Troubleshoot AI Studio issues

• Open the AI Studio experience you are working on (app, web page, funnel step, or agent node).
• Locate the main prompt or system instructions for the generation.
• Ensure the prompt clearly states:
  • The user context or goal.
  • The desired output format (for example, "Return a JSON object with fields headline, body, cta").
  • Any tone, length, or style constraints.
• Remove conflicting or overly broad instructions such as "do anything" or "respond however you like".

Success check: New generations reference your instructions explicitly and follow the requested format more closely.

• Add 1–3 example inputs and outputs directly into the prompt or dedicated example area.
• Use realistic business content rather than generic placeholders.
• For structured outputs, show a complete, valid example with all required fields.
• Rerun the generation with representative test inputs.

Success check: Generated content aligns with the examples and respects your length and formatting constraints.

• If your experience relies on variables, data sources, or prior steps, verify that:
  • Each variable has a non-empty value before generation.
  • Any referenced records or content exist and load without errors.
• Remove irrelevant or noisy context (for example, entire pages of unrelated text) that might distract the model.
• Rerun with a minimal, focused set of inputs and then layer in more context as needed.

Success check: Output becomes more on-topic and less repetitive as you tighten context.

• Check if the response stops mid-sentence or mid-list, which may indicate length limits.
• Reduce requested output size (for example, "3 bullets" instead of "a full report").
• Avoid instructions that might trigger safety filters (for example, asking for disallowed content).
• If you need long outputs, break the task into multiple steps with intermediate summaries.

Success check: Responses complete naturally without abrupt cutoff, and you see consistent structure across runs.

• Open the agentic flow in Agent Studio within ACEIRT™ Fusion.
• Locate the Start node and visually trace connections through the flow.
• Ensure every branch you want to execute eventually connects back to an End node.
• Drag connector handles to link any missing paths from decision or tool nodes to End.

Success check: You can follow at least one unbroken line of connections from Start through intermediate nodes to End.

• Look for nodes that have no incoming or outgoing connections.
• For any node that should participate in the flow:
  • Connect its input to an upstream node.
  • Connect its output to a downstream node or End.
• Delete nodes that are truly unused to avoid confusion.

Success check: No critical node in your intended path appears visually isolated from the Start/End chain.

• Select each node in the Start → End path and review its required fields.
• Confirm that:
  • Each required input maps to an available variable or static value.
  • Output variables are defined and not accidentally overwritten later.
• Fix any nodes that show validation errors or missing mappings.

Success check: The configuration panel for each node in the main path shows no required-field errors.

• Temporarily disable or remove optional branches and complex logic.
• Trigger a run using a simple, known-good input.
• Observe the execution path and output.
• Re-enable branches one by one, testing after each change.

Success check: The flow completes a run and reaches the End node with visible output or state changes.

• Open the canvas and note the exact variable names used in:
  • Node outputs.
  • Node inputs.
  • Conditional logic or routing.
• Look for small differences like singular vs plural, casing, or extra underscores.
• Standardize variable names across nodes so the same concept uses the same identifier.

Success check: Every variable referenced as an input can be traced to a previous node that defines it with the identical name.

• Identify the first node that should assign each key variable.
• Confirm the node:
  • Actually runs on every path where the variable is needed.
  • Writes to the intended variable name (not a temporary or local field).
• If needed, move variable creation earlier in the flow or duplicate it on alternative branches.

Success check: When you run the flow, you see non-empty values in the first node that sets each variable.

• For each node that consumes variables, open its configuration.
• Confirm that:
  • Inputs reference the correct source variable from the prior node.
  • You are not accidentally mapping a constant string instead of a variable.
  • Output mappings do not overwrite important values unintentionally.
• Update the mappings so each input points to the intended upstream value.

Success check: During a test run, downstream nodes receive the expected values and no longer show undefined or empty fields.

• Insert simple "inspect" steps (for example, debug or log nodes, or temporary display nodes) after critical variable updates.
• Run the flow and capture the intermediate variable values.
• Use these snapshots to pinpoint exactly where values go missing or change unexpectedly.
• Remove or disable the debug nodes after resolving the issue to keep the flow clean.

Success check: You can follow variable values across nodes and confirm they match your expectations at each step.

• Open the integrations area for your ACEIRT™ Fusion workspace.
• Locate the integration or MCP provider that the tool depends on.
• Verify it shows as connected and healthy, without error badges.
• If it appears disconnected or invalid, reconnect it and complete any required authentication steps.

Success check: The integration status shows as active, and basic test actions (if available) succeed.

• Review your ACEIRT™ Fusion role and the project-level permissions.
• Confirm that your role allows:
  • Using the relevant tool or integration.
  • Accessing the underlying data or resources.
• If you lack access, contact a workspace admin to adjust your permissions or run the tool on your behalf.

Success check: The tool no longer fails with permission-related messages when invoked from your account.

• Open the node that invokes the tool or MCP.
• Ensure that:
  • All required fields (like endpoint, resource, or account) are set.
  • Credentials or connection references point to the correct integration.
  • Parameters match the expected types and formats.
• Correct any invalid or outdated configuration and save your changes.

Success check: The tool node passes validation in the editor and no longer shows configuration errors.

• Temporarily create a minimal flow that only calls the problematic tool with static, known-good inputs.
• Run this simplified flow and observe whether the tool still fails.
• If it continues to fail:
  • Check whether your environment restricts outbound connections to the external system.
  • Coordinate with your network team if a firewall or proxy may be blocking access.
• If the minimal flow works, reintroduce dynamic inputs from upstream nodes and test again.

Success check: Tool calls succeed in both the minimal test flow and your full flow, returning the expected data.

• Open the app, web page, funnel, or agentic flow in AI Studio.
• Look for indicators that show whether you are editing a draft or a published version.
• Confirm the last published time and who last published, if shown.
• If you are in a draft state, expect that changes will not affect the live experience until you publish.

Success check: You can clearly identify whether you are working on a draft or on the live published version.

• Make sure any unsaved changes are committed using the Save action in the editor.
• Wait for the save confirmation or status indicator to complete.
• Avoid closing the browser tab or switching projects until the save finishes.

Success check: The editor shows no pending-change indicator, and the current state is stored as the latest draft.

• Use the Publish or Deploy action for the specific experience you changed.
• After publishing completes, open the live URL or entry point that end users use.
• If your experience is part of a funnel or larger flow, ensure you start at the correct step or link.
• Hard-refresh the live view to clear any cached content.

Success check: The live experience now reflects your latest changes when accessed from a separate browser or incognito window.

• Confirm whether other team members recently edited or published the same asset.
• Coordinate to decide which draft should be the source of truth.
• If someone else has a newer draft or publish, merge their changes consciously rather than overwriting them unintentionally.
• Establish a shared workflow for who publishes and when, especially for high-traffic assets.

Success check: Everyone on the team can reproduce the same version in both the editor and live environment.