what to watch, and what it prices

Agentforce Observability is the Salesforce utility for monitoring, analyzing, and optimizing AI brokers in manufacturing. We construct and instrument Agentforce deployments at Xavor Company in Irvine, California by way of our Salesforce growth and administration companies. Session tracing data each flip, motion, and reasoning step an agent takes into Information Cloud. Three surfaces carry agent monitoring: dwell conversations, mixture metrics, and particular person session traces. And Salesforce made Agentforce Observability unmetered throughout all Agentforce SKUs.

Tracing prices nothing to activate now, which removes the rationale most groups delayed IT.

What Agentforce Observability captures

Session tracing data each flip, motion, and reasoning step an agent takes into Information Cloud. 4 elements carry that recording, and every one holds a special layer of the identical interplay.

• Agentforce Session Tracing captures and shops detailed agent interplay information in a unified information mannequin on Information Cloud.
• The Session Tracing Information Mannequin logs actions, immediate inputs, LLM outputs, and error messages.
• Agent Platform Tracing captures each Agentforce motion execution as an OpenTelemetry hint tree saved in Information Cloud.
• Agent Analytics makes use of the Session Tracing Information Mannequin to trace each occasion and switch throughout classes.

A hint data what the agent did and the reasoning IT adopted to get there.

One naming be aware price carrying into your personal documentation. Salesforce Assist now publishes Agent Analytics as the present product, with Legacy Agentforce Analytics retiring from Could 2026. Some Salesforce pages nonetheless use the older title.

Learn how to monitor Agentforce brokers

Three surfaces carry agent monitoring: dwell conversations, mixture metrics, and particular person session traces. Every solutions a special query, and groups that use just one discover out late.

Mixture metrics inform you one thing is mistaken, and a session hint tells you the place.

Watching dwell conversations

Omni-Supervisor exhibits energetic agent conversations as they occur, with escalation constructed into the view. Open the Omni-Supervisor tab and choose the Brokers part.

Raised flags seem on the prime of the dialog record. Clicking Switch to Rep strikes the shopper to a human.

Command Middle for Service exhibits all energetic conversations, displaying as much as 100 at a time.

The 4 metrics Agent Analytics experiences

Agent Analytics measures suggestions, escalation, deflection, and abandonment throughout each session. The dashboards sit inside Agentforce Studio.

Agent Analytics additionally surfaces session outcomes and adoption developments. The work of getting agent information right into a queryable state sits within the Information 360 implementation work behind agent analytics.

Studying a single session hint

Clicking a response inside a hint exhibits the reasoning and the trail the mannequin adopted. That makes a hint nearer to a debugger than a report.

Classes and intents carry high quality scores in opposition to every buyer intent. A low rating on one intent isolates the issue sooner than a falling mixture.

Traces exist to isolate failure factors, which is a special job from measuring efficiency.

What Agentforce Observability prices now

Salesforce made Agentforce Observability unmetered throughout all Agentforce SKUs, and the boundary is printed exactly. The change covers session tracing, telemetry information ingestion, the metrics and aggregations powering dashboards, and out-of-the-box dashboard queries. Utilization nonetheless seems in Digital Pockets.

What stays metered is net-new customization slightly than monitoring itself.

Salesforce states observability is unmetered for each buyer, and each drill-down, alert analysis, and session hint now runs at no extra Information Cloud value.

The distinction is price holding. Agent actions nonetheless eat Flex Credit, which Salesforce publishes at $500 per 100,000, with a normal motion drawing 20 credit at ten cents. Watching these actions not prices something.

One consequence follows straight. Groups that beforehand restricted dashboard entry to at least one or two individuals can now open IT to each builder and stakeholder.

The broader return calculation sits in how Agentforce delivers return on AI funding.

How multi-agent session tracing works

A single session can route by way of a number of specialist brokers, and the hint data every handoff. Sub-agent routing seems within the session circulation alongside the turns and actions.

Multi-agent traces are included free for each Agentforce buyer. Salesforce describes them as isolating points associated to ineffective subagents and actions.

A failed handoff between two brokers seems to be similar to a single agent failing till the hint separates them.

We constructed a multi-agent Salesforce deployment with specialist brokers and orchestrated handoffs for a specialty retailer. Three specialists dealt with product discovery, common help, and order queries.

The orchestration layer transitions between them when a buyer pivots mid-conversation, from a suggestion request to a delivery standing question, with out dropping the thread. Studying that transition in a hint is the way you inform a routing drawback from a reasoning drawback.

The place session tracing stops

Three documented limits form how a workforce operationalizes tracing, and every surprises individuals late. None seems in most printed steering.

• Draft and inactive brokers produce nothing: solely energetic agent classes are captured.
• The export API is beta: the Session Hint OTel API accepts one session ID per request, helps no bulk question, and returns classes began inside the earlier 72 hours.
• Digital Pockets lags: IT experiences on a 72-hour delay, whereas the AI utilization information mannequin in Information 360 refreshes each 5 minutes.

A draft agent produces no traces, so the agent you’re testing is invisible.

Two conditions sit beneath all of IT. Session tracing wants Customary Information Mannequin model 1.130 or increased. And audit logging captures solely occasions after you allow IT, with no backfill of what got here earlier than.

Turning findings into agent modifications

4 patterns seem typically sufficient in session information to level at a selected change every time. Every maps to a metric Agent Analytics already experiences.

The primary three are diagnosable from the hint alone. Open a failing session, comply with the reasoning, and the hole normally names itself.

An accurate reply that scores badly is the failure groups misdiagnose most frequently.

The repair there’s hardly ever the mannequin. IT is tone, size, or a solution that resolves the query with out acknowledging what the particular person requested. Designing that layer is a part of Xavor’s agentic AI growth companies.

A manufacturing readiness guidelines

5 checks separate an instrumented Agentforce deployment from one which solely seems instrumented. Run them earlier than the agent handles quantity.

• Customary Information Mannequin 1.130 or increased put in: session tracing will depend on IT.
• Session Tracing enabled: flip IT on below Einstein Audit, Analytics, and Monitoring Setup.
• Brokers energetic slightly than draft: a draft agent generates no hint information.
• Audit logging enabled early: historic utilization shouldn’t be backfilled as soon as you turn IT on.
• A named proprietor for the dashboards: entry is not cost-limited, so the constraint is consideration.

Audit logging captures solely what occurs after you flip IT on, which makes the allow date a knowledge boundary.

Groups extending the interface alongside instrumentation will discover the API floor lined in extending Agentforce with customized UI and APIs.

Instrument earlier than you scale

The price argument for delaying instrumentation not holds, which leaves solely the work.

Turning tracing on takes a setup toggle and a knowledge mannequin model examine. Studying what IT produces, and changing that into agent modifications, is a standing observe slightly than a mission.

Tracing prices nothing to activate now, which removes the rationale most groups delayed IT. Turning what IT data into agent modifications is the half that takes work. In order for you that scoped in opposition to brokers you already run, [email protected] reaches our Salesforce workforce.

Concerning the Creator

Salman is a Salesforce CRM Guide and Structure Lead at Xavor with over 14 years {of professional} IT expertise. Licensed in Salesforce and Azure IoT, he designs advanced software program architectures and delivers high-impact cloud options for demanding world hi-tech purchasers.