Working with AI

With Logic Foundry connected, you talk to your AI assistant in plain language and it works through Logic Foundry’s tools step by step: it looks things up, plans, makes a change, checks the result and continues. The better your request, the fewer wrong turns it takes. This page shows how to ask, how to follow along, and where the limits are.

How the assistant works through a request

flowchart TD
  R[Your request] --> D[Discover<br/>projects, artifacts, tables]
  D --> P[Plan the steps]
  P --> A[Call a tool]
  A --> C{Done?}
  C -- No --> A
  C -- Yes --> S[Summary for you]

A well-behaved assistant:

• Looks names up instead of guessing. It lists or searches projects, artifacts and tables before it refers to them, because names from earlier conversations may not match your environment.
• Reads the schema first. Before writing logic for a table, it reads the table’s columns and choices so it uses real column names and option values.
• Reads before it writes. Before updating an artifact, it reads the current definition and changes only what you asked.
• Explains as it goes. It tells you which tool it is calling and why, and reports errors in plain words before retrying.

If your assistant skips these steps, ask it to: “Look up the real column names before you write the formula.”

Ask effectively

Do	Example
Name the project	“In the Sales project, …”
Name the table and the real business rule	“…create a validation on opportunity that blocks closing as won when Estimated Revenue is empty.”
Say what kind of artifact you want	“Use a decision table, with one row per customer tier.”
Give names and messages	“Call it ValidateWonOpportunity and show the message Enter the estimated revenue before closing.”
Ask for a plan first on bigger work	“Before you create anything, list the blocks and flows you plan to create.”
Ask for impact before a change	“Show me everything that uses CalculateCreditLimit before you change it.”

Some useful read-only questions to start with:

• “Explain what the ApproveDiscount flow does, step by step.”
• “Which recipes run on the quote table?”
• “Which configuration values does the Sales project use, and where?”
• “Which security roles can write to opportunities, and who holds them?”
• “Run the CalculateShippingCost block for a 12 kg order to Germany.”

What the assistant can and can’t change

The assistant can	The assistant can’t
Create projects and every Logic Fabric artifact type	Change artifacts in a project whose solution is managed in this environment
Update definitions of blocks, flows, recipes, events, schedules, configurations, localized resources and service connections	Do anything your own Dataverse security roles don’t allow
Update, publish and activate business process versions (with Process Orchestrator)	Change security roles, teams or user assignments (security tools are read-only)
Run logic blocks and logic flows	Delete artifacts
Publish Dataverse customizations	Deploy between environments (use the Logic Fabric CLI)

Running flows makes real changes

run_logic_flow runs the flow for real. If the flow creates, updates or deletes records, those changes happen in Dataverse and can’t be undone. Run flows in a development or test environment, and read the assistant’s planned inputs before you approve the call.

Review what the assistant built

The assistant reports what it created or changed. Before you rely on it:

1. Open the artifact in the designer and read it, just as you would review a colleague’s work.
2. Run it with test inputs (see Run and test), or generate tests with Logic Assurance.
3. Publish Dataverse customizations at the end of a session if your changes need it.

Example

You ask: “In the Sales project, create a formula block that calculates a 10% discount on an opportunity’s estimated revenue, called CalcOpportunityDiscount.”

A typical run:

1. list_projects confirms the Sales project exists.
2. get_project_details confirms its solution is unmanaged.
3. metadata_get_entity_attributes reads the opportunity columns and finds Est. Revenue (estimatedvalue, Currency).
4. create_logic_block creates the formula.
5. run_logic_block tries it with an estimated revenue of 50,000 and gets 5,000.
6. The assistant reports the new block and the test result.

Related

• Logic Foundry tools
• Best practices
• Logic Assist agent