Agentic AI is at the peak. Here's what to do this quarter

If your inbox filled up with agent announcements last week, there's a reason: Gartner IT Symposium/Xpo ran in Orlando from October 19 to 22, and agentic AI was one of its featured topics.

The announcements will keep coming. The useful question isn't whether agents are real. It's what your team can actually run between now and the end of the year.

Where the curve sits

In April, Gartner's 2026 Hype Cycle for Agentic AI placed agentic AI at the Peak of Inflated Expectations. Citing its 2026 CIO and Technology Executive Survey, Gartner reported that only 17% of organizations had deployed AI agents, while more than 60% expected to within two years. The same article says most deployments remain narrowly scoped and that "fully autonomous agents are not ready for the majority of enterprise use cases." It also flags agentic AI governance, security, and FinOps on the curve, and says the need for that discipline is showing up early in adoption, not only after large-scale deployment.

Symposium added fresher numbers in the same direction. In the opening keynote release (October 19), Gartner cited a Q2 2026 survey of more than 2,000 CIOs and technology executives: 71% of organizations expect to deploy agentic AI within the next 12 months, yet only 21% have agentic AI governance capabilities in place. A separate release the same day reported that only 13% of CIOs say their enterprise has realized significant or transformational value from AI so far. As Gartner VP Analyst Alicia Mullery put it: "Volume is not value."

Outside Gartner, KPMG's Q3 2026 AI Pulse (314 US leaders at companies with $1 billion or more in revenue, self-reported) found 62% of organizations building, deploying, or developing AI agents, up from 53% the quarter before. Nearly half of leaders (49%) say they've defined high-risk use cases where agents aren't allowed to make decisions on their own.

Different surveys, different samples, so don't add them together. They tell the same story anyway: intent is running well ahead of the ability to operate agents safely.

What the peak does and doesn't mean

A peak isn't a verdict that agents don't work. It's a signal that expectations are ahead of readiness, and that some projects started on expectations alone will stop.

Gartner has put a number on that. It's a prediction, not a measurement: over 40% of agentic AI projects will be canceled by the end of 2027 (June 2025), "due to escalating costs, unclear business value or inadequate risk controls." The same release describes "agent washing," the rebranding of assistants, RPA, and chatbots as agents, and estimates that only about 130 of the thousands of agentic AI vendors are real.

The three causes in that prediction are useful because you can plan for every one of them this quarter. Cost, value, and risk controls are design decisions, not luck.

Four things to do before December

Our AI Transformation Playbook is built for exactly this gap between "we want agents" and "we can operate them with gates." Here's how its sections translate into the next 90 days.

1. Make sure someone owns the initiative

The playbook's prerequisite comes before any phase: an executive sponsor, a small team with a named chair, a one-page mandate, and real protected time. We wrote about this in Before Phase 1: name the people who own the initiative.

One mandate item matters more at the peak: "Decide what runs now, what waits, and what never ships." Gartner's Symposium data backs this up. Organizations with guidelines for when an AI pilot should not move to production were 2.3 times more likely to be "vanguards" (Gartner's term for CIOs delivering better business outcomes alongside high CEO-perceived AI value), yet only 31% of CIOs reported having them. This quarter, write that stopping rule down.

2. Locate yourself honestly

The playbook's rule is short: if you're between two phases, pick the lower one. "Safer beats aspirational." Judge it by what runs in production with logs, not by what was demoed last week.

For many teams, that means Phase 1, Fluency, and that's fine. If you haven't inventoried how people already use AI, start there. Start by asking, not banning covers how.

3. Run one Phase 2 Gated Agent, properly

If you're ready for an agent, the playbook's Phase 2, Gated Agents, is the pattern: one workflow with high repetition, a clear definition of done, and failure you can recover from. Draw the gate as a checklist someone could run at 2 a.m. Strip the tools to the minimum. Build a golden set before the first prompt change, and watch for the moment the approve button turns into muscle memory.

Set a cost envelope and a kill switch before you widen. At Symposium, Gartner's Mary Mesaglio called cost "the biggest tactical risk to scaling AI." The playbook's 90-day picture for Phase 2 ends with a line worth taping to the wall: "Widen volume, not autonomy."

4. Classify, then tier

Give every agent an entry in an agent registry, starting with the first: purpose, criticality, autonomy class. Then match oversight to the tier. The playbook says it plainly: "One policy for every agent is governance theater." Gartner's Svetlana Sicular made the same point at Symposium: "Tailor oversight to your level of autonomy."

Three questions for the next agent pitch

When a vendor, or your own board, brings you an agent this quarter, ask:

1. What can it do without a human, and what can it touch? If the answer is "anything," it isn't ready.
2. Who approves its work, and what do they actually see? A gate nobody reads isn't a gate.
3. How will you know it still works after the next change? If nobody has an answer, there's no eval yet.

Clear answers are a good sign. Vague ones point straight at the cost, value, and risk-control gaps in Gartner's forecast.

The quiet advantage

Peaks pass. The teams that come out of this one in good shape probably won't be the ones with the most agents. They'll be the ones who can say who owns the initiative, which phase they're really in, and what happens when an agent gets something wrong.

Get the playbook

The prerequisite, all four phases, and the cross-cutting practices, including Classify, then tier, are in the 231 Labs AI Transformation Playbook. It's free at 231labs.com/playbook.