I taught the empowered team gospel. I installed it. At three different companies I stood in front of a product org and made the case for autonomous cross-functional teams owning outcomes, and I believed it. I still think it was right, for the world it was built for. That world had near-zero interest rates and is gone.
The model was an economic feature, not a law
Here is the thing the empowered team literature almost never says out loud. The model is expensive. You take a group of senior people, give them broad autonomy, tell them not to rush to solutions, fund a discovery cadence, and accept that they will spend weeks learning their way to the right outcome before they commit to a build. That is a beautiful way to work. It is also a luxury good, and it assumes your company can afford the patience.
That patience was affordable in the zero interest rate era because capital was cheap and the market rewarded growth over efficiency. You could carry a large product org running deliberate discovery because money was nearly free and nobody was asking hard questions about cost per outcome. The empowered product team was not a timeless truth. It was a feature of a specific cost of capital, the way free snacks and ten-person platform teams were features of the same era. When the rate environment flipped, the tolerance for slow, autonomous discovery shrank with it.
AI removed the constraint the model existed for
The cost of capital is half the story. The other half is worse for the orthodoxy. Walk back to why empowered teams existed in the first place. Engineering capacity was scarce and expensive. The most important decision in product was how to spend that capacity, and getting it wrong was a six-month, seven-figure mistake. The cross-functional trio, the dual-track discovery, the opportunity solution trees, all of it was a sophisticated system for rationing a scarce resource.
AI collapsed the resource. When an agent produces a working prototype in an afternoon, the bet is no longer six months and seven figures. It is an afternoon. You do not need three weeks of discovery to de-risk a build you can just do and test the same day. So the model got hit from both sides at once. Capital got expensive, which killed the patience it required. AI got cheap, which killed the constraint it was built to manage. A model can survive losing one load-bearing assumption. This one lost both in the same eighteen months.
What is actually breaking
Be precise, because the lazy version says empowered teams are dead, and that is not what is happening. The autonomy is not breaking. Teams should still own outcomes. What is breaking is the size, the ceremony, and the assumption that you need a full cross-functional squad running a discovery cadence to make good product decisions. That apparatus was the expensive part, and it was expensive specifically to manage a scarce build resource that is no longer scarce.
I will own my part. When I installed empowered teams, I taught them as the right way to do product, full stop, not as the right way given cheap capital and expensive build. I stripped the conditions off the conclusion and handed people the conclusion as a law. That is the error the whole canon made. In my current work at Falkster.AI, the empowered trio is a single builder plus a fleet of agents: an outcome comes in, a listening agent extracts it, agents dispatch to build a prototype the same day, and the builder judges the result against the real outcome.
Look at your team's discovery cadence and ask one question. Was this designed to manage scarce, expensive build, or to learn about the customer? If it is the first, try killing one ceremony this sprint and replacing it with a same-day prototype tested against a real customer outcome.