DeepDive

have been running design sprints for more than a decade. And the ones that failed? They all had the same problem.

Someone with authority walked into the room with the answer already decided. The sprint became a performance. Five days of exercises, sticky notes, and prototyping — all to validate what was never supposed to be questioned.

Then I read How Big Things Get Done by Bent Flyvbjerg, and one idea changed how I see every sprint I've ever run: "Commit to not commit."

I believe this is the most important lesson for anyone running sprints today. And it starts with understanding what authority does to a room.

ChatGPT is insanely powerful.

But most people waste 90% of its potential by using it like Google.

These 1,000+ proven ChatGPT prompts fix that and help you work 10X faster.

Sign up for Superhuman AI and get:

Claim your free prompts

In 1941, General Brehon Somervell needed a new building for the U.S. War Department. A big one. Space for 40,000 people. ¹

Somervell was famous for getting things done. He got full authorization within a week. He gave his team a weekend to design it.

And then he sprinted, straight from idea to building. No questioning. No exploring alternatives. No asking whether this was the right approach.

The result was the Pentagon — an $83 million project that became known during construction as "Somervell's Folly." The plan had serious flaws that could have been caught earlier, if anyone had slowed down to look.

But nobody questioned Somervell. He had the authority, the track record, and the momentum.

Flyvbjerg's point is simple:

Instead of looking at other opinions, instead of giving time to question the plan, he went full speed from idea to execution.

I've seen this exact pattern in product teams. Not with buildings, but with Design Sprints.

Jeff Bezos built Amazon's entire culture around questioning ideas rigorously. The famous six-page memo process. The "disagree and commit" philosophy. Every decision challenged from multiple angles.

And yet, when Bezos pushed the Fire Phone, a 3D smartphone that launched in 2014, nobody stopped him.

His design team couldn't find good uses for the 3D feature. Former product leads reported heated debates internally.

But at a certain point, everyone deferred to his track record. Bezos controlled every decision about the phone's design. Even the smallest decisions needed his approval.

The result? A $170 million write-down. The phone launched at the same price as the iPhone 6 and was a complete flop.

The process existed. The culture existed. But authority overrode both.

This is the same thing that happens inside design sprints. When the person with the most authority is also the one driving the direction, other participants have a hard time saying no.

I've seen it happen. The room goes quiet. People sketch what they think leadership wants to see. The exercises still happen, but the questioning stops.

If one perspective starts to dominate, the sprint fails.
(It doesn't matter if that perspective belongs to the smartest person in the room.)

Today, the dominant voice in the room isn't always a person but the AI output that nobody questions.

A team asks ChatGPT for a product solution. The answer comes back fast, articulate, and confident. It looks well-researched. It comes with structure and logic. So it goes on the board. And nobody subjects it to the same scrutiny they would give a human idea.

AI doesn't know your company's history. It doesn't know what was tried in 2019 and failed. It doesn't know the technical constraints your engineers are navigating.

It gives you a confident answer without any of the context that actually matters.

The moment a team treats an AI output as the answer instead of one input among many, they've done exactly what Somervell's team did. They let authority (this time artificial) go unchallenged.

II once ran a design sprint for an oil and gas company. Complex technology. Big stakes.

I made sure technical leaders who had been working in the company for years They knew which approaches had been tried before. They knew which technologies couldn't support certain directions. They knew the constraints that nobody on the product side was aware of.

And during the sprint, they killed ideas that the rest of the room loved.

Those ideas would have died anyway but six months later, after real money was spent.

The technical leader has institutional memory. That's context you can't Google and AI doesn't have.

When the sprint starts with "build X," authority has already won.
When it starts with "why are our users struggling with Y," the room has permission to explore.

The person who asks "are we solving the right problem?" when everyone else is already sketching solutions. Most teams avoid this because conflict is uncomfortable. But the sprints that produce the best outcomes are the ones where at least one person is genuinely pushing back.

Even better > Bring in someone unrelated to the project, someone who questions every idea the same way regardless of who said it.

The whole point of prototyping is to learn, not to confirm. If the test says the idea is wrong, burn it. no matter who brought it to the table. "Commit to not commit" only works if you actually mean it."

Whether it's a officer building the idea of Pentagon in a weekend, a CEO pushing a 3D phone nobody wanted, or a team copying AI outputs onto a whiteboard the failure is always the same. Authority went unchecked.

The design sprint is designed to prevent this. But it only works if you protect the conditions that make questioning possible.

The sprint doesn't protect you automatically. You have to protect the sprint.

Anthropic is doubling Claude's usage. Click to learn more. Source: Anthropic.

Anthropic is doubling usage limits over the next two weeks across Free, Pro, Max, and Team plans outside of peak hours. Separately, the full 1M-token context window is now generally available for Claude Opus 4.6 and Sonnet 4.6, with no increase in price.

Why this matters for teams: A 1M-token context window means you can upload entire design systems, research reports, or sprint documentation into a single conversation. That's a real shift for designers who use AI during discovery more context in, better output out.

But remember what we talked about today: more context doesn't mean less questioning. Treat the output as one voice in the room, not the final answer.