Admin settings move from guidance to enforcement
Administrator settings for AI tools stop being advice the model follows and become controls the product enforces.
A monthly re-read of the MarTech and AI Watches and both weekly series, measured against a register of patterns that persists from month to month.
Where the register stands, September 2026
Each row is a pattern, each column a month, ordered by when the pattern first passed the entry rule.
Passed the entry rule (two weeks, two independent sources); evidence still thin.
New, consistent evidence this month from more than one source.
More evidence than usual, including at least one real release (GA, shipped, rolling out).
No new evidence this month, or only announcements with no follow-up.
Evidence this month points the other way. The pattern stays on the record.
Merged into another pattern, or became established practice. Terminal: the row ends there.
A pattern enters the register only with evidence from two different weeks and two independent sources. An asterisk marks a month reconstructed after the fact: May to September 2026 were rebuilt from the Watch records before the first issue.
Ordered by current status. Open a pattern for its month-by-month history and every piece of evidence.
Administrator settings for AI tools stop being advice the model follows and become controls the product enforces.
Open and click tracking in email moves from an account setting to a per-recipient consent attribute, driven by the French and Italian regulators.
Ad platforms, including ChatGPT Ads, become a channel that engagement and data platforms connect to and manage directly.
Vendors decide whether an agent acts under its own identity or the signed-in user's, and wire that choice into sign-in and provisioning.
Engagement and AI vendors publish what their agents may read, write and keep, next to the capability itself.
Engagement and data vendors buy the layers around them, identity, content, intent and data services, instead of building them.
Agents start work on a trigger, a schedule or a saved session, with nobody watching the run.
Archives, logs, traces and anomaly alerts become standard features, so that what a platform or an agent did can be reconstructed.
Consent, quiet hours and frequency rules move from campaign settings into platform-level objects that agents and journeys must respect.
Vendors switch agent features on by default and remove confirmation steps for users, even as admin controls tighten.
Engagement, data and AI platforms expose themselves to agents through MCP servers, which become the common way in.
Reusable skills become the way vendors package agent capability and the thing administrators scan, approve and distribute.
Agents gain write access to campaigns, content and configuration faster than vendors publish what that access is limited to.
CDP functions run inside the customer's own warehouse or lakehouse instead of copying data into the vendor's cloud.
Watermarks and signed provenance metadata on AI-generated content move from policy into shipped products, ahead of the EU AI Act transparency rules.
The layer that knows the customer and the layer that acts on the customer merge into one product.
Generative and answer engine optimisation turn from agency services into products that marketing platforms sell.
The wave of agent launches peaks and gives way to iteration on existing agents, until the question becomes what an agent may do.