How a Restaurant-Tech Company Built a Photo-Reading AI SMS Support Agent with n8n
A restaurant-tech company built an AI SMS support agent that reads customer photos, answers from its Notion knowledge base, and hands unclear cases to humans.
Make Maia, AI agents, and governance: what to adopt (Oct 2026)
Make Maia memory, AI agent fallback retries, provider-specific prompt caching, and Enterprise custom roles, plus a practical checklist for testing each one.
TABLE OF CONTENTS
Make Maia is the AI assistant with memory built into Make's Scenario Builder. It can build, modify, and troubleshoot scenarios, but it doesn't run them for you. Three other recent features matter for teams running AI agents in Make: agent fallback connections give you one retry, prompt caching can cut repeated processing, and custom roles give Enterprise teams finer access control.
Everything below reflects Make's documentation as of October 14, 2026. Beta features change often, so check the linked docs before you roll anything out.
Make Maia: an AI assistant for building, not an autonomous operator
Maia can create, modify, and troubleshoot scenarios from natural-language prompts. Make’s introduction currently describes Maia as public beta on paid plans, with a time-limited trial on Free plans. Do not assume general availability or identical access in every account; check the current Maia introduction.
Maia works with the current scenario. She cannot view other scenarios’ configurations or connect to and call other scenarios for you. For workflows that span scenarios, design and test those connections separately.
Make’s system card also explains that Maia cannot run, activate, or deactivate scenarios. You remain responsible for validating configuration, running tests, and deciding when an automation should go live.
What Maia memory does and does not store
Maia can remember preferences such as the apps you use, preferred response detail, and language. You can add, edit, delete, or disable memories in the chat’s Memory settings.
This is preference memory, not a workflow database. Make says Maia does not save scenario-specific module IDs or webhook URLs, credentials, or other people’s personal records in memory. Keep durable business state in the appropriate database or data store.
A good use case is asking Maia to explain a scenario for a handoff, then refining it with clear requirements. A poor use case is assuming Maia remembers every customer record or can coordinate your entire automation estate.
Agent fallback connections: one retry, not a reliability guarantee
Make’s October 9 release notes document a fallback AI provider connection when the primary provider or model is unavailable.
Make retries the run once using the fallback connection. Make’s AI Provider is the default fallback connection. Review the setting in your agent rather than assuming a particular secondary model is configured.
Before adopting it:
- Test whether the fallback produces the required output structure.
- Confirm it can handle the tools and instructions your workflow needs.
- Add an error route for a failed retry.
- Make downstream writes idempotent so retries do not create duplicate records or messages.
Fallback is one layer of error handling. It does not guarantee that every run succeeds.
Prompt caching: provider-specific behavior
Prompt caching lets an agent reuse instructions and tool definitions. It can reduce latency and credit usage, but benefits depend on model support and repeated prompt content.
The October 9 release notes distinguish two cases:
- Supported Anthropic Claude models: select a model caching duration. With a conversation ID, cached context applies across repeated calls in the same conversation.
- Supported OpenAI models: prompt content is reused automatically.
Keep stable instructions consistent and separate them conceptually from changing inputs. Do not omit information a run needs merely because an earlier run supplied it: caching is not a replacement for business-state storage. For how to structure those instructions, see our guide to structured prompting for Make AI agents.
Compare credit usage and response quality on representative runs before assuming a saving.
Custom roles: an Enterprise governance feature
Make introduced custom roles in August 2026. Its custom roles documentation states that they are available for Enterprise plans and are not supported on Make private instances.
Organization roles and team roles have different scopes. You can choose from predefined permissions rather than granting a broad default role. Some permissions depend on others.
For example, a role can allow scenario viewing and running without access to billing or user management. However, roles are not per-resource access lists: use separate teams when you need to isolate specific resources. For the agent side of governance, including audit logs and limits, see our guide to Make AI agents governance.
Before changing production access:
- List what each person needs to view, run, and manage.
- Configure the appropriate organization and team permissions.
- Test access with a representative user.
- Confirm someone retains the access needed for support and recovery.
A practical rollout checklist
Confirm feature availability and settings in your own account.
Test Maia-generated changes before activating a scenario.
Keep business records outside assistant preference memory.
Test primary-provider failure and the fallback retry.
Verify output schemas under both provider connections.
Measure caching behavior with representative repeated inputs.
Test error routing and duplicate prevention.
Review team boundaries before tightening access.
Not everything above shipped in October: custom roles arrived in August. Start with the failure or access problem your team actually has, and add controls from there.
Get help with your Make workflows
Fallback connections, prompt caching, and custom roles are each easy to switch on and harder to test properly. If you want a second set of eyes before they reach production, book a ZoomFlow session and one of our Make consultants will walk through your failure paths and access rules with you live.
Related Posts
How to build a Claude scheduling assistant for service businesses
Learn how to use Claude to design a practical scheduling assistant for a service business: data model, daily updates, dispatch workflow, and rollout steps.
How Solo Founders Run Companies With AI Agents
See how AI operations agents for small teams can handle development, support, SEO, and task handoffs while founders keep human QA and customer conversations.
