Weekly signal

This week (Sep 28–Oct 6, 2026) the creative stack kept shifting from single-request generators to agentic pipelines that coordinate multiple models, app connectors, and content provenance. Key signals: Adobe turned up production-grade MCP connectors so creative ops agents can act inside Workfront Fusion and Experience Cloud; Google pushed creative-first agent capabilities (music, voice, connected creative apps) into Gemini and AI Studio; Google Research published a set of agentic frameworks for coherent long-form video (Co-Director, CANVAS, A²RD, VQQA); and industry events are centering agentic workflows for marketing and studios.

What changed

1. Adobe enabled Workfront Fusion as a Model Context Protocol (MCP) server and built Coworker (Adobe’s agent) integrations — Fusion MCP went live Oct 1, 2026. That means MCP-compatible agents (Claude, ChatGPT custom MCP, Gemini, VS Code, custom agents) can read and — where allowed — trigger Fusion scenarios, executions, webhooks and assets with the same permissions as the logged user. Admin controls, read/write labeling, and explicit confirmations for destructive actions are built in.

2. Adobe’s CX Enterprise docs and product updates (Oct 1) also highlight Agent Orchestrator, AI Monitoring, AI Credits accounting, and automatic C2PA provenance attachments for generative outputs across Experience Cloud — useful guardrails for creative teams shipping client-facing assets.

3. Google’s monthly AI recap (Oct 2) consolidated creative-focused agent features: Lyria 3.5 music generation in Gemini/AI Studio/Flow Music, expressive Gemini 3.8 Live audio/Live Extended Thinking models and new Connected Apps (Adobe, Picsart, Squarespace, Webflow) for in‑app agent workflows. Those changes lower friction for agents to create and assemble music, voice, images, and web assets in an integrated session.

4. Google Research published an agentic “AI video co-director” research suite (Sept 24) — Co-Director, CANVAS, A²RD, and VQQA — that treats long-form video as a global planning + world-state tracking problem. The work shows practical components (visual memory, bandit-driven creative selection, closed-loop VLM critique) that materially reduce identity drift and cascading errors in multi-shot video pipelines.

5. Events and industry convenings (Prompt Live New York, Oct 6) are foregrounding agentic use cases for marketing, advertising and studio workflows — a sign that builders and brand teams are moving from experiments to process changes.

What to do with it

• If you run creative ops or agency workflows: pilot Fusion MCP + Coworker on a non-production Fusion org to evaluate how agents can triage brief→scenario execution and surface audit logs; test read vs write MCP tool scopes and confirm destructive-action confirmations.

• For product and studio teams: evaluate Google’s Lyria 3.5 and Gemini Live audio in a sandbox for rapid prototyping (demos, tempates, voice casting). Use Connected Apps to reduce tab-switching in multi-step agent jobs; instrument cost (agent jobs consume AI Credits) and governance early.

• For video teams: study the Co-Director/CANVAS design patterns (visual memory, hierarchical orchestration, VQQA closed-loop critique) and map them onto your pipeline — start by adding a persistent visual-anchor store and a lightweight VLM-based QA loop to catch drift before expensive renders.

• Governance: adopt C2PA/SynthID provenance and instrument AI Monitoring/credits dashboards so you can trace which assets were agent-generated and what agent acted on them. Require human sign-off for final client-facing creatives.

• Community & scouting: send at least one engineer or creative technologist to Prompt Live (Oct 6) to capture vendor roadmaps and share practical agent flows back to your teams.

(Primary sources: Adobe Workfront Fusion MCP docs, Adobe CX Enterprise docs, Google monthly AI roundup, Google Research co-director post, Prompt Live event page.)