Keep up to date with the latest news and thought leadership.
AI features are becoming part of FileMaker development, but the best use cases aren’t random chat boxes in a layout.
The better question is: where can AI reduce friction in a real workflow?
That might mean helping users summarize notes, classify requests, search by meaning instead of keywords, draft responses, or extract useful structure from messy text. FileMaker’s newer AI script steps support several of these patterns, including getting text responses from models, using natural language with database schema, generating SQL, performing AI-assisted finds, creating embeddings, using retrieval-augmented generation, and controlling AI call logging.
The opportunity is real, but so is the risk. AI should support the workflow, not silently become the workflow.
A good first AI workflow should be narrow and easy to verify.
Good candidates include:
Poor first candidates include:
The safest early pattern is “AI suggests, user confirms.”
FileMaker’s AI script steps rely on configured AI accounts. For example, Claris documents that steps such as Insert Embedding and Perform Semantic Find require a named AI account to be configured in the file before those steps run.
That means AI should be treated like an integration, not like a casual layout feature.
At minimum, define:
AI account name
model or service being used
which scripts can call it
what data may be sent
where results will be stored
whether calls should be logged
This matters because AI workflows often touch sensitive business context. You want to know which data is being sent, why it is being sent, and where the result goes.
One practical workflow is note summarization.
Imagine a service team that records long visit notes. Managers may not have time to read every detail, but they need the key points.
A FileMaker-assisted AI flow could look like this:
User writes or imports service notes
↓
User clicks “Generate Summary”
↓
FileMaker sends selected note text to AI
↓
AI returns a concise summary
↓
Summary is stored in a review field
↓
User edits or approves the result
The key is that the AI output should land in a separate field first.
For example:
ServiceNotes::RawNotes
ServiceNotes::AISummaryDraft
ServiceNotes::FinalSummary
ServiceNotes::SummaryReviewedBy
ServiceNotes::SummaryReviewedAt
This preserves the source note and gives the user a place to review the AI result before it becomes part of the official record.
AI can also help sort messy intake records.
For example, an incoming request might need to be classified as:
The script should not blindly accept the AI output. A stronger pattern is:
AI returns suggested category
AI returns confidence or reasoning
FileMaker stores result as a suggestion
User confirms or changes category
Confirmed value drives workflow
A field structure might look like:
Request::SubmittedText
Request::AISuggestedCategory
Request::AISuggestedPriority
Request::AIReason
Request::FinalCategory
Request::FinalPriority
Request::ReviewedBy
This makes AI useful without letting it quietly control routing on its own.
One of the more interesting FileMaker AI patterns is semantic search.
Traditional FileMaker find is exact or structured. Semantic search lets users find records based on meaning. Claris documents script steps for inserting embedding vectors into records or found sets, then performing semantic finds against that embedded data.
That can be useful when users search for concepts rather than exact words.
For example, a user might search:
“customers who complained about late shipments”
Even if the records do not use that exact phrase, semantic search may help find records with similar meaning.
A practical architecture looks like this:
Source text field
↓
Embedding generated and stored
↓
User enters natural language search
↓
FileMaker performs semantic find
↓
Results are reviewed by user
This can be especially useful for notes, support tickets, case histories, knowledge bases, and project descriptions.
This is one of the most important design rules.
Do not overwrite important user-entered or business-critical fields directly with AI output.
Instead, use a staged field pattern:
OriginalValue
AISuggestedValue
FinalApprovedValue
ReviewedBy
ReviewedAt
This gives the workflow a human checkpoint and makes the system easier to audit.
It also makes users more comfortable. People are more likely to trust an AI-assisted workflow when they can see, edit, and approve the result.
If a workflow depends on users typing the “right” prompt each time, the workflow is fragile.
FileMaker can help by storing prompt templates and using structured script logic to assemble prompts consistently. Claris’s AI script step documentation includes support for setting up prompt templates for use in other AI script steps.
A simple prompt template might include:
You are assisting with service request triage.
Classify the request into one of these categories:
– Billing
– Technical Support
– Operations
– Sales
– Other
Return JSON with:
category
priority
summary
reason
Request text:
<<REQUEST_TEXT>>
Asking for structured output, such as JSON, can make the result easier to parse and store in FileMaker.
AI-assisted features should be more restricted when they touch sensitive data.
Useful guardrails include:
FileMaker’s AI features give developers powerful tools, but the application still needs a governance model.
Good fits include:
Riskier fits include:
The closer the workflow gets to a business decision, the more human review matters.
The best AI features in FileMaker will probably not feel like “AI features.”
They will feel like smoother workflows.
A user clicks a button and gets a clean summary.
A manager finds relevant records faster.
A team triages messy requests with less manual effort.
That is the right bar: AI should reduce friction while FileMaker remains the system that structures, governs, and records the work.
Let’s talk about how we can help you streamline, scale, or innovate—on your terms.
Start the Conversation
