McKinsey’s 2026 State of AI survey reports that 40 percent of large organizations are now scaling AI agents, up from 27 percent last year. The figures measure adoption; assessing an agent strategy also requires looking at the work it produces.
Agents have crossed from experimentation into operating plans. Leaders need to work out what changes in the business when software starts taking actions as well as producing answers.
Scaling and absorbing require different work
A company can scale agents in the software sense and still fail to absorb them in the organizational sense.
Scaling means more teams are using them, more workflows are being tested, and more spend is moving from isolated licenses into operating budgets. Absorbing means managers understand where agent work belongs, employees know how to collaborate with it, review loops catch the mistakes that matter, and finance can connect the activity to business value.
McKinsey’s number is useful because it shows a change in buyer behavior. Large companies are putting agents into real work. Gartner has projected that 40 percent of enterprise applications will include task-specific AI agents by the end of 2026, up from less than 5 percent in 2025. The projection concerns application features, while McKinsey reports organization-level adoption.
For a team using an embedded agent, the practical work is to define how its output enters the existing process.
Choose work the team can supervise
Platform selection is easier once the team understands the assignment. Recurring tasks with known inputs, repeatable steps, and a defined reviewer are useful candidates: a weekly sales summary, a first-pass vendor invoice check, a customer support escalation brief, a contract intake review, a meeting-to-task handoff, or a daily exception report.
Not every workflow should get an agent. Ambiguous instructions, undocumented judgment, and systems where a wrong action has an unacceptable cost can make a task unsuitable. Even simple work may cost more to supervise than it saves.
Agents expose management gaps
People often compensate for a messy process with social context: they know who can resolve an ambiguous instruction or which spreadsheet the team actually uses. An agent needs that context documented, along with a route for escalating uncertainty.
Assign responsibility for output approval, instruction maintenance, and quality checks. The workflow owner also needs to distinguish drafting a recommendation from updating a source system, and explain the review job to the people doing it.
Measure the workflow’s operating results
Before expanding a trial, check performance under routine conditions, including the time people spend reviewing and correcting the work. A successful demonstration is one observation in that evaluation.
MIT’s NANDA research found that 95 percent of enterprise generative AI pilots were failing to produce measurable return despite heavy investment. McKinsey’s adoption figures describe a different measure, so a team still needs to establish the return from its own workflow.
For an invoice workflow, compare review-cycle time with the current process. For escalation briefs, track manager preparation time. Other useful measures include sales follow-up speed and exceptions caught before a compliance review reaches the client. Check handoff corrections on day one and after two weeks so the trial captures repeated use.
Write down the assignment
Use a table to record the workflow, its owner, the current problem, the agent’s job, read access, change authority, reviewer, stop condition, and business metric. Keep permissions narrow enough for the first task.
Before the trial starts, resolve missing ownership, unclear output criteria, broad access, or an unscheduled review loop. A vague metric also needs work: the team must be able to tell whether the trial improved the process. Those checks may lead to a smaller first implementation.
An agent can then handle a repeatable part of the workflow while people review its output and remain accountable for the result. Use the evidence from that trial to decide whether to expand.
Research and structure: Mai. Direction and voice: John Lipe.
Sources: McKinsey, “The State of AI: Global Survey 2026,” accessed September 14, 2026; Gartner projection cited in 2026 enterprise agent coverage, task-specific agents in enterprise applications rising from less than 5 percent in 2025 to 40 percent by the end of 2026; MIT NANDA, “The GenAI Divide: State of AI in Business 2025.”