What have you deleted?

Eric Porres had the dashboard every AI leader wants.

At AUTONOMOUS: OBSOLETE in May 2026, Eric described broad AI adoption at Logitech. My takeaway from his talk was that the usage dashboard could not answer the harder question: how had the work changed?

Then he sat with it honestly and asked the harder question: has anything actually changed?

The question cuts through a familiar pattern: people use new tools while the same meetings, reports and approval chains remain. Activity can rise without changing the work that matters.

That was the lesson I took from Eric’s May session. Usage tells you whether people are trying the tools. It does not, by itself, tell you whether the work is better.

So Logitech changed the question. They stopped asking "who is using AI" and started asking "what have you deleted?"

Deleted means the recurring meeting that is off the calendar, the report no one builds by hand anymore, or an approval step the team no longer needs. Eric’s question pushed us to name a concrete change, rather than point to another adoption chart.

That distinction is critical and I think it's exactly right. And it leads to the question that actually compounds: what have you created for others? Not just for yourself. For your team. Finding a repeatable workflow, infusing it with AI, and making it something the whole group runs on.

Deleted and created. That's the system. At Logitech, they built an internal measurement called Pulse to track exactly this. Think of it like heart rate variability for your organization's AI fitness. A vital sign, not a scoreboard.

The bottleneck isn't where you think

Once you start measuring deletion, the real bottleneck shows up fast. And it's not where you'd expect. It's not the models. Eric said the models have been good enough for a while. It's not your frontline employees either. They're curious. They'll try anything.

It's the middle.

The middle manager isn't blocking out of fear of the technology. They're blocking because they're still measured on the old thing. The old report. The old process. Nobody told them they're allowed to stop doing it. Eric's framing: the constraint was never model quality. It's human permission quality.

I wrote about something connected to this in The Leader's Dilemma. My argument was that your people aren't resisting AI. They're resisting ambiguity. And when you don't tell them what's allowed to change, they assume nothing is. The middle manager who keeps running the five-day approval chain isn't being stubborn. They're being rational. They're optimizing for the metric they're measured on.

The sequence almost everyone runs backwards

Eric's answer to that is a sequence that I think almost everyone runs backwards. The instinct is that more training leads to more adoption. It doesn't. Here's what actually works: change the expectation first. Tell your team the report is now AI-drafted and the human edits. That's the job now. The changed expectation forces the workflow to change. The changed workflow produces visible time savings. And then training lands, because now people actually want it.

Expectation → Workflow → Savings → Training. In that order.

The second move is that embedded beats optional. If AI is just a tool people choose to open, usage is fragile. When you build it into the workflow itself, it stops being a choice and becomes infrastructure. It becomes just how work gets done.

The control plane

Eric also emphasized the need for governance. If agents act on sensitive records or make consequential decisions, teams need clear access rules, review responsibilities and audit trails. Faster execution makes that operating design more important.

It's slower at first. Especially in a landscape where every week brings a new model drop. It will feel slow. Eric's line on this one stuck with me: obsolescence without governance is just chaos with a demo.

What actually goes obsolete

And then he reframed what actually goes obsolete. Three things, and only one is software. The per-seat SaaS model, because when an agent does the work you don't buy 50 seats anymore. The usage dashboard, because the entire vanity layer of AI metrics becomes obsolete the moment you measure deletion instead. And the manager who measures activity instead of outcomes.

But here's where Eric landed that I think matters most. What goes obsolete is the small version of the job, so a bigger one can take its place. You won't lose your job to AI. But you might lose it to someone who cleared the old work off their plate while you were still doing it by hand.

That expansion is real, and it's not free. The price is paid in deletion. And here's the hardest part. We've all learned to give permission to fail. Fail fast. That grant is easy because it's about things that don't exist yet. Permission to delete is much harder. It means letting people eliminate what already exists. The report with an owner. The process the manager is measured on.

*AUTONOMOUS: OBSOLETE · Logitech Chief AI Officer · May 2026*