You don’t need more AI agents, you need better decisions

Built all the agents and still waiting for the payoff?

Companies I am talking to at the moment seem to have very similar challenges. They have hundreds of AI agents, and individual teams are automating parts of their job to make things faster, but most companies are still seeing very little ROI from their AI investment.

This is going to be a huge focus for 2027, and this isn’t just what I’m seeing. McKinsey’s 2026 State of AI research found that 80% of respondents saw individual productivity improve, but only 37% saw a positive impact on their organizations operating earnings.

People are certainly getting faster, but the business isn’t getting better, and I would put this gap down to how companies are approaching how they’re redesigning work.

What does redesigning work actually mean?

Most companies are jumping to task, process or workflow automation, with the overall focus being on cost, speed, and efficiency, and this is where I think we’re getting it wrong. When I talk about redesigning work, I mean something quite complex and unglamorous. It starts with who does what in the team, who makes the decisions, who checks whose work, which teams need to collaborate and how does work get handed from one team to the next.

If a leader can’t tell me what each person will be spending their time on after the change, they’re not ready to automate anything yet. The trouble is, most AI conversations are skipping over this question, and treat it as a separate workstream, something for communication, training and change management to think about once the redesign is done.

People are not separate from the work, and if they don’t trust the process you’ve created, then it will not increase adoption, or the ROI companies are looking for. In fact, you’ll just end up with a process that’s more efficient on paper than in practice, and will likely need to keep changing.

Work is also never just a series of tasks. Think about how long it takes to get something through your organization – very little of it is doing the actual work. Most is waiting for sign-off, someone to make a call, or another team to do their bit.

So when you automate tasks, you’re speeding up the part that was already pretty quick, and the waiting stays exactly the same as it was. That’s the gap the McKinsey numbers are pointing to. The tasks get faster – yes, but the process doesn’t because no one took the time to look at how work moves between people, or who makes the final decisions. You can automate ten steps in a process, and it’ll still take three weeks if it sits in someone’s inbox for two of them.

This is a leadership responsibility that needs to get figured out first. How much of your team’s time goes on doing the work, and how much goes on waiting for someone? Whose decisions need to change to shorten that wait?

If we truly want people to adopt AI, we need to bring people into the equation much earlier.

Don’t get me wrong, I understand why companies are splitting AI transformation into different stages. Tech teams work out what’s possible, operations find the efficiencies and then HR or change teams support the people affected, and each of those groups is incredibly useful.

However, in most strategies I’ve seen, by the time anyone asks how people will adapt, the organization has already decided where they’ll spend their time, what they’ll be accountable for and how they’ll work with their colleagues or customers.

I can’t stress enough; every decision about a workflow is a decision about people. If you separate the two, lots of time and value gets lost, and that’s why I’d argue that the people who know how the work actually happens should be in the room before the design is agreed, not after.

So what should we be training managers for?

There’s a big focus right now on building AI fluency and capability, which is important, but it’s not the capability most leaders are missing. Knowing how to use a tool or spot something to automate doesn’t mean you know how to redesign the work well or encourage adoption. If we truly want people to adopt AI then we need to start them.

We need to help managers build capability in redesigning roles, including knowing that just because something can be automated doesn’t mean it should be. That means looking at the whole picture, balancing what the business needs, (speed, cost and efficiency), with what people need from their work, like the parts of the job they’re genuinely good at and enjoy. If you automate the part of someone’s job they enjoy most, they’ll care less about the rest of it, and that ends up costing you more than the time you saved.

Gartner believes more than 40% of AI agent projects will be scrapped by the end of 2027, because they cost too much, nobody can show what they’re worth, or the risks weren’t thought through. It’s only a prediction, but it matches what I’m already seeing.

You can have hundreds of agents and save thousands of hours, and the business can still be doing no better than before. If your team saves five hours a week but nobody’s decided what to do with that time, nothing has really changed. If an agent speeds up one step but the next step still needs three people to sign it off, the work ends up waiting there instead.

More training on AI tools won’t fix this. What fixes it is leaders getting better at deciding how the work should be set up, who does what, and what the business actually needs from it.

Where this leaves us going into 2027

The companies that get the most out of AI next year won’t be the ones with the most agents; they’ll be the ones that think about the people and the work together, and whose leaders know how to make good decisions about both.

That’s what I’m doing with companies right now. I’m working with leadership teams before and during their AI changes, helping them think about the people and the work together, redesign roles in a way that works for the business and their teams, and lead people whose jobs look very different from before.

If you’re seeing this gap in your teams, let’s talk. Send me a message and we’ll set up 30 minutes to look at where things are getting stuck.