US BUYER’S GUIDE · 9 PRODUCTS

AI agents & agent platforms

What can the agent do safely, and how will you know it did the right thing?

Compare grounded answers, tool actions, oversight and usage models for business AI agents.

Research desk · Oct 20, 2026 · Methodology

YOUR SHORTLIST STARTS HERE

Compare the options.

Sort by

Highest documented score first. Fit adds up the 28 feature checks below, including AI, across documented plans. Entry price does not include every capability. Unknown evidence appears as a range.

Product / good fit for	Pricing context	Documented fit	Next step
Microsoft Copilot Studio Organizations building governed agents around Microsoft and connected business systems.	$200.00per 25,000-credit pack/monthCopilot Credit pack	69–100/100 documented fit 69% evidence coverage	Full research →
Intercom Teams combining digital customer service with a configurable AI agent and human support inbox.	$29.00per seat/month plus Fin outcomes and channel usage · billed annuallyEssential	64–98/100 documented fit 66% evidence coverage	Full research →
Botpress Teams building customized customer-support agents and willing to manage conversation and AI usage budgets.	$150.00per month, annual commitment; usage beyond allowances extra · billed annuallyPlus	50–100/100 documented fit 50% evidence coverage	Full research →
Relevance AI Teams building several connected agent workflows with an implementation and governance owner.	Custom quotecustom actions and vendor creditsEnterprise	50–100/100 documented fit 50% evidence coverage	Full research →
LiveChat / Text LiveChat / Text Commerce and service teams combining human support with controlled AI handover.	$19.00user/month · billed annuallyText Essential	46–100/100 documented fit 46% evidence coverage	Full research →
Freshdesk Teams building structured email support with a customer portal and a gradual path to richer routing.	$19.00per agent/month · billed annuallyGrowth	43–100/100 documented fit 43% evidence coverage	Full research →
SuiteDash SuiteDash Service firms wanting client portals, sales and delivery in one workspace.	$19.00workspace/monthStart	41–100/100 documented fit 41% evidence coverage	Full research →
Zendesk Service organizations needing a broad help-desk platform with knowledge, reporting and AI options.	See pricingagent seats and applicable AI / service usageConfirm the current Zendesk suite	41–100/100 documented fit 41% evidence coverage	Full research →
Smartsupp Smartsupp Online stores needing product guidance and multilingual human support.	$25.00workspace/month · billed annuallyPlus	24–100/100 documented fit 24% evidence coverage	Full research →

BEYOND A FEATURE CHECKMARK

How deep does the capability go?

Choose up to 3 for a closer look →

28 weighted checks: core capabilities contribute 32 points and workflow depth contributes 68. AI is included at the weight shown for this category. Expand Evidence for scope and sources, or ask for the same demonstration from each supplier.

Show checksShowing fewer rows does not change the score.

AI agents & agent platforms features for all 9 products
Feature / base weight	Microsoft Copilot Studio 69–100/100 documented fit Core capabilities28.8–32 / 32 pts9/10 checks with established status Workflow depth40–68 / 68 pts8/18 checks with established status	Intercom 64–98/100 documented fit Core capabilities30 / 32 pts10/10 checks with established status Workflow depth33.6–68 / 68 pts7/18 checks with established status	Botpress 50–100/100 documented fit Core capabilities29.6–32 / 32 pts9/10 checks with established status Workflow depth20–68 / 68 pts5/18 checks with established status	Relevance AI 50–100/100 documented fit Core capabilities32 / 32 pts10/10 checks with established status Workflow depth18.4–68 / 68 pts4/18 checks with established status	LiveChat / Text 46–100/100 documented fit Core capabilities29.6–32 / 32 pts9/10 checks with established status Workflow depth16–68 / 68 pts3/18 checks with established status	Freshdesk 43–100/100 documented fit Core capabilities24–32 / 32 pts7/10 checks with established status Workflow depth19.2–68 / 68 pts4/18 checks with established status	SuiteDash 41–100/100 documented fit Core capabilities20–32 / 32 pts6/10 checks with established status Workflow depth20.8–68 / 68 pts4/18 checks with established status	Zendesk 41–100/100 documented fit Core capabilities26.4–32 / 32 pts8/10 checks with established status Workflow depth14.4–68 / 68 pts3/18 checks with established status	Smartsupp 24–100/100 documented fit Core capabilities19.2–32 / 32 pts6/10 checks with established status Workflow depth4.8–68 / 68 pts1/18 checks with established status
Core capabilityKnowledge-source grounding4 points · Ground agent answers in configured business documents or help content.	Supported4 / 4 pts Evidence Agents use configured business knowledge. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Fin answers using configured customer-support knowledge. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Agents can use configured knowledge and help content. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Knowledge bases accept uploaded documents and connected knowledge sources. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Web, file and article sources. Official documentation · Oct 20, 2026Source	Supported4 / 4 pts Evidence Agent Studio explicitly supports knowledge grounding. Official documentation · Oct 15, 2026Source	Not verified0–4 / 4 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source	Supported4 / 4 pts Evidence Knowledge-supported customer agents are documented. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Website, product feed and FAQs. Official documentation · Oct 20, 2026Source
Core capabilityConnected business-record context4 points · Retrieve or use business-system records beyond a static article corpus.	Supported4 / 4 pts Evidence Agents can access context through connected business systems. Official documentation · Oct 15, 2026Source	Limited2 / 4 pts Evidence API Data connectors provide business records; Fin cannot directly query Intercom custom attributes or event data for answers. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Integrations connect agents to external APIs and tools. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Agents use connected business-system context. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Open-ticket customer context. Official documentation · Oct 20, 2026Source	Supported4 / 4 pts Evidence AI Actions can call an external API to retrieve business information. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Allowed fields only. Official documentation · Oct 20, 2026Source	Supported4 / 4 pts Evidence AI agents can connect to CRM systems and test connected actions/triggers. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Product-feed context. Official documentation · Oct 20, 2026Source
Depth checkInspectable source evidence5.6 points · Inspect source references or retrieved content supporting an agent answer. Ask for a demonstration Ask a policy question and open the evidence used for the agent’s response. A scenario to run with the supplier; not a completed product test.	Supported5.6 / 5.6 pts Evidence Evaluation drill-down identifies knowledge sources and tools used in the response. Official documentation · Oct 15, 2026Source	Supported5.6 / 5.6 pts Evidence Fin batch-test detail identifies the content sources used in an answer. Official documentation · Oct 15, 2026Source	Not verified0–5.6 / 5.6 pts Evidence The reviewed evidence does not establish this advanced behavior: inspectable source evidence. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Not verified0–5.6 / 5.6 pts Evidence The reviewed evidence does not establish this advanced behavior: inspectable source evidence. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Not verified0–5.6 / 5.6 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source	Supported5.6 / 5.6 pts Evidence Evaluation results identify the knowledge source used for each generated answer. Official documentation · Oct 15, 2026Source	Not verified0–5.6 / 5.6 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source	Not verified0–5.6 / 5.6 pts Evidence The reviewed evidence does not establish this advanced behavior: inspectable source evidence. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Not verified0–5.6 / 5.6 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source
Depth checkRestricted retrieval context6.4 points · Restrict retrieved knowledge or record fields by identity, audience or configured data scope. Ask for a demonstration Ask the same question with two different permissions and show the difference in retrieved context. A scenario to run with the supplier; not a completed product test.	Supported6.4 / 6.4 pts Evidence SharePoint grounding uses the authenticated user’s existing content access, without increasing permissions. Official documentation · Oct 15, 2026Source	Supported6.4 / 6.4 pts Evidence Audience rules restrict content, guidance, data connectors and automations to specified customer groups. Official documentation · Oct 15, 2026Source	Not verified0–6.4 / 6.4 pts Evidence The reviewed evidence does not establish this advanced behavior: restricted retrieval context. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Not verified0–6.4 / 6.4 pts Evidence The reviewed evidence does not establish this advanced behavior: restricted retrieval context. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Supported6.4 / 6.4 pts Evidence Agent-specific knowledge scope. Official documentation · Oct 20, 2026Source	Not verified0–6.4 / 6.4 pts Evidence The reviewed evidence does not establish this advanced behavior: restricted retrieval context. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Supported6.4 / 6.4 pts Evidence Field allowlist. Official documentation · Oct 20, 2026Source	Not verified0–6.4 / 6.4 pts Evidence The reviewed evidence does not establish this advanced behavior: restricted retrieval context. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Not verified0–6.4 / 6.4 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source
Core capabilityConnected tools and actions4 points · Invoke a tool or perform a documented connected-system action.	Supported4 / 4 pts Evidence Connected-system actions are documented. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Data connectors can fetch or update external-system data. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Integration tools connect external-system actions. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Tools support research, scheduling and record-related actions. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Intent-triggered connected actions. Official documentation · Oct 20, 2026Source	Supported4 / 4 pts Evidence Agent Studio invokes configured AI Actions through serverless apps and external APIs. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Predefined action selection. Official documentation · Oct 20, 2026Source	Supported4 / 4 pts Evidence Agents can perform actions in authorized systems; connected actions are tested with the CRM integration. Official documentation · Oct 15, 2026Source	Not verified0–4 / 4 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source
Core capabilityMulti-step agent workflows4 points · Configure an agent to follow or coordinate a multi-step business workflow.	Supported4 / 4 pts Evidence Configured agent workflows are documented. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Procedures and workflows can call configured connectors. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Connected events can trigger bot workflows. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Workforces coordinate agents with tool and condition nodes. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Trigger/action workflows with delays. Official documentation · Oct 20, 2026Source	Supported4 / 4 pts Evidence Agent Studio builds workflows across chat and email channels. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Predefined action selection. Official documentation · Oct 20, 2026Source	Supported4 / 4 pts Evidence Agent dialogue flows and triggers can be configured and tested. Official documentation · Oct 15, 2026Source	Not verified0–4 / 4 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source
Depth checkHuman approval for selected tool actions5.6 points · Require approval before a configured agent performs selected business actions. Ask for a demonstration Ask for a consequential change and show the agent waiting for explicit approval. A scenario to run with the supplier; not a completed product test.	Supported5.6 / 5.6 pts Evidence Agent flows can combine AI stages with explicit human approval stages; advanced approvals are labeled preview. Official documentation · Oct 15, 2026Source	Not verified0–5.6 / 5.6 pts Evidence The reviewed evidence does not establish this advanced behavior: human approval for selected tool actions. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Not verified0–5.6 / 5.6 pts Evidence The reviewed evidence does not establish this advanced behavior: human approval for selected tool actions. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Supported5.6 / 5.6 pts Evidence Tool-specific approval modes pause selected actions for human authorization. Official documentation · Oct 15, 2026Source	Not verified0–5.6 / 5.6 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source	Not verified0–5.6 / 5.6 pts Evidence The reviewed evidence does not establish this advanced behavior: human approval for selected tool actions. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Supported5.6 / 5.6 pts Evidence Optional Wingman approval. Official documentation · Oct 20, 2026Source	Not verified0–5.6 / 5.6 pts Evidence The reviewed evidence does not establish this advanced behavior: human approval for selected tool actions. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Not verified0–5.6 / 5.6 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source
Depth checkStructured tool inputs and validation6.4 points · Define required tool parameters or validation before an action is sent to a business system. Ask for a demonstration Omit or corrupt a required input and show how the agent/tool handles it before execution. A scenario to run with the supplier; not a completed product test.	Supported6.4 / 6.4 pts Evidence Tools expose configurable input collection and filling, with user or maker authentication. Official documentation · Oct 15, 2026Source	Supported6.4 / 6.4 pts Evidence Connector inputs specify what Fin must collect before executing; responses use configured mappings. Official documentation · Oct 15, 2026Source	Not verified0–6.4 / 6.4 pts Evidence The reviewed evidence does not establish this advanced behavior: structured tool inputs and validation. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Supported6.4 / 6.4 pts Evidence Custom tools define user input fields and input configuration. Official documentation · Oct 15, 2026Source	Not verified0–6.4 / 6.4 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source	Supported6.4 / 6.4 pts Evidence API actions define request parameters and authentication, and can be tested before workflow use. Official documentation · Oct 15, 2026Source	Not verified0–6.4 / 6.4 pts Evidence Allowed fields only. Not verified · Oct 20, 2026Source	Supported6.4 / 6.4 pts Evidence Structured data collection blocks gather configured fields before advancing a generative procedure. Official documentation · Oct 15, 2026Source	Not verified0–6.4 / 6.4 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source
Core capabilityAgent monitoring or traces3.2 points · Inspect agent activity, conversation history or execution traces.	Supported3.2 / 3.2 pts Evidence Monitoring and analytics are documented. Official documentation · Oct 15, 2026Source	Supported3.2 / 3.2 pts Evidence Connector health dashboards expose execution logs, usage and performance. Official documentation · Oct 15, 2026Source	Supported3.2 / 3.2 pts Evidence Studio exposes logs, events, errors and files for inspecting behavior. Official documentation · Oct 15, 2026Source	Supported3.2 / 3.2 pts Evidence Agent tracing and cost visibility are documented. Official documentation · Oct 15, 2026Source	Supported3.2 / 3.2 pts Evidence Support and AI reports. Official documentation · Oct 20, 2026Source	Supported3.2 / 3.2 pts Evidence Agent Studio includes performance measurement and monitoring. Official documentation · Oct 15, 2026Source	Supported3.2 / 3.2 pts Evidence Decision run history. Official documentation · Oct 20, 2026Source	Supported3.2 / 3.2 pts Evidence The AI agents workspace has a performance dashboard; pre-March-2026 Advanced purchases have a different experience. Official documentation · Oct 15, 2026Source	Supported3.2 / 3.2 pts Evidence Response, volume and CSAT metrics. Official documentation · Oct 20, 2026Source
Core capabilityHuman handoff or approval3.2 points · Hand work to a person or require an explicit human approval point.	Not verified0–3.2 / 3.2 pts Evidence The reviewed documentation does not establish human handoff or approval at this feature level. This is an evidence gap, not a finding that the capability is absent. Not verified · Oct 15, 2026Source	Supported3.2 / 3.2 pts Evidence Fin can escalate to human agents. Official documentation · Oct 15, 2026Source	Supported3.2 / 3.2 pts Evidence Human Handoff uses Botpress Desk for new workspaces or the configured legacy HITL integration. Official documentation · Oct 15, 2026Source	Supported3.2 / 3.2 pts Evidence Human approval points are documented. Official documentation · Oct 15, 2026Source	Supported3.2 / 3.2 pts Evidence Human transfer with ticket fallback. Official documentation · Oct 20, 2026Source	Not verified0–3.2 / 3.2 pts Evidence The reviewed documentation does not establish human handoff or approval at this feature level. This is an evidence gap, not a finding that the capability is absent. Not verified · Oct 15, 2026Source	Not verified0–3.2 / 3.2 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source	Not verified0–3.2 / 3.2 pts Evidence The reviewed documentation does not establish human handoff or approval at this feature level. This is an evidence gap, not a finding that the capability is absent. Not verified · Oct 15, 2026Source	Supported3.2 / 3.2 pts Evidence Human transfer preserves context. Official documentation · Oct 20, 2026Source
Depth checkStep-level agent execution trace4.8 points · Inspect a sequence of agent decisions/tool calls and their execution results. Ask for a demonstration Open one failed task and show which tool call failed and what happened next. A scenario to run with the supplier; not a completed product test.	Supported4.8 / 4.8 pts Evidence Historical activity can be inspected as an activity map or transcript. Official documentation · Oct 15, 2026Source	Supported4.8 / 4.8 pts Evidence Connector monitoring includes execution logs and request results for troubleshooting. Official documentation · Oct 15, 2026Source	Supported4.8 / 4.8 pts Evidence Inspect exposes conversation logs, integration events and detailed event data for debugging. Official documentation · Oct 15, 2026Source	Not verified0–4.8 / 4.8 pts Evidence The reviewed evidence does not establish this advanced behavior: step-level agent execution trace. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Supported4.8 / 4.8 pts Evidence Conversation-linked step traces. Official documentation · Oct 20, 2026Source	Not verified0–4.8 / 4.8 pts Evidence The reviewed evidence does not establish this advanced behavior: step-level agent execution trace. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Supported4.8 / 4.8 pts Evidence Decision run history. Official documentation · Oct 20, 2026Source	Supported4.8 / 4.8 pts Evidence Testing tools expose dialogue logs and configured action behavior for diagnosis. Official documentation · Oct 15, 2026Source	Not verified0–4.8 / 4.8 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source
Depth checkContext-preserving human escalation4.8 points · Hand a task/conversation to a person with the history or relevant task context available. Ask for a demonstration Escalate midway through a task and show the context available to the human. A scenario to run with the supplier; not a completed product test.	Not verified0–4.8 / 4.8 pts Evidence The reviewed evidence does not establish this advanced behavior: context-preserving human escalation. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Not verified0–4.8 / 4.8 pts Evidence The reviewed evidence does not establish this advanced behavior: context-preserving human escalation. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Supported4.8 / 4.8 pts Evidence Botpress Desk/human handoff preserves the conversation for a human, with assignment and resume controls. Official documentation · Oct 15, 2026Source	Not verified0–4.8 / 4.8 pts Evidence The reviewed evidence does not establish this advanced behavior: context-preserving human escalation. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Supported4.8 / 4.8 pts Evidence Transcript-linked chat-to-ticket handover. Official documentation · Oct 20, 2026Source	Not verified0–4.8 / 4.8 pts Evidence The reviewed evidence does not establish this advanced behavior: context-preserving human escalation. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Not verified0–4.8 / 4.8 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source	Not verified0–4.8 / 4.8 pts Evidence The reviewed evidence does not establish this advanced behavior: context-preserving human escalation. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Supported4.8 / 4.8 pts Evidence Human transfer preserves context. Official documentation · Oct 20, 2026Source
Core capabilityAgent access controls2.4 points · Restrict user, role or data access for agents and their administration.	Supported2.4 / 2.4 pts Evidence The platform documents governance and access controls. Official documentation · Oct 15, 2026Source	Supported2.4 / 2.4 pts Evidence Restricted-data selection and authentication controls limit data exposed to Fin. Official documentation · Oct 15, 2026Source	Supported2.4 / 2.4 pts Evidence Agent-level access controls and workspace roles restrict administration. Official documentation · Oct 15, 2026Source	Supported2.4 / 2.4 pts Evidence Access controls are documented. Official documentation · Oct 15, 2026Source	Supported2.4 / 2.4 pts Evidence Admin-controlled agent settings. Official documentation · Oct 20, 2026Source	Not verified0–2.4 / 2.4 pts Evidence The reviewed documentation does not establish agent access controls at this feature level. This is an evidence gap, not a finding that the capability is absent. Not verified · Oct 15, 2026Source	Not verified0–2.4 / 2.4 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source	Not verified0–2.4 / 2.4 pts Evidence The reviewed documentation does not establish agent access controls at this feature level. This is an evidence gap, not a finding that the capability is absent. Not verified · Oct 15, 2026Source	Not verified0–2.4 / 2.4 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source
Core capabilityAudit or deployment controls2.4 points · Audit agent-related operations or control deployment environments and configuration.	Supported2.4 / 2.4 pts Evidence Audit integration and environment controls are documented. Official documentation · Oct 15, 2026Source	Supported2.4 / 2.4 pts Evidence Data connectors have draft/live states and configurable rollout audiences. Official documentation · Oct 15, 2026Source	Not verified0–2.4 / 2.4 pts Evidence The reviewed documentation does not establish audit or deployment controls at this feature level. This is an evidence gap, not a finding that the capability is absent. Not verified · Oct 15, 2026Source	Supported2.4 / 2.4 pts Evidence Audit logs are documented. Official documentation · Oct 15, 2026Source	Not verified0–2.4 / 2.4 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source	Not verified0–2.4 / 2.4 pts Evidence The reviewed documentation does not establish audit or deployment controls at this feature level. This is an evidence gap, not a finding that the capability is absent. Not verified · Oct 15, 2026Source	Not verified0–2.4 / 2.4 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source	Supported2.4 / 2.4 pts Evidence Enterprise account audit logs expose configuration changes to permitted administrators. Official documentation · Oct 15, 2026Source	Not verified0–2.4 / 2.4 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source
Depth checkAgent configuration history or versions3.2 points · Inspect configuration changes or published agent versions. Ask for a demonstration Change an agent instruction and show the previous version or change record. A scenario to run with the supplier; not a completed product test.	Not verified0–3.2 / 3.2 pts Evidence The reviewed evidence does not establish this advanced behavior: agent configuration history or versions. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Supported3.2 / 3.2 pts Evidence Publishing a Fin Procedure requires a change note so revisions can be traced. Official documentation · Oct 15, 2026Source	Supported3.2 / 3.2 pts Evidence Versions record configuration changes and support comparison and restore; retention varies by plan. Official documentation · Oct 15, 2026Source	Supported3.2 / 3.2 pts Evidence Version Control records agent and tool changes and can restore a previous published version. Official documentation · Oct 15, 2026Source	Not verified0–3.2 / 3.2 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source	Not verified0–3.2 / 3.2 pts Evidence The reviewed evidence does not establish this advanced behavior: agent configuration history or versions. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Not verified0–3.2 / 3.2 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source	Supported3.2 / 3.2 pts Evidence Generative procedures retain version history; published changes are tracked. Official documentation · Oct 15, 2026Source	Not verified0–3.2 / 3.2 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source
Depth checkCredential or connection access boundary4 points · Separate agent credentials and connection permissions from unrestricted general workspace access. Ask for a demonstration Show who can use versus change the agent’s connected-service credentials. A scenario to run with the supplier; not a completed product test.	Supported4 / 4 pts Evidence Tool connections can require each end user’s authorization instead of shared maker credentials. Official documentation · Oct 15, 2026Source	Not verified0–4 / 4 pts Evidence The reviewed evidence does not establish this advanced behavior: credential or connection access boundary. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Not verified0–4 / 4 pts Evidence The reviewed evidence does not establish this advanced behavior: credential or connection access boundary. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Not verified0–4 / 4 pts Evidence The reviewed evidence does not establish this advanced behavior: credential or connection access boundary. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Not verified0–4 / 4 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source	Not verified0–4 / 4 pts Evidence The reviewed evidence does not establish this advanced behavior: credential or connection access boundary. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Not verified0–4 / 4 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source	Not verified0–4 / 4 pts Evidence The reviewed evidence does not establish this advanced behavior: credential or connection access boundary. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Not verified0–4 / 4 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source
Core capabilityAgent behavior configuration2.4 points · Configure instructions, behavior or agent workflow logic.	Supported2.4 / 2.4 pts Evidence Agent behavior and workflows can be configured; credits and environment limits apply. Official documentation · Oct 15, 2026Source	Supported2.4 / 2.4 pts Evidence Fin supports guidance, personality settings, knowledge and automations. Official documentation · Oct 15, 2026Source	Supported2.4 / 2.4 pts Evidence Agent-building and behavior configuration are documented. Official documentation · Oct 15, 2026Source	Supported2.4 / 2.4 pts Evidence Agents, tools and workforces can be configured. Official documentation · Oct 15, 2026Source	Supported2.4 / 2.4 pts Evidence Admin-controlled agent settings. Official documentation · Oct 20, 2026Source	Supported2.4 / 2.4 pts Evidence Agent Studio configures agent behavior and deployment. Official documentation · Oct 15, 2026Source	Supported2.4 / 2.4 pts Evidence Selectable model/provider. Official documentation · Oct 20, 2026Source	Supported2.4 / 2.4 pts Evidence The AI agents workspace provides agent configuration. Official documentation · Oct 15, 2026Source	Supported2.4 / 2.4 pts Evidence Preview before publishing. Official documentation · Oct 20, 2026Source
Core capabilityAgent testing or evaluation2.4 points · Test or evaluate configured agents with documented evaluation, test or comparison tools.	Supported2.4 / 2.4 pts Evidence Evaluations run test sets, compare responses and export results. Official documentation · Oct 15, 2026Source	Supported2.4 / 2.4 pts Evidence Batch Test evaluates questions, sources and automations; tests support uploaded or historical questions. Official documentation · Oct 15, 2026Source	Supported2.4 / 2.4 pts Evidence ADK evals simulate conversations and assert on replies or workflow state. Official documentation · Oct 15, 2026Source	Supported2.4 / 2.4 pts Evidence Agent evaluations and A/B testing are documented; commercial allowances are custom. Official documentation · Oct 15, 2026Source	Supported2.4 / 2.4 pts Evidence Preview without executing skills. Official documentation · Oct 20, 2026Source	Supported2.4 / 2.4 pts Evidence Freshworks describes Agent Studio as a place to build, test and monitor agents. Official documentation · Oct 15, 2026Source	Supported2.4 / 2.4 pts Evidence Sample-record test. Official documentation · Oct 20, 2026Source	Supported2.4 / 2.4 pts Evidence A Test AI agent control and test-environment workflows are documented; the separate conversation simulator is EAP. Official documentation · Oct 15, 2026Source	Supported2.4 / 2.4 pts Evidence Preview before publishing. Official documentation · Oct 20, 2026Source
Depth checkRepeatable agent test sets3.2 points · Run a saved collection of test cases against changed agent behavior. Ask for a demonstration Run the same saved test set before and after changing instructions and compare failures. A scenario to run with the supplier; not a completed product test.	Supported3.2 / 3.2 pts Evidence Saved test sets can be rerun after changing knowledge, instructions or tools. Official documentation · Oct 15, 2026Source	Supported3.2 / 3.2 pts Evidence Saved batch tests run questions from past conversations or imported/custom test sets. Official documentation · Oct 15, 2026Source	Supported3.2 / 3.2 pts Evidence ADK conversation scripts define repeatable evaluation cases with response and tool assertions. Official documentation · Oct 15, 2026Source	Supported3.2 / 3.2 pts Evidence The documented Evals release adds repeatable test cases and criteria for agent validation. Official documentation · Oct 15, 2026Source	Not verified0–3.2 / 3.2 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source	Supported3.2 / 3.2 pts Evidence The saved query list can be rerun unchanged or edited, with up to 100 queries per run. Official documentation · Oct 15, 2026Source	Not verified0–3.2 / 3.2 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source	Not verified0–3.2 / 3.2 pts Evidence The reviewed evidence does not establish this advanced behavior: repeatable agent test sets. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Not verified0–3.2 / 3.2 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source
Depth checkInspect evaluation evidence or outcomes4 points · Inspect per-case responses, tool results or evaluation outcomes beyond one overall score. Ask for a demonstration Open a failed test and show the expected outcome, actual response and supporting trace. A scenario to run with the supplier; not a completed product test.	Supported4 / 4 pts Evidence Per-case evaluation detail exposes actual responses, expectations, reasoning and used knowledge/tools. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Per-question test detail shows answers, sources, guidance and automation context. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence The evaluation runner reports per-case pass/fail against explicit expected responses, tool calls and parameters. Official documentation · Oct 15, 2026Source	Not verified0–4 / 4 pts Evidence The reviewed evidence does not establish this advanced behavior: inspect evaluation evidence or outcomes. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Not verified0–4 / 4 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source	Supported4 / 4 pts Evidence Test detail includes the query, answer status, response, source and user rating; results can be exported. Official documentation · Oct 15, 2026Source	Supported4 / 4 pts Evidence Sample-record test. Official documentation · Oct 20, 2026Source	Not verified0–4 / 4 pts Evidence The reviewed evidence does not establish this advanced behavior: inspect evaluation evidence or outcomes. Request the demonstration above; absence of public evidence does not establish absence of the feature. Not verified · Oct 15, 2026Source	Not verified0–4 / 4 pts Evidence Not established by reviewed sources. Not verified · Oct 20, 2026Source
Depth checkLimit the number of agent reasoning iterations2.5 points · Set a low iteration cap on a test agent with a looping tool. Show where execution stops and how the failure is reported. Ask for a demonstration Set a low iteration cap on a test agent with a looping tool. Show where execution stops and how the failure is reported. A scenario to run with the supplier; not a completed product test.	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source
Depth checkStream a response before generation completes2.5 points · Demonstrate "Stream a response before generation completes" with test data. Show the result, an exception case, and the required plan. Ask for a demonstration Demonstrate "Stream a response before generation completes" with test data. Show the result, an exception case, and the required plan. A scenario to run with the supplier; not a completed product test.	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source
Depth checkSwitch to a configured backup model after failure2.5 points · Demonstrate "Switch to a configured backup model after failure" with test data. Show the result, an exception case, and the required plan. Ask for a demonstration Demonstrate "Switch to a configured backup model after failure" with test data. Show the result, an exception case, and the required plan. A scenario to run with the supplier; not a completed product test.	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source
Depth checkTest agent behavior against prompt-injection cases2.5 points · Demonstrate "Test agent behavior against prompt-injection cases" with test data. Show the result, an exception case, and the required plan. Ask for a demonstration Demonstrate "Test agent behavior against prompt-injection cases" with test data. Show the result, an exception case, and the required plan. A scenario to run with the supplier; not a completed product test.	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source
Depth checkDelete one conversation's persistent agent memory2.5 points · Demonstrate "Delete one conversation's persistent agent memory" with test data. Show the result, an exception case, and the required plan. Ask for a demonstration Demonstrate "Delete one conversation's persistent agent memory" with test data. Show the result, an exception case, and the required plan. A scenario to run with the supplier; not a completed product test.	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source
Depth checkRun independent agent tools in parallel2.5 points · Demonstrate "Run independent agent tools in parallel" with test data. Show the result, an exception case, and the required plan. Ask for a demonstration Demonstrate "Run independent agent tools in parallel" with test data. Show the result, an exception case, and the required plan. A scenario to run with the supplier; not a completed product test.	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source
Depth checkRedact sensitive data before model processing2.5 points · Demonstrate "Redact sensitive data before model processing" with test data. Show the result, an exception case, and the required plan. Ask for a demonstration Demonstrate "Redact sensitive data before model processing" with test data. Show the result, an exception case, and the required plan. A scenario to run with the supplier; not a completed product test.	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source
Depth checkEnforce a token or spend budget per agent run2.5 points · Demonstrate "Enforce a token or spend budget per agent run" with test data. Show the result, an exception case, and the required plan. Ask for a demonstration Demonstrate "Enforce a token or spend budget per agent run" with test data. Show the result, an exception case, and the required plan. A scenario to run with the supplier; not a completed product test.	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source	Not verified0–2.5 / 2.5 pts Evidence This specific check has not been verified. Missing evidence does not mean the capability is absent. Ask for a demonstration in the proposed plan. Not verified · Oct 21, 2026Source

WHAT WE MEASURE

The criteria behind the score.

The established 20 checks contribute 80 points. Eight new operational checks contribute 20 points. AI checks are included in their own group and are excluded when AI priority is set to zero. These assess documented capability scope, not measured performance. Adjust group priorities in the comparison tool.

See the complete calculation →

Business knowledge & context 20%
Ground responses in approved information and relevant business records.
Tools & workflow actions 20%
Connect to systems and execute bounded, useful operations.
Monitoring & handoff 16%
Trace work, manage failure and hand control to a person.
Permissions & operational controls 12%
Restrict access, audit changes and govern deployment.
AI configuration & evaluation 22%
Configure or evaluate model behavior and understand its limits.
Operational detail 10%
Specific workflow, integration and control checks beyond the established rubric.

THE PRACTICAL BUYER’S GUIDE

How to choose ai agents & agent platforms.

An agent platform is not just a chat interface. It connects instructions, business knowledge, tools and an execution policy. A customer-service agent may be packaged around a help desk and its handoff rules. A general business agent builder may require your team to design the workflow, connect systems and define operational controls. Start with one bounded job and a measurable completion condition. State what the agent may read, what it may change, what requires approval and when it must stop. Compare platforms at that level before comparing models or promotional autonomy claims. The cost of a working service includes the platform, model consumption, integrations, evaluation and the people who maintain it.

Define the contract for the job

Write an input, expected outcome and stop condition for each task. For example, an agent might identify an order, retrieve its status and explain the shipping policy. Issuing a refund is a separate permission and a separate test. Include the identity checks and the source of truth for each decision. Record which systems the agent can access and whether it uses a service account or the requesting user’s permissions. A platform that can technically call an API is not automatically configured to make the right business decision. Use a small, explicit action set before allowing broad tool selection.

Evaluate the failures that matter

Create a repeatable test set with ordinary requests, missing information, conflicting documents, unavailable systems and attempts to obtain unauthorized data. Assess correctness, completeness, escalation quality and action safety. Measure cost and latency as well, using the same workload for each finalist. Review traces with the business owner, not only the builder. Ask whether you can retain a versioned test set and rerun it after changing a prompt, model, integration or knowledge source. Vendor benchmark results can inform questions, but they do not substitute for evidence on your workflows. Keep the production rollout reversible.

Compare a configured service with a configured service

One vendor may bundle a ready-made customer agent while another sells a flexible builder. Include the work needed to reach equivalent functionality: knowledge preparation, connector setup, authentication, approval logic, monitoring and support. Ask who owns a failure outside business hours and what happens when a quota is exhausted. Confirm whether model costs are included, passed through or represented by credits. Track billable actions, conversations or outcomes using a sample month, including retries and longer requests. The most useful commercial comparison is the cost of the defined job at acceptable quality, not a generic price per token or agent.

What changes the total cost?

Platform subscription or minimum commitment, AI credits, model consumption and billable actions or outcomes.
Connector development, enterprise identity, permissions, observability and evaluation work.
Human review, knowledge maintenance, escalation coverage and ongoing changes to external systems.

Where purchases go wrong

Equating a text-writing assistant with an agent authorized to execute business actions.
Launching without a repeatable failure test set, rollback path and named operational owner.
Treating a vendor’s usage unit as equivalent to a competitor’s unit without a worked billing example.

Bring these questions to the demo

Show the exact tools and records this agent can access, and a request it must refuse or escalate.
What is logged for each action, and can we inspect the source, parameters, result and cost?
What happens when a connector fails, information conflicts or the usage budget is exhausted?
Can we test a changed configuration before publication and restore the previous working version?

Questions buyers ask

Is an AI agent platform the same as workflow automation?

There is overlap. Deterministic workflows follow defined rules, while agent behavior can select actions from model-generated reasoning or context. The useful distinction is the actual execution path and control model.

Does a high feature-fit score prove an agent is reliable?

No. It indicates documented capabilities under this rubric. Reliability must be evaluated on representative tasks and monitored after deployment.

Should we start with a general builder or a packaged service agent?

Use the workflow and ownership model. A packaged agent can reduce setup for its intended domain; a builder offers flexibility but can require more design, integration and operations work.

AI agents & agent platforms

Compare the options.

Microsoft Copilot Studio

Intercom

Botpress

Relevance AI

LiveChat / Text

Freshdesk

SuiteDash

Zendesk

Smartsupp