Managers used to get away with managing work through status.

Not always well, but often enough. Who has the task? Is it done? What is blocked? When will it ship? Do we need to escalate?

AI makes that style weaker. More work is produced faster. More output looks finished before it is trustworthy. More exceptions appear at the edges. More context sits inside tools, prompts, generated drafts, and review comments. A manager who only tracks status will see motion without understanding the system.

The manager's job moves toward work-system design.

That does not mean every manager becomes a process architect. It means managers must design how work flows when humans and AI share the job.

They need to decide where AI assists, where it reviews, where humans handle exceptions, and where automation is safe. They need to define quality bars. They need to allocate review capacity. They need to make hidden work visible. They need to adjust roles. They need to notice when output volume is rising but decision quality is not.

This is practical management, not futurism.

A manager redesigning a support workflow should know which cases are safe for AI-drafted replies, which need full human review, which should route to specialists, and which should never be AI-assisted without approval. They should know how corrections improve the macros, prompts, and knowledge base. They should know whether handle time improved at the cost of reopen rate or customer trust.

A manager redesigning a sales workflow should know whether AI research improves account judgment or merely increases outreach volume. They should know which CRM fields can be automated, which deal risks require rep judgment, and which manager reviews are still worth the time.

A manager redesigning a finance workflow should know which variance explanations can be drafted, which calculations need locked sources, which anomalies trigger escalation, and how review comments improve next month's packet.

The manager becomes responsible for the operating conditions of good work.

That includes the input quality. Bad inputs create bad AI workflows. If tickets are messy, account data is stale, policies conflict, or source-of-truth ownership is unclear, AI will magnify the problem.

It includes the mode choices. Managers should not let every person invent their own assist/review/exception/automate pattern. Local improvisation is useful early. At scale, the team needs a shared design.

It includes review capacity. If senior people are overloaded, the manager must redesign the review system rather than praising them for heroics.

It includes metrics. Old productivity metrics can become dangerous. If AI reduces drafting time, measuring only volume may push people to create more low-quality work. Managers need metrics that include quality, rework, cycle time, exception rate, review load, and customer impact.

It includes learning loops. Every correction should have a place to go. Otherwise the workflow depends on individual memory and repeated cleanup.

The trap is treating AI adoption as enablement. Train the team, share prompts, encourage usage, celebrate wins. That helps, but it does not replace management design.

A useful manager routine might look like this:

  • review one AI-assisted workflow each week
  • inspect where work waits or gets reviewed
  • sample outputs against the quality bar
  • look at exception volume
  • ask what hidden work people are doing
  • update one prompt, playbook, source, or routing rule
  • remove one unnecessary handoff or review step when trust is earned

Small loops beat grand transformation decks.

This shift also changes what good managers need to be good at. They need enough technical literacy to understand system behavior, enough operational taste to simplify workflows, enough judgment to set risk thresholds, and enough coaching skill to help people move up the value curve.

The manager does not need to be the best prompt writer. That is too small. The manager needs to understand the work system well enough to make AI useful without making the organization more fragile.

In the AI era, weak management will hide behind tool usage for a while. Strong management will show up in cleaner flows, sharper roles, better review, fewer exceptions, and higher-quality decisions.

The difference will be obvious in the work.

This is part 9 of 10 in Work Design for the AI Era.