Your Company Is Carrying Chaos Debt
Unseen Adoption: Employees are building AI workflows independently, often creating valuable solutions before leaders establish ownership or oversight.
Compounding Costs: Chaos debt grows through undocumented dependencies, while talent debt reflects unrealized gains from processes nobody redesigns.
Useful Evidence: Unapproved AI tools reveal where official systems fail, giving leaders practical redesign priorities instead of reasons for shutdown.
Visible Governance: Zapier combined AI usage data, private recommendations, and flexible spending limits to encourage learning without compliance theater.
Start Auditing: Leaders should inventory builders, owners, dependencies, and data access while embedding visibility into every AI workflow.
There's a question worth putting to your executive team, and it sounds simpler than it is. What has been built with AI inside the company over the past six months? Not purchased, not licensed. Built.
The answers tend to be partial, because the building has been happening in places that don't report on it. Someone in finance put together a reconciliation workflow, a recruiter set up a project that drafts job descriptions, two people in customer support wired something together that handles first-pass responses, and it worked well enough that another team started routing tickets through it.
There’s a good chance that none of that was on your initial adoption roadmap.
Create a Free Account to Keep Reading—and Keep Leading Smarter
Unlock this piece and join a community of forward-thinking leaders discovering tools, playbooks, and insights for thriving in the age of AI.
Have an account? Log In
Analiese Brown, chief people officer at Campminder, has a name for what piles up in that situation. She calls it chaos debt, and the way she describes it forming is likely something you’ve seen before..
When everyone in an organization accesses AI simultaneously, and no operating model exists for its use, what gets built, or who accounts for the outputs, a proliferation of tools, workflows, and decisions emerges that nobody fully sees. It can look and feel like productivity. But underneath, it is deeply fragmented.
That first part is what makes this expensive. Adoption numbers go up, people report saving time, and the perception in the halls is that the transformation is working. In a narrow sense it is working, which is exactly why managers and senior staff allow things to carry on for a while.
What You Owe
Debt is a fair word for it, because the balance grows on its own. The reason it compounds is that people build on top of whatever is already working. One person's workflow becomes the data source for someone else's report, and that report becomes the thing a team makes decisions with. There was never an intention to create a dependency.
By the time anyone tries to map it, the dependencies run through work that has never been documented and isn't owned by anyone in particular. The person who built the original thing may have changed teams, and the logic that makes it run lives in their notes.
Bradford Wilkins, vice president of people and organization at Cognite, points to a cost that moves in the opposite direction. He describes talent debt, a term he attributes to Boston Consulting Group (BCG) and EY, as the unrealized potential sitting inside workflows that nobody has gotten around to redesigning.
So there are two liabilities here. One builds up from moving fast without ownership, and the other builds up from not moving at all.
The Instinct That Makes It Worse
When leaders find the sprawl, the reflex is usually to shut it down. Approved tools only, requests routed through IT, a policy memo to follow.
The trouble with that reflex, as I see it, is that it throws away the most useful information the company has produced. Everything people build without asking is a record of where the official tools let them down.
The reconciliation workflow exists because the finance system doesn't reconcile, and the support tool exists because the help desk software can't draft. Read as requirements rather than as violations, those tools amount to a map of what needs redesigning, drawn by the people closest to the work.
If you suppress the building, you lose the map, and what you're left with is the same fragmentation somewhere harder to see.
Brown's argument against treating governance as bureaucracy lands right here.
In an AI-augmented organization, guardrails are what make speed sustainable.
That's useful, because it stops framing governance as the tax you pay for moving quickly and starts treating it as the thing that lets you keep going.
What It Takes to See the Whole Picture
Zapier got to a point where individual usage was strong and nobody could see the whole picture. Brandon Sammut, the company's chief people officer, walked us through what they built in response.
The team pulled AI usage data from everywhere people were working with it, including Zapier's own product along with Anthropic's and OpenAI's, into a single data lake, and then automated reporting on top of it. The design choices they made are more interesting than the infrastructure.
Reports go out privately by direct message at the start of each month, and there are no leaderboards or rankings against colleagues, which Sammut is emphatic about.
The report doesn't just describe what happened last month, it makes recommendations. If someone in a benefits role has been running routine tasks on the most expensive model available, the report says so and suggests something cheaper that will handle the job. It keeps track of what it recommended and adjusts the following month.
There is one cap, and it's sized to catch runaway spend from a misconfigured agent or a stolen key rather than to limit anyone. Getting it lifted takes a message in a Slack channel.
What Zapier turned down is as instructive as what they built. Usage leaderboards were on the table, and so were hard caps, and both would have produced compliance behavior instead of information.
The Audit
The practical version of Brown's problem comes down to a handful of questions that are harder to answer than they look.
- What has been built in the last six months, and by whom?
- Who owns each of those workflows once the person who made it changes roles?
- Which of them are sitting inside a process the business now depends on?
- What data can these workflows reach?
Brown's own approach suggests a place to start. The career development tool that her director of people built shows the team the questions employees are asking it, so they can see what's happening and step in when something needs a person rather than a response.
The visibility was designed into the tool rather than added after something went wrong, and that's the difference between an inventory you take once and a system that stays legible.
Who Took On The Debt?
Campminder's governance model is still being shaped and socialized, which Brown describes as deliberate.
“You don't bolt a policy onto a culture and call it done, because people need a hand in building it to feel any ownership over it,” she said.
That patience gets easier to justify once you look at who ran the balance up. Chaos debt didn't accumulate because people were going around the rules. It accumulated because people were trying to do their jobs well in an environment where the tools showed up faster than an agreement about what to do with them.