In their new book Superagency, LinkedIn co-founder Reid Hoffman and his coauthor Greg Beato argue for an ambitious and optimistic approach to AI, on the grounds that it has the potential to increase human agency, “the capacity to make your own choices, act independently, and thus exert influence over your life.”
“In some instances, you might want to work closely with an AI—such as when you’re learning a new language or practicing mindfulness skills,” they write. “In others, such as optimizing your home’s energy consumption based on real-time energy prices and weather forecasts, you might prefer to let an AI handle that by itself. Either way, the AI is increasing your agency, because it’s helping you take actions designed to lead to outcomes you desire.”
We spoke with Hoffman—who is also a co-founder of AI startups Manas AI and Inflection AI, a Microsoft board member, a venture capitalist at Greylock, and a former OpenAI board member—about how his vision for AI could apply to a workplace context. Here are excerpts from our conversation, edited for space and clarity:
What are examples of how AI could increase the agency of workers?
We already see some of it today. Writers, marketers, report writers already when they start using GPT-4 and other tools are finding that it’s an amazing accelerant, that they help solve the blank-page problem, etc. We already have it integrated into various forms of email communication for your everyday person. Then one of the particular lenses that we’re seeing—through products like Microsoft Copilot and others—is the amplification of engineering—people who code—which obviously is a lens into the future for any other kind of detailed professional work, whether it’s legal, medical research, etc.
By the way, the coding stuff is going to be important, not just because it’s a deeply trained professional use case, which will then echo into other deeply trained professional use cases. But also, in a sense, one part of how the AI tools will be a form of agency amplification is the most prevalent coding language will be English and we will all have the ability to add code to what we’re doing. Not just the people who are deeply expert in it.
How do you see the spread of AI agents—which can perform tasks without the involvement of humans—impacting work and workers over the next roughly 18 months?
Part of what ChatGPT showed was that a form that people were familiar with—namely the equivalent of a chatbot because we’re familiar with this from messaging—suddenly caused the thing to explode in terms of accessibility, usability, and utility. We’re going to see a similar thing with agents. Because there will be everything from your own personal agent—such as Inflection and Pi, personal intelligence—to multiple workplace agents. There will be agents for helping assemble knowledge across meetings. There will be agents doing research. There will be agents of the knowledge base for groups in the company.
That agentic format is the parallel for what we’re going to see with ChatGPT because people then go, ‘I know how to add that into my workflow.’ For people who are using Slack or its equivalent, ‘I know how to use that effectively. I know how to bring that in and have that be something that can help me become a lot faster, more on target, more effective, bringing in domains of knowledge and capability.’ So over the next 18 months—and every quarter from now through the entirely visible future—agents will proliferate. An end state will be that a professional won’t deploy except with a set of agents. It’s almost like you say, ‘I’m a professional, but I don’t work with a phone or a computer.’ Of course, every professional works with a phone and computer. It’s the same working with a set of agents that will be part of how I produce, how I analyze, how I collaborate.
You said ‘produce, analyze, collaborate’—is that how we should think of the utility they provide in a work context?
What the pantheon of agents will be is an interesting thing to predict. But there will be agents for people—like your inbox equivalent—also both outside and inside a company. There will also be agents for functions and teams. So there will be more agents, more agent functions than there are people.
For a corporation of a thousand people, I wouldn’t be surprised if there were 2,000 or 3,000 agent functions in that mature environment, or even possibly more. You say there’s one for each individual enabling them, but then there’s also team agents and function agents. There’s inside for us to work and then there’s outside engaging for our company. You’ve already mentioned customer service as one example. That’s an obviously early one that will be growing at a very fast pace. But there’s the notion of, ‘Hey, you just want to talk to my company. We have an agent!’
Going back to individual agency, what are the best mechanisms for workers to have a voice in how AI is deployed in their organizations?
There are a couple. One of the really central things—and this is part of the reason why I’m publishing Superagency—is to get people to go grab the agency steering wheel and get active. Start using the AI tools and start getting a sense of what that is. Versus treating it as like trying to imagine what the issues would be with flying in an airplane when you’ve never gotten in one. Getting some experience is very important. Within that, part of the thing that you want to do internal to a company is discuss what are the ways that we feel most enabled and empowered by these agents.
There will also be a natural, ‘We could use them to monitor your productivity and see are you working as hard as you could be, or working the number of hours that we expect you to do.’ Obviously that kind of thing will probably create in many corporate cultures, bad corporate cultures. So you want to avoid that.
Part of the phrase we use in Superagency is an AI that’s for you and by you, versus on you. As opposed to saying ‘I don’t want this AI thing because I don’t want to be learning it,’ it’s like, ‘I want to be using it to help me and help the company and here’s the things that I want it to do and not do as we learn and iterate to do that.’ We should have good, honest conversations between management and employees about how it is we’re collaborating together here.
Purchase Superagency from Bookshop or Amazon.
Read a full transcript of our conversation, including more about how AI changes what it means to be a successful manager, what the most important competitive differentiators will be between companies, what AI means for worker collaboration and job loss, and how AI impacts the way we assess individual worker performance.
Read a report about AI in the workplace released by McKinsey to coincide with the release of Superagency.