The Copilots Have Shipped: A Buyer's Review of AI in the CMS

The chat window came to the content platform this year, and the arrival is now general enough to review. Jetpack’s AI Assistant writes and revises inside WordPress’s block editor. Adobe is threading Firefly and its GenAI services through AEM’s content supply chain, generation in the DAM, variation tooling in the authoring flow. Drupal’s module ecosystem has assembled its OpenAI integrations with its usual communal speed. Every DXP keynote this fall leads with a copilot, the industry having converged on both the pattern and the word inside four quarters.

The buyer’s problem is that the category is being marketed by vendor and experienced by job. So this review is organized the way we organize every tooling assessment: by the job, not the logo. With maturity ratings the public evidence and early rollouts support. And with the governance work, December’s policy homework, now a year old and due to graduate, carried throughout. The reviews the keynotes will not give you are precisely the ones about what these features do to your workflow’s control surfaces. Three tiers, one configuration assignment, and the pattern note. In that order.

Mature enough for governed use now

The jobs where the skill is real, the risk is manageable under the December lines, and our deployments already pay. Drafting and revision assistance. The in-editor tier (‘draft this section, tighten this paragraph, adjust this tone’) works as advertised. The block-editor and authoring-flow hookups put it where writers actually work. That beats the paste-from-chat-window routine it replaces, on both ease and oversight. The December rule holds without modification: assistance before the review gate, never generation past it.

Boiling text down. The abstract from the article, the excerpt from the page, the meeting-notes tier. Reliably useful, low-risk under review, and quietly valuable at scale. The metadata and summary debt every content inventory finds, finally cheap to pay down, reviewed.

And translation support, the sleeper strength. The machine-translation tier built into content workflows has quietly become the strongest case in the set. Draft translations of workflow-managed content, made in-platform, flow into exactly the review-state machinery our translation-pipeline coverage built. The human reviewer holds the same quality gate as ever. The pattern across all three: the mature tier assists inside existing workflows and existing gates. That is why it is mature. The platforms did not have to invent governance for it, merely respect what was there.

Promising with conditions, and demo-grade despite the keynotes

• Image generation in the DAM pipeline: promising, with strings. The generate-and-vary tooling arriving in the asset workflows is genuinely useful for the exploring-and-supporting tiers. The conditions are exactly Firefly’s. Licensing stance and legal-cover terms matter, Adobe’s edge is precisely there. The proof boundary holds: the admissions-brochure test does not care which platform generated the image. And the disclosure policies extend to the DAM word for word.

• The agentic promises: demo-grade, despite the keynotes. The auto-tagging of whole sites. The variants that write themselves. The agents that reshape content with no one watching. Every platform demos them. None of our testing supports production trust yet. The error rates on taxonomy work, the brand risk of unsupervised variation, the audit-trail gaps. The advice is the usual one for capability that rewards waiting. Let the platforms mature it on other people’s sites. Revisit on the annual re-grade. Treat ‘agentic’ in a 2023 keynote the way we treat every frontier adjective: as a roadmap claim dressed as a product.

• The evaluation rule, transferable: grade each copilot feature by where it sits relative to your review gates. Inside the gates, assisting humans who decide, is buyable now. Past the gates, deciding and publishing, is a governance question with a feature’s name attached, and the burden of proof runs accordingly.

The configuration assignment: December’s homework graduates

The governance work is now due in its working form. Last December we sequenced the policies. Disclosure, quality gates, expertise. Workflow placement and legal. This year’s platform releases convert them from documents into configuration. The copilot features ship with permission surfaces, and the assignment is to set them deliberately.

Who can call up generation: mapped by role, per the same expose-and-lock approach every new capability has received. The drafting tier broadly. The generation tier with care. The agentic tier off. What gets logged: invocation records, provenance metadata where the platform supports it, per the audit-trail needs your compliance posture already implies. And where human review cannot be skipped: the publish gates, enforced in workflow states rather than in memos, so the December line between help and pipeline is a system property rather than a promise.

Sites that run this configuration pass this quarter will have converted a year of policy work into enforced practice. Sites that leave the defaults will have made the same decisions by omission, at defaults chosen by vendors whose incentives are adoption-shaped rather than governance-shaped. The pass takes a sprint. The defaults last until someone audits them.

AI capability is now a parity feature

The pattern note, stated plainly. AI capability is becoming a parity feature at suite speed. Four quarters from ChatGPT’s release to copilots in every major CMS is the fastest capability spread we have recorded. Compare the multi-year absorption timelines of mobile editing, REST APIs, or media management.

When capability turns commodity this fast, the edge shifts to wiring and safety design. Which platform’s permission model is governable. Whose audit trails are real. Whose review-gate hookup respects existing workflow. Whose defaults are trustworthy. Exactly the layer where the platforms’ instincts visibly diverge. Adobe selling legally shielded pipelines to enterprises. WordPress shipping assistant features to the long tail. Drupal assembling composable pieces for the governed middle. And exactly the layer this audience should be evaluating on.

Which yields the buyer’s verdict compressed to three words, and our advice for every copilot conversation ahead: buy governance, not demos. The demos will keep improving on everyone’s platform more or less equally. The governance architectures are diverging now. They are harder to retrofit than any feature, and they are, per this entire year’s file, where the institutions that adopt this era well will turn out to have made their actual choices. The copilots have shipped. Configure them like you meant it.