## Weekly signal

The useful signal this week is that agentic marketing is moving from “AI creates assets” to “AI runs parts of the operating loop”: analyze performance, decide next actions, coordinate channels, enforce brand rules, and increasingly mediate shopping or ad discovery. Coverage is current through May 11, 2026; May 12 had not occurred at scan time.

## What changed

1. Adobe pushed agentic CX orchestration closer to marketer workflows. Adobe Research’s May 4 Summit recap highlighted Proactive Insights and Root Cause Analysis for automatically surfacing business-performance changes, Brand Intelligence for continuously learning brand rules, Brand Concierge for conversational experiences, and CX Enterprise Coworker for building and executing multi-step marketing workflows from a selected outcome such as launching a campaign or responding to a market signal.

2. Adobe also published a practical multi-agent design-review pattern. Agentic-DRS uses specialist agents for typography, color, alignment, spacing, composition, and other design dimensions, coordinated by a meta-agent. The interesting builder takeaway is not just “AI critiques creative,” but the architecture: multiple narrow reviewers, structured design descriptions, relevant examples, and measurable feedback quality.

3. Salesforce/Tableau reframed analytics as an agent action layer. Tableau’s Agentic Analytics Platform turns governed business logic, semantic models, and metadata into “trusted knowledge” that agents can use to answer and act across Slack, Teams, Google Workspace, Salesforce, Claude, ChatGPT, and other surfaces via MCP-style delivery. For marketing teams, that is a direct challenge to dashboard-centric reporting.

4. AI assistant surfaces are becoming paid media surfaces. Axios reported that OpenAI launched a U.S. beta self-serve Ads Manager for ChatGPT campaigns, with direct advertiser buying, agency access, ad-tech partners, CPC buying, and new measurement tools. This matters because discovery, search, and shopping increasingly happen inside answer engines and agent interfaces, not only web search pages.

5. The martech stack is being rewired around context. MarTech’s analysis of the State of Martech 2026 says the landscape grew only 0.7% to 15,505 tools, but nearly 1,500 were added and more than 1,300 disappeared. It argues AI is becoming the value layer above SaaS infrastructure, while context engineering is now the strategic bottleneck. It also cites 90.3% of marketing organizations using AI agents in some capacity, but only 23.3% in full production.

## What to do with it

Audit one marketing workflow end to end: inputs, decision rules, brand constraints, data access, human approvals, and final system writes. Do not start with a general “marketing agent.” Start with a narrow loop such as weekly performance diagnosis, creative QA, audience refresh, campaign anomaly response, or sales-ready account research.

Build a marketing context layer before expanding autonomy: brand rules, product truth, offer eligibility, inventory, campaign history, creative examples, audience definitions, suppression rules, and measurement events. Then give agents read access first, draft/write access second, and autonomous action only after logs, rollback, and approval thresholds are working.