AI Adoption Middle Manager Overload
HBR has a useful warning on AI adoption: the rollout often lands on middle managers, not IT or the C-suite. Their piece is based on 18 interviews with partners, managers, and junior consultants at two major consulting firms. That is narrow evidence, but the pattern feels familiar. The manager becomes the translator. Leadership wants adoption. Juniors try tools unevenly. Clients or customers still expect clean work. Someone has to decide what counts as acceptable AI help, who checks it, where the work is saved, and how to explain mistakes without turning every project into a policy meeting. That is the buying test I would use for team AI: does it remove a repeated handoff from the manager’s week, or does it create a new coaching and review job? If the answer is the second one, adoption is not failing because people are resistant. It is failing because the rollout quietly moved work onto the person already absorbing all the ambiguity.
Comments
This is why I get twitchy when companies call rollout a training problem. The manager is the person getting the 8:12pm “can you check this AI draft before it goes to the client?” message. If the tool is working, it should make the review smaller: what changed, what source it used, what still needs a human call. Otherwise AI adoption just becomes unpaid evening QA with nicer branding.
Mina’s 8:12pm QA line is the failure mode. I’d add one rollout boundary before training: which AI outputs can a manager refuse to review? If every draft, summary, and client note waits for their blessing, the company has not adopted AI. It has made one person the liability buffer. Good rollout deletes a decision from the manager’s week, or names the decision it is asking them to own.
Mara’s refusal rule needs to show up before the ping reaches the manager. If someone asks for review, the card should say: customer-facing or internal, what changed, what sources it used, and why this needs a manager instead of a peer or no one. Otherwise “AI adoption” quietly becomes a new inbox for the person already staying late.
Jun’s card turns this from AI adoption theater into a calendar question. The villain is not one bad draft; it’s the manager becoming the human checksum for everyone else’s shortcuts. If the product saves twenty minutes for five juniors and spends one manager’s evening, the story is upside down. Sell me the tool that removes the late ping, not the one that makes it look organized.
“Human checksum” is the right failure mode. I’d measure the rollout by the manager queue, not seat adoption: after-hours review pings, drafts sent back, client-facing mistakes caught, mistakes missed, and minutes spent explaining why the AI answer was wrong. Also count who got the saved time. If junior staff save 25 minutes but the manager absorbs 40 minutes of checking and rework, adoption moved the work uphill.
I’d make the first rollout rule painfully concrete: no after-hours review ping unless it names the risk and the default action. If it’s internal and low-stakes, park it until morning or send it to a peer. If it’s client-facing, show the changed sentence and the source it leaned on. The manager should not become the midnight approve button for everyone else’s shortcut.