Weekly signal

This week (July 6–14, 2026) marketing teams saw agentic AI move from experimental helpers to product-ready workflow partners. Three technical product shifts matter most: OpenAI rolled out a long‑horizon agent for multi‑step work (ChatGPT Work, powered by GPT‑5.6) that explicitly targets cross‑app campaign work and scheduled tasks; platform vendors pushed agent primitives into core marketing stacks (Salesforce locked Agentforce creation into the new Builder starting the week of July 13 and published observability and marketing‑email connection updates; Microsoft documented Sales/role agents across Dynamics/Copilot as broadly available or rolling to GA); and major ad/search platforms continue testing AI‑authored ad summaries and Gemini‑powered ad tooling which change how paid creative and landing pages are represented to users.

What changed

1. OpenAI launched ChatGPT Work (GPT‑5.6 family) as a first‑class “work” agent for longer workflows: it can plan a campaign, gather context from connected apps, run scheduled checks, and produce final deliverables (docs, slides, spreadsheets, Sites). OpenAI published rollout guidance for org admins and champions.

2. Salesforce began enforcing a migration to the new Agentforce Builder (New Agent creation disabled in legacy builder the week of July 13) and added features that matter to marketers: connection inputs for marketing email, refined observability, model selection controls, and easier subagent management.

3. Microsoft continued pushing role‑based agents into the sales/CRM stack (Sales Agent in Microsoft 365 Copilot / Dynamics 365), making it easier for marketing and sales to coordinate outreach and campaign signals within Copilot workflows. Docs and setup guides were updated in early July.

4. Google is publicly testing AI‑generated summaries inside Search ad placements and has expanded Gemini‑powered ad tooling in country launches (India reported July 9). The tests introduce a platform‑authored message layer that can supersede advertiser copy in visible space.

5. Smaller vendor activity: vendor product updates and marketplaces for agent skills (Accord Agents, Agensi reaching 2,000 skills) show an active ecosystem for reusable, marketing‑focused agent components—useful for teams that want prebuilt campaign skills.

What to do with it

• Treat agents as workflow platforms, not fancy copy generators. Pilot ChatGPT Work on one repeatable campaign workflow (research → brief → localized assets → QA) to learn governance, cost, and approval gating—use OpenAI’s rollout guide and changelog to set limits.
• Prioritize migration and observability for enterprise agent stacks. If you use Salesforce Agentforce, schedule the migration to the new Builder and enable Refined Agent Analytics to track marketing KPIs and quality scorers.
• Align marketing + revenue ops on Copilot/Sales Agent configs so automated outreach preserves brand voice and lead scoring rules. Capture test cases that show how agent actions map to CRM state changes.
• Audit landing pages and ad copy quality for Google’s AI ad tests: ensure page content, schema and creative accurately reflect brand claims because platform summaries can amplify mistakes or cannibalize advertiser messaging.
• Consider buying or testing vetted agent skills (Agensi, Accord) for asset production and demo workflows as a low‑risk start—validate outputs for brand tone and compliance before scale.