Use the tools your team chooses.
You choose the agent framework, test runner and release process. We agree the source connections, data formats and integration work for your deployment.
For AI vendors, implementers and internal teams
Real workflow data.
Use real cases to build and test your agent.
Bring case records, expert reasoning and approved rules into development. Connect through MCP or the command line. Draft AI tests, called evals, with our Claude and Codex skills.
the decision.
From knowledge to requirements
Define the checks.
Name the next action.
Ask how experts handle gaps and exceptions. Link their reasons to the records and approved rules.
Give developers the checks, allowed actions and person to contact when the agent must stop.
See the interview methodFrom real work to tested automation
Prepare the workflow.
Then test the agent.
- 01
Reconstruct the workflow
Collect the agreed records, documents, messages and activity logs. Check which case each event belongs to. Map the routine steps and exceptions. Measure case volume, time spent and completion quality; label any estimates.
- 02
Capture expert judgment
Review cases with the people who handled them. Ask what they checked, why they chose an action and what would have changed their decision.
- 03
Resolve the rule
Have the person responsible for the process check the explanation against policy. Ask them to approve when the agent may act and when a person must decide.
- 04
Choose the step and method
Choose recurring work with a clear benefit. Decide whether each step needs a fixed rule, a system connection, AI help or a person. Specify allowed actions, required approvals and what to do when the step fails.
- 05
Access the data
Connect your AI assistant through MCP, or read the data using FieldSignal’s command-line interface (CLI).
- 06
Draft the tests
Use a FieldSignal skill in Claude or Codex to draft tests. Check the inputs and expected actions. Resolve open decisions and exclude information that became available only after the action.
- 07
Test the change
Adapt the cases to your test tools. Check routine cases, exceptions and steps where a person must take over. Repeat the tests after changing prompts, models, tools or policy. Use each failure to find the rule or implementation to fix.
- 08
Pilot and measure
Run a small pilot. Compare quality, staff time and running costs with the workflow before automation. Count review, corrections and requests for help. Expand only after the results meet your agreed requirements.
Planned capabilities
Make the evidence easier to inspect.
Planned capability
Show the work
Show a real case on screen. The planned AI interviewer will ask about your decisions and draft a map. Check it against the records.
Planned capabilityProcess metrics
Estimate the time saved, effort to build and costs that remain. Check the estimates against measured pilot results.
Your development environment
Use your own tools.
Review the tests and name the owners.
Have the process owner approve the rules and developers check the results. Assign someone to review exceptions and measure the change.
Developer docs are coming soon.
Contact us for current access and setup instructions. Check the current documentation before choosing commands, connections or test tools.
Docs — coming soonStart with one workflow
Start with one deployment.
Bring a workflow, its records and someone who can demonstrate it. Agree the instructions and tests your developers need.
Agree the scope, responsibilities and price before starting.