The Precursive AI Component

The Precursive AI component is added to Salesforce record pages by an administrator. Depending on how the admin has placed it on a given page, you will see it in one of two layouts:

• Card - a standard component that sits inline with the rest of the page, alongside other components.
• Docked - a floating panel pinned to the bottom-right of the screen. It can be collapsed to a small header strip and expanded back when you want to look at it.

Both layouts display identical content - the only difference is positioning on the page. Which layout you see is determined by the admin per page placement and is not something you change yourself.

When you open a record, the component displays the most recent insight generated for that record by the active Zen Agent.

Switching Between Zen Agents

If multiple Zen Agents are configured for the same record type, or for related record types, you can switch between them using the dropdown at the top of the component.

Agents are grouped in the dropdown by source object. For example, if you are viewing an Opportunity, you may see agents listed under "From Opportunity" as well as agents listed under "From Account". This means you can view insights from an Account-level agent directly from the Opportunity record without navigating away. The component remembers which agent you last selected for each object type, so your preference is kept the next time you return to a similar record.

Refreshing an Insight

Each insight is generated at a point in time. The generation date is displayed on the component. If an insight is more than 24 hours old, a warning indicator appears to let you know the data may be stale. Click Refresh to generate a new, up-to-date insight for the current record.

The component will show a loading indicator while the insight is being generated. Once complete, the updated insight is displayed. The Refresh button has a 60-second cooldown after use - it will be temporarily disabled with a countdown indicator before becoming available again.

Cached Responses

When you refresh, the system first checks whether the underlying context data has changed since the last insight was generated for this record today. If the data has not meaningfully changed, the previous insight is reused rather than generating a new one. This is shown as a "cached response" indicator on the component.

Caching exists to preserve the agent's weekly entitlement usage. Without it, every refresh would consume an LLM request even when nothing about the record had changed.

If the agent's configuration has been updated since the previous insight was generated, the cache is bypassed and a fresh insight is created.

A refresh of a single record consumes a request from the linked Zen Agent's entitlement (unless a cached response is returned).

Understanding the Insight

Insight Content

The insight is displayed as formatted text, typically including a summary of the record's current status, identified risks or opportunities, and any relevant context gathered from related records.

Note on AI accuracy: AI-generated outputs may contain inaccuracies (hallucinations). To address this, Precursive provides citations to underlying Salesforce data and incorporates user feedback loops, enabling you to verify outputs and continuously improve reliability over time. Always review citations to confirm the AI's reasoning aligns with the actual data.

Citations

Key statements within the insight are backed by citations - references to the specific Salesforce records and fields that the AI used to reach that conclusion. Citations appear as dotted underlines on relevant text. Hovering over a citation shows the source record and a link to navigate directly to it.

Citations help you verify the AI's reasoning and quickly navigate to the underlying data. If a statement in the insight is not what you expected, checking the citation tells you exactly what data the AI was looking at.

Score Badge

Some Zen Agents include a custom score configured by the agent builder - for example, an "Account Health" score or a "Project Risk" score. When configured, the score is displayed as a colour-coded badge alongside the insight.

Hovering over the badge shows the AI's rationale for the score it assigned, so you can see why the agent rated the record the way it did.

The colour bands are based on the score value:

• 70-100: Green
• 40-69: Amber
• Below 40: Red

The meaning of the score and what it measures depends on how the agent has been set up by your team.

Sharing Context

You can share the underlying context data that was sent to the AI for this insight by selecting "Share context..." from the Zen Agent dropdown menu. This downloads a file containing all the Salesforce data gathered for this record.

This is useful if you want to take the context into a general-purpose LLM such as ChatGPT for more open-ended questions or deeper exploration than the insight alone provides.

A note on handling the downloaded file: the context file contains real Salesforce data and may include sensitive information depending on what your Context Sources gather. Treat the file as you would any sensitive export - delete it from your device after use, do not paste it into untrusted services, and follow your company's AI usage policies.

Providing Feedback

You can rate each insight using the thumbs up or thumbs down buttons on the insight card.

Clicking either button opens a feedback modal where you can:

• Confirm or change your rating
• Select a feedback category for negative feedback: Inaccurate, Not relevant, Incomplete, Formatting issue, Too verbose, Too brief, Citation, Action, or Other
• Add a free-text comment to explain your feedback in more detail

Your feedback is saved against your user record. You can see at a glance whether you have already rated an insight, and you can update your rating at any time if your view changes. Feedback submitted from email digest links is also captured and attributed correctly.

Feedback is reviewed by Builders to improve agent quality over time.

Acting on Recommended Actions

If the Zen Agent has Actions configured, the insight may include one or more recommended actions based on what the AI found. Recommended actions are displayed in the component alongside the insight content, with a rationale explaining why the action is being suggested and citations showing which data it is based on.

Supervised actions are surfaced as recommendations. You decide whether to trigger them by clicking the action button. The action result is confirmed in the component once complete.

Autonomous actions are run automatically by the AI when it generates the insight, but only when its confidence in the recommendation is high enough. These appear in the component marked with an "AI Added" or "AI Updated" badge, indicating the action has already been taken. Your admin or Builder can tell you which actions in your org are supervised and which run autonomously.

Why You Might Not See an Insight

There are a few reasons why an insight may not be visible on a record page:

• No insight has been generated yet - the Zen Agent may not have run for this record, or the record may not match the agent's filter criteria.
• The insight is being generated - if you have just triggered a refresh, wait for the loading indicator to complete.
• Access control - if your org uses a private sharing model for insights, you will only see an insight if you have access to all of the Salesforce records and fields that were used to generate it. If you do not have the required access, the component will display an access denied message. Contact your administrator if you believe you should have access.
• The component is not on the page layout - contact your administrator to have the Precursive AI component added to the relevant record page.

You do not have the required permission set - you need the Precursive AI Standard permission set. Contact your administrator to request access.