Weekly signal

This week (covering 2026-09-28 through 2026-10-06) the marketing landscape moved from experimentation to product-grade agent deployments that directly affect ad distribution, CRM workflows, and small-business marketing stacks. Two major platform vendors shipped always-on agents and expanded agent-to-tool connectors, while partners and analysts flagged training/data-risk questions that marketers must treat as operational issues.

What changed

1. OpenAI launched "dots": always-on, personal and specialist agents in ChatGPT (rolled out to Pro, Business Premium and enterprise pilots). Dots run on GPT‑6 Astra, have their own cloud computer/browser, and can connect to plugins and work across ChatGPT, Slack and Teams. They include permission controls, auto-review for sensitive actions, and IT-ready specialist-dot features for organizations.

2. OpenAI continues to treat ChatGPT as an ad surface: it expanded ChatGPT Ads (advertisers can run ads inside ChatGPT) and is testing Sponsored Agents — a labeled ad click that opens a branded business-owned agent conversation instead of a landing page. OpenAI also announced integrations with CRM and ecommerce tools to make ChatGPT Ads usable from existing workflows.

3. Meta launched Muse for Small Business, adding skills and dozens of connectors (Shopify, Stripe, QuickBooks, Klaviyo, Canva, Facebook/Instagram ad accounts and Instagram professional analytics) so Muse can read business context and propose or draft marketing work — with the company emphasizing that publishing, sending, or spending requires owner approval. Muse is shipping for small businesses in the US/Canada initial footprint.

4. Independent analysis and rights-watchers called attention to data/Training defaults and consent: third-party connector behaviour and whether connected business data is used for model training or kept read-only matters for privacy, compliance, and contracts — marketers should not assume connectors are training‑opted‑out by default.

What to do with it

1. Treat agents as a new acquisition and conversion channel, not a replacement for your stack. Run a tight pilot (US-first where Sponsored Agents test) and measure cost per qualified conversation, lead-to-close, and downstream ROAS. Use HubSpot/CRM integrations where possible so the CRM remains the system of record.

2. Lock down approvals and scopes before enabling connectors. Require explicit review gates for actions that publish content, spend ad budget, change prices, or access PII — test the owner-approval flows that both Meta and OpenAI expose.

3. Audit connector training and privacy settings now. Verify whether a connector is read-only or allowed to train models, update vendor contracts, and add clauses that forbid undisclosed reuse of customer data for model training. Document consent for customer data flows.

4. Build QA & governance for agent outputs: brand-voice tests, policy-review sampling, and ad creative approval rules. Agents can scale creative quickly, but brand risks and policy flags are real — include human-in-the-loop checkpoints.

5. Instrument attribution and analytics: integrate agent conversations and actions into existing attribution models (CRM events, UTMs, conversation IDs) so you can evaluate performance quickly and stop leaky spend.