Today, people use AI to help them do their work. Eventually, most AI will be used by other AI systems, says Matt Wood, PwC’s global and US commercial technology and innovation officer.
In a recent episode of The Most Interesting Thing In AI podcast, Wood, who has a PhD in machine learning, laid out a vision for how work could get done in organizations with AI. In the bottom layer, AI systems will work with other AI systems to handle operational work, like supply chain logistics. In the layer above that, humans will work with AI systems on creative work, like brainstorming and exploring problems. In the top layer, humans will collaborate with other humans through debates and meetings, but be better prepared because the prep work and data crunching has already been done by or with AI.
We spoke with Wood about this vision, and the types of work he thinks humans and AI will eventually do. Here are excerpts of that conversation, edited for length and clarity:
How should workers think about which parts of their job AI will eventually handle versus what they’ll continue doing?
One of the things that we’ll all have to get comfortable with is that we are going to have to reassess our own professional identity and reassess where we take professional pride. Because, to answer your question, AI systems are going to be not just more efficient, but better than us at very specific tasks in the same way that computers are better at specific tasks today.
This idea that an AI system is good for a first draft or helps us get started—all of that is true. But it will be a very small sliver of how we use AI going forward, because computers are just better at some things than us.
To me, it makes all the sense in the world to lean into where computers are better. There is some onus on us to be open-minded where we see the AI outperforming us. It won’t be everything.
The idea that an AI is just going to completely wipe out all economically valuable work is not going to be the case because we’re not going to want that. We’re not going to be on the WALL-E starship floating around on our beds, choosing not to work. My bet is that we will choose to work on a set of increasingly compelling problems, and the AI systems will be there to assist us in exploring those problems.
We’re not there yet. The best option at the moment is to acknowledge the shortcomings and limitations in the short term and work really hard to work around those. And assume that right around the corner is an AI that never makes a mistake and is available at effectively zero cost. Not because it’s coming tomorrow, but that’s the path that we are clearly on.
What type of work do you see falling in the bottom layer you identified, where AI systems are working with other AI systems to complete work?
It will largely be operational and transactional work. That’ll be the first layer—most of the work that is transactional, high-volume, repetitive, ongoing. Not just because they’re small tasks. There will be AI systems that just run for a couple of months. They may actually end up taking about the same amount of clock time as a human does, but the quality of their output will be better than a human. Not planning for a world where the majority of AI systems are consumed by other AI systems is betting against gravity.
What type of work do you think humans will be better at than AI?
I suspect that humans will continue to want to spend time with other humans. There is a real human need to band together to discuss, debate, and reason through problems. I think the future of work looks a lot more like humans spending time on that type of activity and a lot less time on the prep for that activity. The prep will be done one-on-one with a series of agents that help me be better prepared for those debates and those discussions.
My guess is it’s going to look like that sort of work, versus [whether] we will retain creativity or judgment.
I think that machines will eventually be more creative than humans. Machines will exhibit higher judgment relative to most humans. [But] I think it will be that we love to get together and we love to get in front of a whiteboard and just teach each other, tell stories to each other about how we think about a problem, and influence each other in relatively small or large groups. No doubt there will be a whole swath of AI systems that are listening in to support that debate in real time. I don’t think it will ever get to the point where AI is out of the loop. Increasingly, we will rely on it to support those types of activities.
Read more on jobs, AI, and how it’s changing work:
- Six expert AI and work predictions for 2026
- What we know now about AI we didn’t a year ago
- The higher-order skills AI can’t touch—yet
LinkedIn co-founder Reid Hoffman on what’s next for AI and work