Agent adoption

How fast are AI agents actually being adopted?

Agentic AI is the loudest claim in enterprise software and the hardest to verify. This tracks the share of newly documented enterprise AI deployments that are agentic — quarter by quarter, and broken out by industry, use case and business domain.

Agentic now

39%

of the last 12 months' deployments
vs 2023–24

+8 pp

was 30%
Agent deployments

1,176

of 3,905 documented
Multi-agent

38%

of agent deployments
Data updated 3 minutes ago

Agents went from 7% of newly published enterprise AI deployments in Q1 2022 to 35% in Q2 2026 — a 4.7× rise. Across the trailing 12 months the agent share sits at 39% against 30% for the 2023–24 era, a move of +8 pp.

The curve

Agent share of newly published deployments, by quarter

Each bar is the share of that quarter's newly documented enterprise AI deployments that our classifier reads as agentic. The bars are on an absolute 0–100% scale, so the slope is the real slope.

Quarters with fewer than 25 published deployments are not charted; the final quarter (*) is still filling and is greyed out. Publication-dated: this tracks when evidence was published, not when a system went live.

By industry

Which industries have entered the agent era

Agent share of the last 12 months' deployments per industry, against the same industry's 2023–24 baseline. A dot on the change chip marks a move large enough to clear a 95% significance test.

  1. Energy & Utilities+33 pp51%
    19 of 37
  2. Professional Services+6 pp49%
    47 of 96
  3. Automotive+9 pp49%
    16 of 33
  4. Manufacturing+19 pp48%
    39 of 81
  5. Logistics+20 pp46%
    19 of 41
  6. Tech & Comms−0 pp44%
    79 of 180
  7. Consumer & Food+30 pp42%
    25 of 60
  8. Public Sector−0 pp37%
    26 of 71
Read the industry note

Industries need at least 30 published deployments in both windows to be compared. Full narrative: the agent-adoption note.

By use case

Which use cases went agentic fastest

The same trailing-12-months vs 2023–24 comparison, per primary use-case type. Some categories were agentic from the start; others crossed over only in the last year.

  1. Automotive operations automation+11 pp92%
    24 of 26
  2. Workflow automation−4 pp67%
    43 of 64
  3. Clinical documentation+14 pp50%
    13 of 26
  4. Patient engagement+4 pp50%
    14 of 28
  5. Customer service automation−4 pp49%
    26 of 53
  6. AI platform+10 pp43%
    25 of 58
  7. Compliance automation+3 pp41%
    11 of 27
  8. Business process automation+29 pp40%
    8 of 20

Use-case types need at least 20 published deployments in both windows. Showing the 13 qualifying types, highest current agent share first.

By business domain

Which business functions agents reached first

Agent share by the business function the deployment serves. A deployment can serve more than one function, so these shares are not a partition of the corpus.

  1. IT & Security+13 pp50%
    110 of 220
  2. Sales+6 pp48%
    65 of 136
  3. Finance+21 pp47%
    128 of 274
  4. Legal & Compliance+11 pp42%
    228 of 538
  5. Research & Development+12 pp41%
    185 of 447
  6. Operations+9 pp41%
    439 of 1,070
  7. Supply Chain+9 pp40%
    138 of 348
  8. Customer Service−3 pp39%
    135 of 344

Business domains need at least 20 published deployments in both windows.

ⓘ How agent adoption is measured

A deployment counts as agentic when our classifier reads the documented system as taking multi-step actions on a goal, not when the vendor calls it an agent. That is a stricter bar than marketing language, so these shares run below vendor claims.

Every figure is publication-dated: it tracks when source-linked evidence was published, not when a system went live. Newly published evidence about older deployments therefore lands in the quarter it was published.

The "now" window is the trailing 365 days (from 2025-07-30); the baseline is everything published from 2023 up to that cutoff. Change chips carry a dot when a 95% two-proportion test clears; without it, treat the move as noise.

1,176 agent deployments of 3,905 classified · 444 multi-agent.