Creating Agent Mode Rules for AI Chat (and Testing Them)

Agent Mode Rules are plain-language coaching rules that shape the wording and behavior of AI Chat when it is operating in the Agent persona. They are how you teach AI Chat your team's voice, your preferred response patterns, and the constraints you want every agent-facing response to follow.

This article covers what Agent Mode Rules are, where to configure them, the two places they now apply, how to test them, and how they differ from Triage Rules.

• What Agent Mode Rules Are
• Where to Configure Agent Mode Rules
• Where Agent Mode Rules Apply
• Company- and User-Scoped Rules
• Writing Effective Rules
• Testing Agent Mode Rules
• Agent Mode Rules vs. Triage Rules

What Agent Mode Rules Are

Agent Mode Rules are short, natural-language instructions that AI Chat reads at the start of every Agent-mode response. They do not change what data AI Chat can see and they do not change routing or categorization decisions — they shape how AI Chat phrases and structures its answer.

Typical uses include:

• Setting a tone (e.g., "Be professional and friendly but not too formal")
• Enforcing format conventions (e.g., "Use numbered steps for procedures")
• Calling out preferred terminology or vendor language
• Reminding the AI of constraints that apply to every response, such as "Always include the ticket number when referencing a ticket"

Where to Configure Agent Mode Rules

Agent Mode Rules live in one place. The rules for an area are now a single Markdown document rather than a grid of separate rules added one at a time:

1. Open Settings > Rules
2. Review the current rules in the read-only display, which scrolls to show the full document
3. Click Edit to open the full-screen editor in a blade
4. Add or revise your rules in plain language, then save

The editor opens the whole rules document at once, so you write and organize all of your rules in a single place instead of creating and editing one rule at a time. Once saved, the document applies immediately to the next AI Chat Agent-mode turn and to the next Fast Feedback agent summary that triage generates.

Saving is safe. The editor skips saving when nothing has changed, and if a save fails it reports an error and stays open so your edits are not lost. The first time you save, any rules you previously kept as separate rows are migrated into the single document automatically.

Organizing Rules With Comments

You can structure the document for your own readability using comments. Any line that starts with a hash followed by a space — for example, # Tone guidance — is treated as a comment for your own organization and is stripped out before the text is sent to the AI. This lets you group related rules under headings, leave notes to your teammates, or keep example templates commented out inline for reference.

One nuance matters here. A hash that is not followed by a space is treated as real rule content and is preserved, so priority-style phrases like #1 priority or #urgent still reach the AI. Only lines that begin with a hash and a space (# ) are removed as comments.

Where Agent Mode Rules Apply

As of the 2026-05-14 release, Agent Mode Rules govern two surfaces:

• AI Chat in Agent mode — every Agent-persona conversation, whether started from the home AI Chat page, from a Ticket Details AI Chat dropdown, or from inside a PSA pod (which is always locked to Agent)
• The Fast Feedback agent summary — the internal note that triage posts back to your PSA after analyzing a new ticket. This means the same wording, tone, and formatting standards you set for live chat also apply to the briefing your techs see when a ticket lands

You only have to manage one rule set to keep both surfaces consistent.

Company- and User-Scoped Rules

The Rules tab under Settings > Rules contains Tenant-wide rules that apply to every Agent-mode response. If you need a rule that applies only when AI Chat is operating in the context of a specific company or a specific user, you can add scoped rules from the relevant detail page:

• Company-scoped rules live on the company's detail page in the Companies tab. They apply only when AI Chat is reasoning about that company — for example, when a tech opens an AI Chat conversation anchored to one of that company's tickets.
• User-scoped rules live on the user's detail page in the Users tab. They apply only when AI Chat is reasoning about that user.

Scoped rules layer on top of the Tenant-wide rules in Settings > Rules. The Tenant rules always apply; company and user rules apply additionally when the matching entity is in scope.

Writing Effective Rules

Some patterns that work well:

• Keep each rule focused on one behavior. Short, single-purpose rules are easier for the model to follow than long, multi-clause ones
• Prefer positive instructions ("Use numbered steps") over negative ones ("Do not use prose")
• Include an example if the behavior is subtle. "Refer to the customer by their first name, e.g., 'Hi Sam,'" is clearer than "Be familiar"
• Group related rules under comment headings (lines beginning with # ) so the document stays readable as it grows
• Test the rule with realistic prompts before assuming it works (see the next section)

Some patterns to avoid:

• Don't try to enforce factual content with rules. If AI Chat is giving wrong answers, the fix is usually a better KB article, not a tone rule
• Don't stack contradictory rules. If two rules pull in opposite directions, the model will pick one inconsistently

Testing Agent Mode Rules

The most reliable way to test a rule is to use it in a real AI Chat conversation:

1. Open AI Chat from the home page or refresh if it is already open so the rule load is current
2. Switch the persona toggle to Agent
3. Ask a question that should exercise the rule. For a tone rule, ask anything; for a format rule, ask something procedural; for a terminology rule, ask something that would naturally use the wrong word
4. Read the response and confirm the rule was followed
5. If the response did not follow the rule, click Edit in Settings > Rules and tighten the wording — make it shorter, more specific, or add an example — then save and try again

To test how a rule affects the Fast Feedback agent summary, either wait for a new ticket to come through your triage pipeline or re-run triage on a recent ticket and read the resulting internal note in your PSA.

Agent Mode Rules vs. Triage Rules

It is important not to confuse the two rule systems in ServiceAI:

• Agent Mode Rules (this article, Settings > Rules) are coaching rules. They shape wording and tone for AI Chat and the Fast Feedback agent summary. They do not change what gets routed where or how a ticket is categorized
• Triage Rules (Settings > Triage) are decision rules. They control routing, categorization, priority, and field updates during triage. They do not affect the wording of any chat or summary

If you want to change how AI Chat sounds, edit Agent Mode Rules. If you want to change what triage does with a ticket, edit Triage Rules.