AI made you faster. Your team still isn't moving faster. Here's why.

AI is making individuals faster than ever. It's not making teams faster in the same proportion. That gap is one of the biggest problems I see in product organizations today, and it comes down to two things: AI amplifies whatever misalignment already exists on your team, and creation has scaled a lot faster than decision-making. Here's how I think about fixing both, and where I still think a person, not the model, has to make the call.

The real bottleneck isn't creation, it's decisions

We commissioned Forrester last year to talk to 518 product, engineering, and design leaders across the industry, and the data backed up something most of us already feel: 75% of these leaders said most AI tools focus on individual productivity, while team productivity stalls or even slows down as individual output goes up.

At the pace work moves now, just getting everybody up to speed around the same source of truth is getting harder, not easier, and more meetings won't fix that. Small differences in context can turn into real misalignment fast, and AI accelerates that just as much as it accelerates the good stuff.

At the same time, creation has scaled with AI. Decision-making hasn't kept up. The PMs, engineering managers, and senior folks who are supposed to make the call are turning into organizational bottlenecks, and that correlates directly with burnout.

You can try to fix this with better prioritization habits, but at some point that just creates more alignment tasks, more status updates, and more meetings, which actually slows the team down further. The goal should be the opposite: with as few meetings and as little coordination overhead as possible, get to decision-ready work that everyone can see, challenge, and build on from the same shared source of truth. What I actually care about is making sure the whole team moves faster in the same direction, not just making one person faster.

What I mean by an agentic canvas

I think about this in three simple pieces. Connectors bring context from the tools your team already uses, like Slack, GitHub, Amplitude, or Granola, into the canvas, so AI doesn't have to work from an isolated prompt. Miro Sidekicks are AI teammates you work with conversationally: think out loud with one, ask it to make sense of a pile of information, and have it create docs, diagrams, prototypes, or entire boards the team can then work with. Miro Flows are repeatable AI workflows that live directly on the canvas, so instead of a one-off interaction with AI, you build something the whole team can see, run, reuse, and evolve together.

What actually matters here is where that work lands, not which model is doing it. If I have something sitting in my own AI chat session, nobody else can really use the context I've built up there. On a shared canvas, everybody can lean in: see the inputs, reuse the outputs, copy something, rerun it, change it, or build on top of it. That sounds like a small distinction, but it's fundamental. You stop using AI purely as an individual productivity tool and start using it collaboratively with your peers.

The two flows I keep coming back to

Watch it live: I originally walked through the two flows mentioned below in a Product School builder week session. 

I run two Flows on repeat that show what this looks like in practice: one automates a big chunk of the product development lifecycle, or PDLC, from messy research and feedback all the way to a prioritized backlog and new prototype variants, and the other turns scattered cross-functional updates into a single, decision-ready view.

For the first one, I deliberately use generic inputs, things like NPS data, PDF research, and a prototype screenshot, because they're the kind of thing almost any product manager already has lying around. The specific format isn't the point. The point is that you take a messy collection of inputs, hand it to AI, and ask it to make sense of it or turn it into the next useful artifact.

The most common mistake I see people make when they first try to build a Flow like this is overcomplicating it right away. In almost every case, you're just trying to turn one thing into another: signals into synthesis, research into themes, themes into concepts, a concept into a prototype. Play with that one transformation before you try to design some elaborate end-to-end pipeline.

The second Flow exists because cross-functional alignment is where teams lose the most time. Different functions track timelines in different tools, updates land faster than anyone can read them, and somebody always ends up manually piecing it all together before a status meeting even starts. AI can take that scattered context and turn it into something the whole team can look at together, which is genuinely valuable because it cuts out so much manual synthesis work.

What changes in both cases is that the work stops being trapped inside somebody's individual machine or AI session, not just that one person moves faster. Engineering managers, ICs, designers, and PMs can all see the same source material and the same outputs, lean in, challenge something, or build directly on top of it, instead of waiting for someone to export a CSV and pass it around.

Where human judgment still has to do the work

This is the part I want to be direct about: you should not blindly trust the output, on either of these flows.

If AI is synthesizing research, pulling information from different sources, identifying themes, or proposing strategies and prototypes, somebody still has to sanity-check it. Was the source material actually complete? Do the conclusions hold up? Does the result reflect the real nuance of the problem? I would never advocate taking an AI-generated conclusion at face value, especially when it's informing an important product decision. You wouldn't send your boss a block of AI-generated text without reading it first, and you shouldn't make a real product decision without reviewing the reasoning behind it either.

The same caution applies to the cross-functional side, maybe even more so. I'd be careful about ever calling an AI-generated synthesis a perfect source of truth. If it's pulling information from multiple places, you can't automatically assume every detail is correct, complete, or interpreted the right way. The goal here is a much better shared context that gets the team decision-ready faster, not an unquestionable verdict, so humans can spend their time validating the facts and making the actual call instead of spending the first twenty minutes of every meeting just catching each other up.

The point of AI here isn't to remove human judgment. It's to dramatically compress the work it takes to get to the point where good human judgment can actually be applied.

Where this goes next

One thing becoming very clear to us is how much AI customization is going to matter. Generic AI can already do a lot, but teams and organizations work in very specific ways, with their own processes, terminology, standards, and expectations. Giving people more ability to customize how AI works for them, through things like skills and custom skills, is only going to matter more.

Longer term, I expect the relationship teams have with AI to shift from everybody individually prompting a generic assistant toward teams building shared AI capabilities around the way they actually work. The interesting question stops being "how do I personally use AI to do this task faster" and becomes "how does our team encode the best way we do this work, so AI can help everyone execute against it." That's where multiplayer AI gets genuinely powerful: not just shared access to a model, but shared context, shared workflows, and increasingly shared, customized capabilities that become part of how the team operates.

Start with the process everyone dreads

If you want to try this yourself, don't start by inventing some sophisticated AI-native process from scratch. Start with the most pressing manual process your team already runs regularly, the thing you repeatedly spend time on in a sync meeting: collecting updates, synthesizing research, turning feedback into themes, moving information from one format into another. If the manual effort is obvious to everyone, that's the best place to see whether AI workflow automation can compress it.

Our AI Playbooks are a good place to find concrete starting points if you want examples to adapt directly. Pick the update or synthesis your team already does every week, and try building your first Flow with it on a free Miro board.