AI is moving faster than anyone predicted.
In a massive new study analyzing 1,000 jobs and nearly 20,000 tasks, Cognizant found that 93% of jobs are already impacted by AI … with $4.5 trillion in U.S. labor value potentially automatable today.
But here’s the twist: AI isn’t replacing entire jobs. On average, only 39% of a role’s tasks can be automated. The future isn’t AI alone: it’s humans plus AI.
But will it be fewer humans?
- Get the deepest insights concisely on the TechFirst Substack newsletter
- Subscribe to the TechFirst YouTube channel to never miss an episode
And, watch our conversation here:
In this episode of TechFirst, host John Koetsier sits down with Babak Hodjat, CTO of Cognizant, to unpack:
- Why construction and transportation are seeing surprising AI growth
- Why programming jobs may have hit an automation plateau
- What “agentic AI” actually means — and why it matters
- How management roles are more automatable than we thought
- The rise of vibe coding and democratized software creation
- Why compute power, not ideas, may be the biggest bottleneck
We also explore how companies can safely capture AI’s upside, why training matters more than ever, and what happens when digital twins, LLMs, and human expertise combine.
This isn’t hype. It’s a data-driven look at where AI is actually changing work right now.
Transcript: AI, jobs, and the $4.5 trillion impact
John Koetsier:
Is AI going to take your job? Well, there’s good news and there’s bad news. The bad news is that the AI invasion is happening way faster than we anticipated. The good news is it can’t do everything. And in fact, there are some things where we can work with it to do more than we otherwise thought we could.
Hello and welcome to Tech First. My name is John Koetsier. Cognizant just released a major AI report.
AI is moving way faster than we anticipated. Ninety-three percent of all jobs—they studied a thousand jobs—will be impacted. And AI can already unlock $4.5 trillion in value. That’s absolutely huge. That’s a few percent of total world GDP. It’s also growing fast, even in careers like construction, which we might have assumed were safe. But it can’t do everything. People still matter, at least for now.
To chat more and dive into the data, we’ve got Cognizant CTO Babak Hodjat. Welcome, Babak. How are you doing?
Babak Hodjat:
Great, thank you very much.
John Koetsier:
Super pumped to have you with us today. Thank you so much. Big report. You looked at a thousand jobs, 20,000 different tasks. You figured out what AI can do, what it can’t do, where it fits, where it doesn’t fit. What were the biggest surprises in the data for you?
Babak Hodjat:
Well, we had done an analysis in 2023, actually, and this was revisiting that analysis. As you said, it was on almost 20,000 tasks across a thousand jobs—basically the jobs that are reported in the U.S. annually.
The surprise was that some of the tasks we were expecting to be automated later on are already being automated. Some of the breakthroughs, for example in agentic AI and multimodal AI and so forth, are already impacting some of these tasks.
There were also surprises the other way, where certain jobs we expected to be exposed more seem to have hit sort of a plateau in their exposure, which underscores the role of humans. But in general, the rate of exposure, the rate of automation, and the tasks we can now augment using AI have gone well beyond what we expected.
John Koetsier:
Give some examples of jobs you thought would be more automated that are not at the top of the list.
Babak Hodjat:
Some of the jobs we expected to be at higher automation—jobs in computers, programming, and mathematics—we expected them to be exposed more, but they are not as exposed as we anticipated. That was a bit of a surprise. We think that’s part of what I mentioned earlier, this plateau that we’re seeing.
John Koetsier:
Plateau, right? Because LLMs came in, knocked the door down, everything exploded. It’s amazing and incredible. And yet there’s a limit to what they can do. They’re impressive—the latest ChatGPT, Gemini, all that—but they still make basic errors and can’t do everything.
Babak Hodjat:
I should also add that these are areas where AI has been adopted faster. AI folks are programmers themselves, and most of our AI benchmarks relate to programming. So they’ve been adopted faster and look like they’ve saturated in adoption.
Math is similar. There are many reasoning and math benchmarks. In other jobs, we don’t necessarily have the same benchmarks, and adoption hasn’t been as fast. There’s a difference between what’s automatable and what’s actually being adopted.
John Koetsier:
It was interesting to see that some jobs most of us probably thought were safe—like construction—are being impacted more significantly. My assumption is AI can help a carpenter figure out spacing, framing, materials. Are there other examples?
Babak Hodjat:
Yes, very surprising areas like labor and transportation. It’s still relatively small in overall exposure, but it’s growing significantly. In transportation, we saw a rise from 6% exposure in 2023 to 25%. In construction, from 4% to 12%. Those are big jumps.
We have self-driving cars, and AI systems are starting to prove themselves. Since 2023, we’ve also been able to give tools to AI—that’s what agency is. Agentic systems can actually do things on our behalf. That explains a large portion of the disruption we’re seeing.
John Koetsier:
You can see that with self-driving cars and drone delivery, which is already real in parts of the U.S. and Europe. There’s route management, solving the traveling salesman problem, all of that.
So 93% of jobs are impacted, but the average percentage of a job impacted is 39%. Talk about that difference.
Babak Hodjat:
We look at jobs as collections of tasks. Some tasks are automatable or impacted by AI, but not all tasks within a job can be automated today. That’s why the percentage at the job level is smaller.
It also points to the fact that it’s humans plus AI, not AI alone.
John Koetsier:
Right, because if my job is 39% automatable, someone still needs to connect the dots. You may not have an agentic system with sufficient contextual knowledge.
Babak Hodjat:
Exactly. Humans are very contextual. We have effective memory and deep expertise in our niches. And creativity. There are fundamental limitations in today’s state of the art that prevent AI from handling everything.
John Koetsier:
I noticed management can’t be fully automated, although many CEO tasks can be.
Babak Hodjat:
That was a surprise. AI doesn’t really care where you sit in the management pyramid. Systems are better at reasoning and long chains of decision-making autonomously.
Tasks like budget allocation, pipeline review, scheduling—things we think of as strategic human endeavors—can be heavily augmented or automated.
John Koetsier:
One big use case is planning scenarios. What if I allocate here? What if I invest there? Seeing the potential outcomes.
Babak Hodjat:
Correct. Large language models give us a generalist digital twin of the world. They offer breadth but not deep contextual knowledge of your specific organization.
You need to feed that context in, and input windows are limited. Other machine learning approaches can create more contextual digital twins, but they require significant engineering and data.
The combination of a generalist LLM, specialized machine learning tools, and a domain expert human creates much stronger decision-making.
John Koetsier:
You can imagine an LLM trained deeply on economics, finance, business journals, company data. It won’t be perfect but could be incredibly helpful.
The top-line number caught my attention: $4.5 trillion in labor AI can handle today. That’s U.S. only, correct?
Babak Hodjat:
Yes, U.S. work tasks that are automatable today.
John Koetsier:
U.S. GDP is around $30 trillion. That’s about 15–16% of GDP. That’s huge.
Babak Hodjat:
It’s potential. Adoption isn’t there yet. Some sectors are growing faster than others. Also, innovation in AI has been so rapid—new breakthroughs every few months—that this $4.5 trillion potential will likely grow.
John Koetsier:
Globally, that could be massive. World GDP is about $125 trillion. Maybe it’s 20%, 30%, even 50% in a decade. That’s world-changing. And we haven’t even talked about humanoid robots.
Babak Hodjat:
The report does consider embodied AI progress, but that area is still catching up. There’s significant research happening.
Internationally, countries may adopt faster because they learn from early adopters like the U.S. However, processing capacity is currently a limiting factor. AI systems are compute-heavy.
The good news is that we’re optimizing rapidly. It’s like early computing—entire rooms for a few kilobytes of memory. Today, we carry powerful multiprocessor systems in our pockets. We’ll get there. It’s only been three or four years of rapid progress.
John Koetsier:
That explains the massive rush to build data centers globally. There’s enormous labor potential to automate.
Let’s make this human. How does a company realize the most value from AI?
Babak Hodjat:
It’s about empowerment. That means bringing in partners, innovating internally, giving people access to tools, and allowing them to decide how to augment their work safely.
Safety is critical. AI can be implemented in risky ways or in disciplined, productive ways.
Empowerment also means skilling. We have a strong training heritage at Cognizant and continue investing in it.
Last year, we broke a world record in vibe coding. Forty percent of the 50,000-plus participants in our hackathon had never written a line of code professionally.
There was Cognizant before that event and Cognizant after. People recognized AI’s power, its limitations, and where they fit in. Training and hackathons are powerful preparation tools.
John Koetsier:
I’ve written code, though I wouldn’t call myself a developer. I’m doing some vibe coding for apps and SaaS projects, and it’s mind-blowing.
A CTO might review it and find lots of garbage code. I suspect vibe-coded apps contain a high percentage of messy code. But it’s impressive.
Babak Hodjat:
That’s true. But code is becoming more fungible. You can generate it, test ideas, then have professionals refine it.
I remember 30 years ago talking to assembly programmers who refused to use Fortran because it produced inefficient compiled code. We’re seeing the same pattern again. But it opens many doors.
John Koetsier:
Everything old is new again. For one app I’m building, I’m sure 80% of the code is garbage. But maybe that’s fine—it’s a prototype. Extract the spec, rebuild cleanly.
It’s a wonderful, weird, wild world. Thank you for taking the time to explain what’s happening.
Babak Hodjat:
My pleasure. Thank you very much for having me.