Use the AI agent and review results

Give the agent a specific task

Include the sample or group, population, markers, and comparison you want. For example: “Compare the frequency of the CD4 population between the control and treated sample groups, using its parent population as the denominator.” Use the actual names from your workspace.

If the request changes the analysis, inspect the proposed action and any approval prompt before applying it. Ask for the missing information when a result does not identify the samples, population, or calculation used.

Inspect the work

1. Check that the agent is working with the intended workspace and sample selection.
2. Review the tool results and any changes to gates or settings. Inspect the corresponding plots and counts.
3. Ask follow-up questions to clarify a method or refine the analysis. Verify the revised output against the underlying data.
4. Save the reviewed analysis and keep enough context to identify the inputs and settings later.

Prepare statistics and figures

Use Statistics & figures for population tables and figure outputs, or Compare populations for cross-sample overlays. Choose the populations, channels, and sample groups explicitly. Check whether frequencies are reported relative to a parent or another reference population.

For MFI summaries, state whether the measurement is a mean or median and which parameter it describes. Keep sample identifiers and group labels in exported tables so that downstream comparisons remain traceable.

Check before reporting

Verify axis labels, transforms, units, population names, and sample counts. Inspect saved or exported files before sharing them. Include the relevant gate hierarchy, compensation or unmixing choices, and analysis settings with your methods.

For statistical comparisons of exported population tables, use the Flow summary guide and the Group comparison reference. Make the biological replicate and any pairing explicit in your analysis request.

Example of a reviewable request

“Use the current workspace. Show a table for the control and treated groups with one row per sample: sample ID, CD4 event count, parent event count, and frequency of parent. Identify any sample missing that population. Keep the gate boundaries unchanged.” Replace these labels with names that exist in your workspace.

Inspect the returned table against representative samples. If you then request a statistical comparison, state what an independent sample represents and whether observations are paired. Keep that request with the output.

Resolve an incomplete answer

1. Identify the missing element: sample selection, population path, denominator, channel statistic, or method.
2. Ask a targeted follow-up using the actual workspace labels.
3. Inspect the underlying table or plot after the response, including any changed settings.
4. Record unresolved warnings before exporting; a fluent explanation is not evidence that the intended calculation ran.

Export handoff checklist

Keep sample identifiers, groups, population paths, frequency denominators, and the exact channel statistic in the output record. If a table says MFI, confirm whether it refers to a mean or median before using it downstream.

Open the downloaded file and check its labels against the workspace. When continuing in an assay workflow, preserve the mapping from each summary row back to the original flow sample.