How Should Managers Use AI In 2026? Three Rules That Are Now Backed By Data

Managers in 2026 should use AI to absorb routine work, free time for the parts of management AI cannot do, and reinvest the saved capacity in coaching and judgment work. The mistake most managers are making is using AI to do more of the same management work faster, instead of using it to do different management work entirely. The data behind this answer comes from Deloitte's 2026 Global Technology Leadership Study, Gallup's State of the Global Workplace 2026, and the Q1 2026 layoff patterns at Freshworks, Atlassian, and the broader software cohort.

That is the snippet-length version. The full answer requires unpacking what each of the three rules looks like in practice, what the evidence is for each, and what specifically managers should stop doing as agentic AI absorbs the workflow tasks that used to define management coordination. Three rules, all backed by named institutional research from the last six weeks.

What's the biggest mistake managers are making with AI right now?

The most common mistake is treating AI as an efficiency tool for existing management work, rather than as a structural change in what management work is.

A manager using AI to draft team updates faster, summarize meetings faster, or process status reports faster is operating inside the old model — same work, less time. The 2026 evidence is that the firms producing competitive advantage from AI are operating inside a different model: same time, different work. Freshworks announced the week of May 4 that it cut roughly 500 employees, 11% of its workforce, while Q1 revenue grew 16% year-over-year to $228.6 million. CEO Dennis Woodside told analysts that more than half the company's code is now AI-written. The savings are being redirected into Freshservice, the company's AI-powered IT service management product. Atlassian made a comparable cut the prior month with similar logic.

The pattern matters for the manager-level question because it shows what the organizational endpoint looks like: a leaner team where the remaining people do work that compounds, supported by AI that handles the routine. The manager's job in that endpoint is not "the same management work, faster." It is the work AI cannot do — coaching, judgment calls, cross-functional translation, and the kind of trust-building conversations that determine whether high performers stay.

Deloitte's 2026 Global Technology Leadership Study, released April 30, supplies the institutional confirmation. The study, which surveyed more than 660 senior tech executives, found that 79% of leaders report delivering measurable enterprise value as their top priority, and 71% of organizations now have five or more tech leaders in the C-suite — a structural fragmentation that pushes coordination work outward to the manager layer (Deloitte press release, April 30, 2026). The expanded coordination demand is precisely the work managers should be using AI to absorb, freeing capacity for the work that actually requires a human manager.

The mistake, then, is identifiable. It is the manager who runs the same five status meetings, but with AI-summarized notes; same one-on-ones, but with AI-prepared talking points; same approval workflows, but routed through an AI assistant. The cycle time is faster. The actual job has not changed. And the data says the manager next door — the one who used AI to eliminate three of those status meetings outright and reinvested the time in coaching the two underperformers on the team — is producing measurably different outcomes.

How do you measure whether AI is actually working for your team?

Stop measuring time saved. Start measuring what the saved time was reinvested in.

The Deloitte study found that 42% of senior tech leaders report low or no ROI on AI investments, despite the elevated value mandate. The number is uncomfortable but diagnostically useful: it points to the measurement gap. Most organizations are tracking AI inputs (licenses bought, employees trained, tasks automated) rather than outputs (decisions improved, judgment work done, coaching hours delivered). The inputs are easy to count and tell you nothing about whether the AI is producing the leverage the firm is paying for.

For a mid-market manager running a 12-person team, the practical measurement framework has three layers. The first is time displaced — how many hours per week is the team spending on AI-assisted versus manual work. This is the easy number. Most managers can produce it from a Slack-and-calendar audit in a single afternoon.

The second is time reinvested. This is the number most managers do not track. If AI saved the team eight hours of meeting summarization per week, where did those eight hours go. If the answer is "we ended up doing more of the same work," the AI investment is producing zero leverage; the team is on the wrong side of the Freshworks pattern. If the answer is "we ran two coaching sessions, did one cross-functional review, and shipped one experiment we had been deferring," the AI investment is producing the leverage the firm bought it for.

The third is outcome-side measurement. The Deloitte data is clear that 79% of tech leaders are now accountable for enterprise value outcomes. The same accountability flows down to the manager layer, even if the formal measurement does not. A manager who can show, at the end of Q3, that AI-saved capacity was reinvested in two specific outcomes (a hire that closed faster, a customer escalation that was resolved on the manager's judgment rather than escalated upward, a cross-functional initiative that landed because the manager had time to drive it) is producing the kind of evidence that boards and CEOs are now looking for.

Gallup's State of the Global Workplace 2026 report, published April 8, adds the failure-mode evidence.

The report found that fewer than 15% of employees in AI-integrated organizations say their manager actively supports their AI use (Gallup release, April 8, 2026). Translated into the measurement frame: in 85% of organizations, managers are not making AI a topic of active coaching, which means the AI-saved time is not being reinvested in the high-leverage activity. The measurement framework that works has to start with the manager's coaching cadence around AI, not with the tooling.

What should managers stop doing now that AI agents can do it?

The cleanest 2026 answer is structured around four categories of work that managers can stop doing themselves, plus one category they should never delegate.

Stop: Routine status synthesis. Meeting notes, weekly updates, quarterly recaps, project summaries — all of this is now agentic-AI work. Managers spending more than 30 minutes per week generating these artifacts manually are mispricing their time against the 2026 cost of AI tooling.

Stop: First-pass screening of routine inputs. Sales pipeline reviews, candidate resumes for first-pass fit, support ticket triage, supplier responses to RFPs. The agentic systems available in 2026 do this work to a quality that is consistently good enough for first-pass screening, with the manager's judgment reserved for the cases that get flagged for review. The manager who screens 60 resumes manually in 2026 is the manager who screened 60 in 2019. The cost has changed.

Stop: Routine drafting work. First drafts of policy memos, performance review templates, project briefs, internal communications. AI handles the structural draft; the manager handles the substantive judgment about what the document should say. The 2026 split is roughly: AI drafts, manager edits and decides. Managers still drafting from scratch are absorbing time that should be going somewhere else.

Stop: Calendar and meeting logistics. Scheduling, agenda preparation, follow-up emails, action-item tracking. Agentic AI tools released through Q1 2026 — including the kinds of agentic capabilities Harvard Business School's Tsedal Neeley and Expedia's Ritcha Ranjan describe in their February 2026 working note — handle this layer at a quality that no longer requires human attention (HBS Working Knowledge, February 11, 2026). The exception is meetings where the human relationships at stake make the logistical care itself part of the message.

Never delegate: The conversation where a person needs to be heard, challenged, or supported. Coaching the underperformer. Telling someone they are not getting the promotion. Walking through what is going wrong on the team. Explaining a strategy decision that affects someone's career. AI cannot do this work; the value is precisely that a human is doing it. Managers who use AI to script or shortcut these conversations are eliminating the part of management that justifies the role.

The synthesis is straightforward but uncomfortable. The 2026 manager job has narrowed to three things: judgment work that requires human stakes, coaching work that builds the team's capacity, and decision work that requires accountability. Everything else is now AI-assisted or AI-completed. The managers who recognize this narrowing, and reinvest in the three things, will be valued. The managers who use AI to do the same job faster will be progressively automated out of the job they are working harder to keep.

What this means for the next 90 days

Three actions for any manager working through this question in May–July 2026.

Action 1: Run the time audit. Ten-minute exercise: list every recurring activity on the calendar. Mark each one with a category (status, screening, drafting, logistics, judgment, coaching). Add up the hours. The first three categories should now be largely AI-assisted. If they are not, the next 30 days is about getting them there.

Action 2: Define the reinvestment. Before automating any task, decide what the saved time is going to. Specific commitment: this hour saved per week becomes one additional 1:1 with a high-potential team member. This four hours saved per quarter becomes the cross-functional initiative that has been deferred. The reinvestment has to be specific, or the time disappears into the same shape of work.

Action 3: Make AI a coaching topic. Per the Gallup data, 85% of managers are not actively coaching their teams on AI use. The fix is small: one of the next four 1:1s with each direct report includes the question, "where could AI absorb 30% of what you are spending time on, and what would you do with the time." The conversation surfaces the team's actual leverage points and signals the manager's expectation that AI is part of the job, not a tool the team uses on its own time.

The 2026 institutional evidence has converged. Managers who use AI to do the same job, faster, will lose the job. Managers who use AI to do a different job — narrower, higher-judgment, more coaching-dense — will keep the role and grow into the expanded mandate the data describes. The question is not whether to use AI. It is whether the use is structural or cosmetic. The data now answers the question, if managers are willing to read it.