Adobe Summit 2025: The Agents Arrive in AEM

Fifteen years of covering this platform teach you how to read a Summit, and the reading rule has held. Adobe’s March announcements reliably sketch the enterprise category’s direction about two years early, at premium prices, with the gap between keynote and shipping doing its usual work. The evidence runs from the personalization-suite announcements of the mid-2010s through the content-supply-chain framing that came before the industry’s.

This year’s direction is unambiguous: agentic. Adobe’s new Agent Orchestrator tops the bill. AEM Sites Optimizer ships agents that spot site problems (speed issues, broken pages, SEO gaps) and carry out fixes with approval. Journey Optimizer gains agents that run tests. And the conductor ambition, agents working in concert across the suite, under rules Adobe is selling as hard as the skill, threads every keynote.

We have covered this platform since its first CQ5 review in 2010, and it is changing its stance. From a platform that serves experiences to a platform that acts. We will cover it the way we cover every Summit. Sort the shipped from the announced. Lead with the rubric that now opens every agent evaluation. And ask the readiness question that has decided this platform’s value for fifteen years and decides it again.

Shipped versus announced

Concrete and near. Sites Optimizer’s problem-spotting tier is real, and you can watch it work. Constant scanning of a site for the problem list every audit knows by heart. The vitals slips and broken links. The metadata gaps, the weak templates. Surfaced as ranked findings with proposed fixes. Read it as the audit-and-watch layer automated. That is genuinely valuable at scale (the two-hundred-site university, the multi-brand enterprise) where the finding backlog always outruns the capacity to find.

Carefully gated. The execution tier (the agents that fix what they find) operates behind human approval steps in everything shown and shipped. That is both the correct design, per every governance piece since the copilot reviews, and the tell about maturity. Adobe is not yet betting brand-critical sites on autonomous action, and neither should you.

Aspiration on familiar timelines. The run-the-whole-show tier. Agents planning campaigns, making variants, working across the suite’s clouds. Keynote-complete, roadmap-priced, and belonging in the two-year reading the Summit rule assigns. Much of it will arrive trimmed, renamed, and useful, per the pattern’s long history.

The rubric that now leads: governance before capability

• Agent authority lines, explicit and inspectable: what may the agent see, propose, and change? On which sites, within what limits? The expose-and-lock rule arriving at the action layer, and the first setup talk of any rollout. Agents inside brand-critical plumbing raise the stakes on every question the 2023 CMS-AI review asked. A mis-scoped copilot drafts a bad paragraph. A mis-scoped agent ships one.

• Audit trails as a first-class requirement: every agent finding, proposal, and action logged and reviewable. The compliance stance the regulated tiers already know to demand, now applied to software that acts. It is the question that separates enterprise-grade agent tooling from demos.

• Rollback paths, rehearsed: what un-does an agent’s change, how fast, verified how? CrowdStrike showed what skipping that rigor costs at world scale. Apply the lesson to this newest class of automatic change. Trust like an engineer: specifically, boundedly, with a rollback path.

• The approval math, priced honestly: human-gated agents turn a problem of finding things into a problem of reviewing them. The staffing line for the reviewers belongs in the business case beside the license.

The readiness question, fifteen years running

We have asked this question in every AEM review since 2010, because it keeps deciding outcomes. The readiness question decides the value, and it has three parts this round.

Data foundations. The agents reason over your site’s signals: the analytics, the content structures, the experiment history. Sites with the wired-up, well-modeled base our measurement coverage prescribes give agents something to be right about. Sites with sediment give them confident inputs to be wrong with.

How grown-up the governance is. The line-setting, approval-routing, and audit habits above are things a team can do, not features a platform ships. The platform ships the surfaces, and the organization supplies the policy. We saw the same thing with copilot configuration: the defaults last until someone owns them.

The wiring capacity. The orchestration value assumes the suite’s clouds are actually wired to your operations. The CDP populated, the journeys instrumented. The perennial Summit gap between suite-as-demoed and suite-as-deployed is widest exactly here.

Teams strong on all three should pilot Sites Optimizer this year, with the rubric above as the contract. Teams weak on any should spend this round on the base, the same advice we have given at every capability wave on this platform. The capability keeps arriving on schedule. The readiness keeps being what separates the reference customers from the shelfware statistics.

Fifteen years of reading Summits

This year’s Summit is worth reading beyond Adobe’s walls. Parallel developments (the Drupal initiative organizing, the WordPress abilities work, the MCP ecosystem maturing) suggest the whole industry is walking the same road at different price points. If the enterprise category follows on the usual two-year lag, then agents inside the CMS are the 2026-2027 buying conversation. The governance rubric above is the checklist to write into RFPs now, while vendors are still deciding what to promise.

The direction is real, as it always is. Fifteen years of Summits have produced a strong record on direction and a comedy record on timelines. Our method (sort the shipped from the announced, lead with governance, decide by readiness) is tuned for exactly that gap.

Some readings hold for fifteen years. The one that says the platform’s value lives in the buyer’s operating maturity, not the keynote’s ambition, has. This year’s agents, for all their real newness, land squarely inside it. The agents arrive to questions that were already here. The sites that answer them before deployment will be the ones the next Summit’s case studies are built on.