Dear Suzannah

We use an AI tool every day. What should we do before it changes or shuts down?

Treat the tool like any other business dependency. List what it touches, what data matters, who owns the process, and how you will move the work before a deadline forces the decision.

Here’s the deal

OpenAI published an Atlas retirement notice with an August 9, 2026 stop date, while Microsoft has also published migration deadlines for parts of its AI platform. The useful lesson is not about one product. It is that AI workflows need exit plans.

What this actually helps you do

For a small company relying on a browser-based AI tool for research, logins, and repeat admin work, this creates a controlled way to move the recurring process to a supported tool without losing access, context, or accountability.

Exactly three benefits

• Protect business continuity when a tool changes or disappears.
• Reduce the chance that important data, browser state, or process knowledge gets stranded.
• Give the replacement tool a clear acceptance test instead of hoping the migration worked.

Step-by-step

1. Inventory the trigger, inputs, outputs, connected accounts, files, and people involved.
2. Separate portable information from tool-specific information. Save critical URLs, prompts, reference files, and instructions in company-controlled storage.
3. Choose the replacement based on the job. Confirm that it can perform the required actions with the right permissions.
4. Run the old and new process side by side on several real cases.
5. Document the cutover date, owner, rollback option, and any work that still requires a person.
6. Retire old access only after the new process passes the acceptance test.

Tips and tricks

• Keep a one-page dependency map for every important AI-assisted process.
• Store source files and business instructions outside the AI product whenever possible.
• Add retirement and migration dates to the same calendar where you track renewals.

Common mistakes

• Waiting until the final week to discover that bookmarks, tabs, or history do not transfer.
• Assuming a replacement tool has the same permissions and behavior.
• Leaving the old tool connected after the migration is complete.

Human review checklist

• Can we name the process owner?
• Do we know what data must be preserved?
• Has the replacement completed real test cases?
• Can we reverse the change if the new process fails?
• Have old permissions been scheduled for removal?

How to measure success

• Failed or interrupted tasks during migration
• Percent of required data and instructions successfully transferred
• Time needed for the replacement process to reach the old quality level

FAQ

Do I need a migration plan for every AI tool?

No. Start with tools that touch customer work, financial work, proprietary data, production systems, or recurring operations.

What if the vendor says data will transfer automatically?

Verify the exact items that transfer. Product notices often distinguish conversation history, browser state, files, credentials, and settings.

Should I keep the old tool active as a backup?

Only for a defined transition period. Permanent duplicate access increases confusion and security exposure.

Glossary

Dependency: A tool, account, file, or system a business process relies on.

Cutover: The point when the new process becomes the primary process.

Rollback: A defined way to return to the previous safe state if the migration fails.

Sources and further reading

• OpenAI ChatGPT Business release notes
• Microsoft Foundry migration guidance
• Microsoft Copilot Studio roadmap

Related: How to Build an AI Rollback Plan Before Launch.

Practical closing note

If an AI tool matters to daily operations, its exit path is part of the system. Save what matters before the vendor’s deadline becomes your emergency.