Relay shuts down, YouTube tightens AI: July 21 dispatch
Relay sets deletion dates, YouTube defines inauthentic AI video, MCP changes its plumbing, and Adobe puts AI guidance inside its camera.

This sheet contains partner links. A purchase through one earns Runbook a commission at no additional cost to you. How we make money.
Relay users need a replacement plan before their automations and stored data disappear. YouTube publishers need a template audit. And anyone connecting AI assistants to business systems should delay a protocol upgrade until next week. Set aside 30 minutes today to protect the workflows and content that already make or save your business money.
Relay sets dates for shutdown and data deletion
Relay will delete free accounts after August 15 and paid accounts after September 14. The automation service confirmed those dates in its shutdown notice, along with a July 16 stop on new accounts and paid upgrades. Existing workflows keep running during the wind-down, but anything you fail to export will go with the account.
This matters if Relay moves leads, sends customer notices, or updates the software that stores your customers and sends follow-ups. A working automation can be invisible right up to the day it stops. Do not start by choosing a replacement. Start by recording every trigger, action, approval, connected account, and owner. Then compare that inventory with the options in this Zapier alternatives breakdown.
Open each workflow in Relay and export it as JSON, a structured text file another system can read, plus the AI prompt. Export run history, then download each Relay Table as a CSV spreadsheet file. Rebuild one revenue-critical flow in Make, Zapier, or n8n and run the old and new versions side by side for three successful jobs. The Make versus Zapier guide will help you place that first rebuild.
Your move
YouTube defines the AI videos it will not monetize
YouTube now names three problem areas: mass-produced template videos, upsetting or manipulative material, and AI-made presenters discussing sensitive subjects such as health, law, or finance. Its current channel monetization policy says reviewers can examine your newest and most-viewed videos, the content producing most watch time, and even titles and thumbnails.
The business consequence is lost ad income, not a blanket ban on AI production. Original explanation and a clear story still count. If your workflow swaps a few words into the same script, voice, images, and sequence for dozens of uploads, the template itself is now a financial risk. Owners publishing tutorials should pay attention because repeating information already common on YouTube without adding firsthand substance can also fail the test.
Open YouTube Studio > Analytics > Content and list the ten videos with the most watch time. Add columns for repeated script, repeated visuals, original demonstration, and human review. Rewrite the weakest format before producing another batch. If you use an AI presenter, keep it away from sensitive advice and review the setup tradeoffs in the AI avatar video guide.
MCP prepares a breaking plumbing change
MCP will change on July 28. Model Context Protocol, the common connection format that lets an AI assistant use outside tools and data, is moving to a stateless core, meaning the receiving server will no longer need to remember every conversation connection. The official release-candidate announcement also warns that the revision contains breaking changes.
Most owners should make no change today. Your hosted AI tools will handle this underneath the screen. But if a contractor built a private connection from an assistant into your calendar, customer database, or reporting system, ask which MCP version it uses. Record that answer beside the connection owner and renewal date.
Do not upgrade the live connection before the final specification lands. Copy it into a test workspace, check login, list the available actions, and run one read-only task before allowing it to create or edit records. This is the same discipline used in a dependable lead-to-customer-system automation: test the handoff before trusting the follow-up.
Adobe adds AI guidance at the camera
Adobe put an experimental AI Playground into Project Indigo version 1.1 on July 20. The Adobe Research announcement says invited iPhone users can remove distractions, change style or lighting, request a photo critique, and get suggestions while they are still standing at the scene.
For a business, the useful part is the reshoot prompt. A staff member photographing a finished job or product can correct framing before leaving, which is cheaper than trying to repair a weak image later. The limit is substantial: access reaches only a small percentage of users for a few weeks, edits require an internet connection, results are roughly 2,000 pixels on each side, and the feature only works on photos captured in Indigo.
If the sparkle icon appears after you update to version 1.1, test Photo Guidance on internal material first. Save the untouched original and the edited result. Do not put this experiment into a required publishing workflow yet, and do not treat its critique as approval for a customer-facing claim.
On the bench
- Watch Relay exports for missing app credentials. The files document the steps, but you will still reconnect each business account in the replacement.
- Test one YouTube format with an original demonstration and named reviewer before scaling production.
- Recheck MCP connections after July 28, when the final specification is scheduled to ship.
For the next practical build, use the automation guides to wire the replacement flow and document its owner. Get the next build.
About Runbook
AI tools and automation builds for marketers. What to use, how to wire it, and the workflow to copy this week. How we work
GET THE NEXT DISPATCH
Run the next build before your competitors read about it.
One short email when an AI tool or automation actually changes the work, with the build to copy.
No send unless there is a build worth running.
// keep_reading
Related builds

Zapier flags AI builders, Google splits data: July 7 dispatch
Zapier's AI-builder survey, Google's Search data setting, Reddit's spam filters, and database app builders change your stack checks.

Zapier alternatives for automation in 2026
Zapier is easiest to start, but Make, n8n and GoHighLevel can be cheaper or cleaner for the right automation job.

Make vs Zapier for marketing automation in 2026
Make is cheaper at volume, Zapier is easier to hand to a team. Here is the pricing math, workflow fit, and setup path for each.