail, deep transformation.

Google Agent 1 5:29

Okay, give me an example of that, because transformation is such a buzzword. It usually just means someone painted the walls.

Google Agent 2 5:34

So the report talks about a mining company. Now, mining, you think big trucks, dirt, heavy machinery.

Google Agent 1 5:41

Right. Not exactly Silicon Valley.

Google Agent 2 5:44

But that's where the edge is. This company didn't just use AI to schedule shifts better, you know, for efficiency. Instead, they took their traditional mining equipment and turned it into these intelligent connected platforms.

Google Agent 1 5:56

So the truck isn't just a truck anymore, it's a rolling laboratory.

Google Agent 2 6:00

It's a node in a network. And that changes everything. You're not selling digging anymore, you're selling predictable extraction, you're selling data integration. They disrupted their own market.

Google Agent 1 6:10

And that's the difference between I saved 10 minutes and I just found a whole new revenue stream.

Google Agent 2 6:15

And that distinction is so critical because of what the report calls the revenue gap. 74% of organizations hope to grow revenue with AI. Hope. But only 20% are actually doing it today.

Google Agent 1 6:28

Ouch. That is a massive delta between hope and reality.

Google Agent 2 6:32

It is. Most companies, about 66%, are just getting those efficiency gains. They're saving money, sure, but making new money, that takes imagination, not just automation.

Google Agent 1 6:43

Well, speaking of automation, we have to talk about the elephant in the room, or maybe the robot in the cubicle. The numbers in this report on jobs, they honestly made my stomach drop.

Google Agent 2 6:52

This is the section where the numbers get really stark.

Google Agent 1 6:54

Stark is putting it mildly.

Google Agent 2 7:02

Fully automated, not augmented, replaced.

Google Agent 1 7:05

And if you look out three years, that number jumps to 82% of companies expecting that level of automation. That's not just efficiency, that is people packing up their desks.

Google Agent 2 7:14

Aaron Powell It is. But here's the disconnect. And this is the part that, as an analyst, worries me the most. Despite expecting this, this seismic shift, 84% of companies have not redesigned jobs around AI capabilities. Wait, wait, so they know that robots are coming to take the tasks, but they haven't figured out what the humans are supposed to do with all their newfound free time.

Google Agent 1 7:34

Precisely. They are sleepwalking straight into a talent crisis. The report gives this great example of a loan officer.

Google Agent 2 7:41

Aaron Powell Okay, walk me through that.

Google Agent 1 7:43

So five years ago, a loan officer's value was their judgment, their gut instinct, 20 years of experience, looking at a file and saying yes or no.

Google Agent 2 7:51

Aaron Powell They were the gatekeepers.

Google Agent 1 7:53

Now, in 2026, an AI system gives the recommendation, approve or deny, based on millions of data points, way more than a human could ever process. So the human's core value, their judgment has been outsourced.

Google Agent 2 8:06

So what's left for the human to do? Just click, okay. If the job isn't redesigned, then yes, they become a rubber stamp. And that just kills morale. But it also raises all these new questions. When do you override the AI? How do you explain an AI's decision to a customer? Companies just aren't answering that.

Google Agent 1 8:24

And there's another piece to this, right? If you automate all the entry-level work, you get the broken ladder problem.

Google Agent 2 8:30

This is critical. The biggest barrier to AI integration is insufficient worker skills. But how do you learn a job? You start with the grunt work.

Google Agent 1 8:38

Tata entry, first level support, you learn the ropes.

Google Agent 2 8:41

Exactly. If AI does all the data entry, if it answers all the simple questions, where's the training ground for junior employees? If we automate the bottom rung of the ladder, how does anyone ever climb up?

Google Agent 1 8:52

We're sawing off the bottom of the ladder while we're standing on it.

Google Agent 2 8:55

We are. And the report suggests we need entirely new structures. We're seeing this trend toward uh pod-based models. About 53% of companies are considering it.

Google Agent 1 9:05

Pod-based? What does that even look like?

Google Agent 2 9:07

Well, you don't need a manager watching 20 people do data entry anymore. Instead, you might have a small pod of three experts a strategist, a creative, a technical lead managing, a whole fleet of AI agents.

Google Agent 1 9:20

So it's less of a pyramid and more like a special ops team.

Google Agent 2 9:23

Correct. But most HR departments are still trying to hire for the pyramid.

Google Agent 1 9:27

That is a massive cultural shift. And speaking of shifts, let's zoom out from the office to the map. Another huge theme this year is sovereign AI. Usually tech is all about the global village, but this suggests AI is getting very, very territorial.

When Tech Works But Economics Fail

Google Agent 2 9:43

Aaron Powell It is. Sovereign AI is this idea that nations and companies need strategic independence. They need to build AI using their own laws, their own infrastructure, their own local data.

Google Agent 1 9:53

Aaron Powell But that feels like a complete reversal of the last 30 years. The whole promise of the internet was that it was borderless.

Google Agent 2 9:58

Aaron Powell It was. But think about it. If you're a French bank, do you really want your critical infrastructure to depend on a model that lives in a data center in Virginia? A model that's subject to, I don't know, U.S. export controls or shifting political winds.

Google Agent 1 10:11

Aaron Powell Right. You don't want to be locked out of your own brain if diplomatic relations go sour.

Google Agent 2 10:15

Trevor Burrus Exactly. So the question is shifting from how smart is the model to whose passport does this model hold?

Google Agent 1 10:24

Whose passport does the model hold? That is a wild way to think about a piece of software.

Google Agent 2 10:30

Aaron Powell But it's the new reality. 77% of companies say the location of AI development is a key factor now. 66% are concerned about relying on foreign-owned AI.

Google Agent 1 10:40

So are we seeing like a splinter net for AI?

Google Agent 2 10:43

In a way, yeah. We're seeing the rise of these distilled small language models. So instead of one giant global brain, companies are asking for smaller, local models that live on their own servers in their own country.

Google Agent 1 10:55

So I can have a model that speaks perfect French, understands French law, and never sends a single bite of data across the Atlantic.

Google Agent 2 11:01

That's the goal. It's about control. In the Americas, only 11% of companies rely on foreign sourced solutions. But in EMEA, that's Europe, Middle East, Africa, that number's up at 32%. So they're feeling the pressure to build their own capacity fast.

Google Agent 1 11:16

Digital sovereignty. It makes sense. Okay, we've talked about the where. Let's talk about the what. The tech itself is changing. The new buzzword is agentic AI. And I need you to break this down because it sounds like the robots are getting legs.

Google Agent 2 11:30

Metaphorically, they are.

Google Agent 1 11:31

Yeah.

Google Agent 2 11:32

This is a big shift from chatbots that talk to agents that act.

Google Agent 1 11:36

Okay. Give me a concrete example. How is an agent different from the chat GPT we've been using?

Google Agent 2 11:42

Think of it like this. The AI we've been using is like a really smart intern. It waits for you to give it a very specific command. Summarize this PDF. Write this poem. It does it, and then it stops.

Google Agent 1 11:54

It's passive. It waits for me.

Google Agent 2 11:55

It's passive. Agentic AI is like a seasoned employee. You give it a high-level goal, plan a travel itinerary for the team, stay within this budget, book the flights, and put it on our calendars.

Google Agent 1 12:05

And it just goes and does it.

Google Agent 2 12:06

It goes. It breaks that goal down into steps, it checks flight prices, it reasons, oh, that flight is too expensive, I need to check another airline. It accesses your calendar API, it sends emails, it can execute transactions.

Google Agent 1 12:18

That is, I mean, that's incredibly useful, but it's also a little terrifying. If an intern messes up a summary, I just rewrite it. If an agent messes up the bookings, you lose real money or you damage your reputation.

Google Agent 2 12:30

The blast radius of a mistake is just so much wider.

Google Agent 1 12:34

And are companies ready for this? Handing over the company credit card to an algorithm seems bold.

Google Agent 2 12:39

And that is the governance gap. This is probably the most crucial point in the whole report. Agents are scaling way faster than the guardrails. 74% of companies plan to use agentic AI within two years.

Google Agent 1 12:53

Two years. That's basically tomorrow.

Google Agent 2 12:55

But and this is the scary part. Only 21% have a mature model for governance.

Google Agent 1 13:00

21%. So only one in five companies has a plan for how to control this thing they're about to unleash.

Google Agent 2 13:06

Correct. We're about to hand over autonomy to software without clear boundaries.

Google Agent 1 13:09

It's like building a Ferrari engine and then putting it into a go-kart with no brakes.

Google Agent 2 13:13

And hoping the driver figures it out on the fly. And it's not just digital agents. We're also seeing physical AI. Robots, drones. 58% of companies are already using it. It's projected to hit 80% in two years.

Google Agent 1 13:28

And the report mentioned Asia Pacific is really leading the charge on that physical side.

Google Agent 2 13:32

They are, yeah. While the West is very focused on the generative, you know, the text and images, Asia is all in on the physical application. Robotics and manufacturing, autonomous logistics, they're integrating the AI brain into the robot body.

Three Tiers: Surface To Deep Transformation

Google Agent 1 13:46

Okay, so let's try to pull this all together. It's January 2026. We have access to the tools, but we're not using them. We're stuck in pilot purgatory. The workforce is facing a massive shakeup that HR isn't ready for. Nations are building AI borders, and we're about to unleash autonomous agents without knowing how to control them.

Google Agent 2 14:04

But that is the state of the enterprise in a nutshell. It's a transition year, a year of friction.

Google Agent 1 14:09

It really feels like the last few years were the dreaming phase, and now we all woke up and realized we have to actually build the house.

Google Agent 2 14:16

And the blueprint keeps changing while we're pouring the concrete.

Google Agent 1 14:19

Fantastic. So for everyone listening, what's the one thing they should be thinking about?

Google Agent 2 14:24

I want you to go back to that governance statistic. We are about to give autonomy to AI agents to make decisions and execute real tasks within two years. And only one in five companies has a plan for how to govern them.

Google Agent 1 14:38

So are we building the engine before we've built the brakes?

Google Agent 2 14:41

I fear we might be. And if you're a leader, you have to ask, is my organization just hitting the gas because everyone else is, or have we actually checked the brakes? Because when you have an agent that can act on its own, speed is great, but control is everything.

Google Agent 1 14:56

That is the question for 2026. Are you reimagining the work or are you just automating the old way of doing things and do you have the brakes ready?

Google Agent 2 15:04

Indeed. It is going to be a fascinating year.

Google Agent 1 15:07

It certainly is. Check your brakes, everyone. Thanks for listening to the deep dive. We'll see you next time.