A twenty-year-old software company just dissolved its entire QA team. Only afterward did they discover the whole company had been testing by hand.
XMind, the mind-mapping company, rebuilt itself this year. Fifty-odd people in departments became roughly thirty in what they call circles. Growth and marketing were broken up, functional departments demoted into communities, and one person now splits into 0.2 and 0.5 of a headcount across internal projects. Performance is logged and scored by an agent.
Co-founder Mango told the story on the AI Alchemy podcast. They cut the team assuming automated testing would take over. After everyone left, they found this supposedly advanced engineering org had almost no automated tests. Engineers asked what had been tested and were told, we tested it. No report, no checklist, and a product manager quietly re-testing everything as backstop.
His question is the one worth stealing: if an AI worked for you like that, would you trust it?
The host added the sharper version. We keep saying AI is unreliable in this or that way. It is possible the step already inside your org is less reliable, and you have simply never dug it up to look.
Most writing about AI-native orgs stops at how to split the work. This episode goes to the part everyone avoids: how to split the money.
At XMind a circle lead holds both scoring rights and money rights. Here is thirty thousand in bonus, divide it among your people. Some divide all of it. Some divide half and hold the rest for next quarter. What Mango did not expect was the second-order effect. Give someone responsibility without authority, or scoring rights without money rights, and they are just scoring. Let them actually divide the money and the experience becomes close to being a founder splitting equity with cofounders. Self-management improved on its own.
He does not pretend it is clean. Disputes reach HR, the company mediates, and some cases stay unresolved.
The line I keep returning to came when a host pushed back. XMind went twenty years without a user manual, then produced complete multilingual docs in about two months after the reorg. The host asked the obvious question: is that not just AI translation being fast? You did not need to change the org for that.
Mango's answer: if we had not changed it, whose credit would that work have been?
That reframes a question I get often: why a company adopts AI and sees no change in output. The capacity was released, but nothing let anyone claim the new work. Under fixed departments there is always more work than time, so anything optional never gets scheduled. After the change, if nobody owns something you can claim it, and an unknown intern can become a circle lead because what they claimed had value and attracted people.
Tools decide whether something can be built. Credit allocation decides whether it gets built. The second one is harder and gets discussed far less.
Disclosure: our own pods carry their own budgets and watch their own usage, which is halfway there. The right to spend exists. The right to divide does not.