Last month, a friend told me that by the end of the year, his startup will have seven employees and over 20 AI agents. He has led several previous successful organizations and has a reputation as a wonderful people manager. Yet he felt unmoored imagining how to maintain a tightly cohesive team in an organization where AI will have a large presence.
He’s not alone in what he’s feeling. When we delegate increasingly significant tasks to AI agents, or welcome their presence and input in team channels, we allow them (knowingly or not) to profoundly influence our decision-making and relationships at work. We need a way to steer that wisely.
The good news is that we can proactively manage AI as a powerful collaborator without giving it the status (or indeed, the dignity) of a human teammate. In fact, getting productive outcomes from AI at work depends on finding this new way forward: one in which the presence of technology is intentionally harnessed to unlock the potential of our human colleagues.
Leading ethical and humane use of AI at Salesforce has given me insight into thousands of companies’ adoption efforts. Research for my book Manage the Machine helped me make sense of what I learned. Here are four early principles for navigating the challenge.
Assign human accountability
You wouldn’t open a new store and not put a manager in charge of it, nor would you hire a new employee and give them no one to report to. And yet, many companies have simply given workers AI tools capable of executing highly consequential tasks without being clear who answers for results.
The starting point is simple: individuals should remain accountable for their work output, with or without AI. But someone also has to own the tools themselves. Here, the management model I’m seeing emerge is a matrixed one. If technology was previously seen primarily as an IT problem, today’s companies are realizing that accountability should also live closer to where work gets done.
At Asana, for example, the IT and legal teams set company-wide standards for AI and provide a set of tools that departments can customize. But then, for example, Asana’s marketing organization has its own group that designs AI agents that serve the needs of marketing. Marketers who understand the day-to-day challenges of their work are better positioned to design both human and AI workflows to achieve desired outcomes. They can also take a holistic approach to problem solving if things go wrong, looking at issues from both a technological and organizational perspective.
Teach delegation, and let AI agents earn their autonomy
Aarthi Belani, a partner at law firm Baker McKenzie, told me how she thinks about onboarding new AI tools. She maps out where a tool is reliable and where it isn’t, limits its library to authoritative sources, determines how much context each task requires, and widens that scope only as the tool proves itself.
This may seem obvious, but it is the same discipline that makes any delegation work. You have to know what your delegate can and can’t do, and whether you want it to escalate or push through uncertainty. In a world where autonomous agents can transact large amounts of money or represent your brand to key customers, you also want every person working with them to know where that delegation should end.
Fascinatingly, developing these skills with AI can sometimes make people more effective in working with human teams as well. Aaron Kemmer, the CEO of Magic, a service that provides customers with hybrid virtual assistants (both humans and AI), observes that when customers learn how to delegate well to AI assistants, they actually ask for more help from humans, too. This is because customers have gotten more comfortable with the general skill of how to hand work over productively.
Scope AI’s influence in decision making
Steering AI effectively isn’t just about deciding what tasks to delegate (or not). It’s also about designing how you want AI to influence your team’s decisions.
We’ve known for a while about “automation bias”– the tendency for people to over-defer to AI recommendations, sometimes setting aside their own critical thinking. We also know that AI learns from historical data, so when people over-defer to it, they risk locking in fixed ideas that may no longer serve them. But when harnessed thoughtfully, AI can also help us challenge fixed ideas and improve our decision making.
Part of the answer may be in having AI ask us more questions rather than only provide answers. Researchers at MIT and Columbia report that people are better at spotting flawed reasoning when an AI interface is designed in this more interrogative style.
Some leaders take this a step further, setting up AI to play devil’s advocate, helping to represent perspectives that haven’t been considered in the room. For instance, a prominent venture capitalist described how her firm uses AI in investment committee meetings. The agent draws from historical diligence data and performance records of portfolio companies. But it is also in the room listening to deliberations about new potential investments. From time to time, the agent will notice perceived inconsistencies in the discussions and bring them up as questions, including where patterns of bias may be at play in discussing a founding team.
Sometimes, this VC shared, it can be hard in the moment for her to challenge a colleague directly. But when AI brings up the question, it allows the team to have a more holistic conversation.
Reinforce, don’t replace, human connection
We know that belonging and clear communication are hallmarks of high-performing teams. But in a recent Workday study of more than 2,000 employees at large companies, the top concern about AI was reduced connection to colleagues, ahead of potential job loss. One in five Gen Z employees in the same study reported feeling lonelier at work since AI tools were introduced. Other studies indicate that AI can unintentionally disrupt how human colleagues interact and coordinate with each other.
Part of the answer is to thoughtfully engineer nudges and interactions that preserve and reinforce the bonds that matter at work. Think: Slack search results that not only show an employee the answer to the question they asked about a particular project and its timelines, but also the right person or team to contact if they want to learn more.
Other times it’s about placing limits on places where human connection is essential. AI management coaches, for example, can be surprisingly effective (as I learned myself) on helping managers rehearse hard conversations or reminding them to check in or provide kudos to team members. But they shouldn’t serve as a substitute for those crucial interactions. If anything, as AI becomes more pervasive, companies will need to invest more in human-only time – onsites, gatherings, social activities that preserve the cohesion that makes work tick.
As AI becomes more pervasive in knowledge work, my friend’s worry about keeping his team aligned and energized is likely to become more common. The answer lies in a deliberate approach to how we manage AI. We should be the ones deciding how it shapes our decisions, our work product, and our relationships, rather than the other way around.
When we make these choices deliberately, AI can do what the best technology has always done: extend what people can accomplish, together.
