Adobe Summit 2026: CX Enterprise, Coworkers, and the Agentic Suite

Adobe Summit gave the agentic era its enterprise vocabulary this month. Three new nouns matter.

CX Enterprise is Adobe’s name for customer experience run as one system, with coordinated AI orchestrating across it. Coworker packages agents into goal-directed workflows that operate across scattered data and content. The name choice is telling: an agent as a colleague, not a feature. And Experience Manager was recast as a platform for AI context, brand governance, and delivery. New agents shipped for AEM Sites, and ‘discoverability for the agentic web’ was a headline theme.

We have covered this platform since 2010, and the rhyme is hard to miss. Summit 2012 sold suite coherence for the marketing cloud era. Summit 2026 is selling the same coherence for the agent era. That is not a criticism. Coordination is genuinely valuable and genuinely hard. It just means the buyer’s homework is the same as it has always been.

What actually shipped versus what was announced

Adobe Summits reliably sketch where the enterprise category is heading, about two years early and at premium prices. So sort the announcements into three piles.

Concrete and available: the AEM Sites agents. They scan a site for problems (broken links, metadata gaps, slow templates) and propose fixes for human approval. At the scale of a two-hundred-site university, automated finding is genuinely valuable, because the problem list always outruns the people available to compile it.

Carefully gated: the execution side. Agents that fix what they find still run behind approval steps everywhere Adobe showed them. That is the right design, and it is also an honest indication of maturity. Nobody is betting brand-critical sites on autonomous action yet, and neither should you.

Still aspiration: the fully orchestrated whole, with agents planning campaigns and coordinating across clouds. That is keynote material on the usual timeline. Some version of it will arrive eventually, trimmed and renamed. Price it as roadmap, not product.

The questions to ask before any pilot

With agents, governance is the evaluation. Four questions belong in every conversation.

• What are the authority boundaries? What may the agent see, propose, and change? On which sites, and within what limits? A mis-scoped assistant drafts a bad paragraph. A mis-scoped agent ships one.

• Where is the audit trail? Every agent action should be logged and reviewable, from finding to proposal to approval. Regulated organizations already know to demand this. Everyone should.

• What is the rollback path? What un-does an agent’s change, how fast, and has anyone actually rehearsed it?

• Who reviews, and do they have time? Human-gated agents turn a finding problem into a review-capacity problem. The reviewers belong in the business case next to the license.

What a sensible pilot looks like

For clients who pass the readiness bar, we are scoping pilots the same way this quarter.

Start with one site, not all of them. Turn on the detection tier and let it run for a month before anyone approves a single fix. The first deliverable is not remediation. It is a judgment check: did the agent find real problems, and did it rank them the way your team would have?

In our first test runs, the answer was mostly yes with useful exceptions. The agents are good at the mechanical tier: broken links, missing metadata, slow templates. They are weaker at knowing which problems matter to your institution, which is exactly the judgment your team supplies in the approval step. That division of labor is the whole point, and a month of watching it teaches you more than any demo.

Only then turn on gated execution, on low-risk fixes first, with the rollback path tested before the first approval. Boring, deliberate, and reversible. That is what a good agent pilot looks like in 2026.

The variable that decides everything

Here is the constant that has decided every AEM engagement we have seen: organizational readiness determines whether the platform delivers or gathers dust. The agent era raises that bar in three places.

Data foundations first. Agents reason from your content structures and your analytics. Sites with governed, well-modeled content give agents something to be right about. Sites full of sediment get confident automation of their own mess.

Governance maturity second. The boundaries and review habits above are organizational muscles, not platform features. The platform ships the settings, but the policy has to come from you.

Integration third. The orchestration story assumes the clouds are actually connected to your operations. The gap between suite-as-demoed and suite-as-deployed is always widest right there.

If your organization is strong on all three, pilot the Sites agents this year with the four questions as your contract. If it is weak on any of them, spend this cycle on foundations instead. The capability arrives on Adobe’s schedule. The value arrives on yours.

The direction is correct and the premium is real. The open question is your organization’s readiness, and no keynote has ever been able to answer for it. Some frameworks hold for decades because they are about buyers, not vendors.