Best enterprise AI agents in 2026: governance, pricing and platforms

Published by 3Gen Internet Corporation, maker of OpenAssistant, Skej and AGNT. Comparisons use public documentation, without hands-on testing.

Microsoft Copilot Studio with Microsoft 365 Copilot agents leads this list for the enterprise job, ahead of Salesforce Agentforce and Google Gemini Enterprise. The large platforms lead because they publish the most about admin control, audit logs, security statements and price. Where a vendor publishes no price, we write “not stated”. If what you want is an assistant for each person that acts through email, text, calls and Slack, OpenAssistant is our separate pick, covered in its own section below.

Options at a glance

For enterprise buyers we emphasize what a security or IT reviewer asks first: governance, admin and audit controls, security statements, and whether the price is published. We consider documented capabilities, evidence limitations, price, integrations, and privacy and administrative controls. A platform moved up when its own pages state audit logging, admin control over which agents run, a data-training position and a price. It moved down when a page was silent on those points. Silence is recorded as “not stated”, not as “no”.

The OpenAssistant row is a pick for teams that want an assistant per person acting through email, text, calls and Slack. It is not ranked against the platforms above because it does a different job. The big platforms win on deep system integration.

1. Microsoft Copilot Studio and Microsoft 365 Copilot agents

Microsoft describes Copilot Studio as a platform for creating and managing custom agents “with no code” (Microsoft Copilot Studio). Our reason for ranking it first is the governance detail Microsoft states in public.

What it does. Copilot Studio builds autonomous agents and publishes them internally, to social channels or to websites. It lists 1500+ pre-built data connectors and MCP servers (Microsoft Copilot Studio). Microsoft 365 agents can be declarative or custom engine agents that take actions like sending emails or updating records (Microsoft Learn). Pre-built agents include Researcher, Analyst and Facilitator (Microsoft Copilot pricing).

Governance and audit. Microsoft says to “capture every agent interaction and security-relevant event in detailed audit logs”, which flow into Purview Compliance Audit, and lists Entra ID, Defender and conditional access policies. It also lists version control, release pipelines and approval gates, and budgets and usage limits for user groups (Microsoft Copilot Studio). Microsoft Agent 365 became generally available on 1 May 2026 as a central control plane to observe, govern and secure agents, licensed per user; the price is not on that page (Microsoft Learn). Admins can set data policies and customer-managed encryption keys (Microsoft Learn). For Microsoft 365 agents, admins “have full control to select which agents are allowed” (Microsoft Learn).

Security statements. Microsoft says prompts, responses and Graph data are not used to train foundation LLMs, that Copilot only surfaces data a user can already view, and lists GDPR, ISO 27001, HIPAA and ISO 42001 (Microsoft Learn). The Copilot Studio governance page refers to “multiple standards certifications” without listing them (Microsoft Learn).

Price. Standalone Copilot Studio is sold in capacity packs of 25,000 Copilot Credits at $200 per pack per month, tenant-wide. Pay-as-you-go and a pre-purchase plan with up to 20% savings also exist, and an Azure subscription is required to use agents (Copilot Studio pricing). At the pack rate, one credit is $0.008 ($200 divided by 25,000). Microsoft’s rate table charges 1 credit for a classic answer, 2 for a generative answer and 5 for an agent action (Microsoft Learn), so an agent action works out to $0.04 at the pack rate (our arithmetic). Unused credits do not carry over month to month (Microsoft Learn).

For people, the Copilot Studio pricing page lists a Microsoft 365 Copilot add-on at $30 per user per month paid yearly, and says internal agents come with it while outside channels need a standalone plan (Copilot Studio pricing). The business pricing page lists a different Microsoft 365 Copilot Business add-on at $18 per user per month paid yearly, or $25.20 monthly (Microsoft Copilot pricing).

Watch out for. Microsoft’s own pages show two Copilot add-on prices for different SKUs, plus credit meters, so budget with a Microsoft rep. Microsoft 365 Copilot is now named Microsoft Copilot in the privacy documentation. The Microsoft 365 Copilot business and agents marketing pages were blocked to plain requests, so this entry rests on pricing pages and Microsoft Learn.

2. Salesforce Agentforce

Salesforce calls Agentforce “the AI agent platform that delivers 24/7 autonomous support at enterprise scale” (Salesforce Agentforce). It ranks second because its price list is the most detailed of any large vendor, while its security page is thinner than Microsoft’s.

What it does. Agent jobs listed include customer service, contact center, field service, employee service, sales and IT service. Agents can escalate to human agents, and a low-code builder uses Flows, Prompts, Apex and MuleSoft APIs. Salesforce also lists MCP server support from AgentExchange partners (Salesforce Agentforce).

Governance and audit. Salesforce lists observability to “monitor, analyze, and optimize agent performance in near real time” (Salesforce Agentforce). Its sales-agent page says agents act under the same permissions as the employee using them, with a suggestive or autonomous mode for pipeline updates (Salesforce sales agents). An audit-trail statement did not appear on the pages we read.

Security statements. The Trust Layer lists data masking, TLS in-flight encryption and zero data retention with LLMs, and says prompts and responses are never stored or used to train the third-party models (Salesforce Trust Layer). Certifications are not stated on the pages opened.

Price. Salesforce lists three routes: Flex Credits, conversations or per-user licensing (Agentforce pricing). The page lists these figures in USD:

• Flex Credits: $500 per 100,000 credits. An Agentforce action is 20 credits and a Voice action is 30 credits, so one action is $0.10 and one Voice action is $0.15 (our arithmetic: 20 credits at $0.005 each, 30 credits at $0.005 each).
• Conversations: $2 per conversation. Flex Credits and Conversations cannot be mixed in the same org. One conversation costs the same as 20 standard actions ($0.10 times 20 is $2.00), so the cheaper route depends on how many actions a conversation takes.
• Agentforce add-ons for Sales, Service and Field Service: $125 per user per month for unmetered employee usage. Industries add-ons: $150.
• Agentforce User License: $5 per user per month, which requires Flex Credits. Agentforce Coworker User License: $50 per user per month.
• Max Edition: from $550 per user per month, including the add-on and 2.75M Flex Credits per org per year.
• Salesforce Foundations is listed at $0 with Agentforce Builder and Prompt Builder.

Unused Flex Credits do not roll over to later terms. The page says it is “subject to change” and tells buyers to contact sales for detailed pricing (Agentforce pricing).

Watch out for. The $125 add-on covers employee usage; customer-facing agents are billed by credits or conversations. The pricing figures only appeared when the page was read in a browser.

3. Google Gemini Enterprise

Google describes it as “One universal agent for the way you work” across email, docs, code, data and business apps (Gemini Enterprise). It ranks third on governance and security statements, but Google states no per-seat price.

What it does. Long jobs run on a secure Google Cloud machine, and Google lists over 100 integrations, including Google Workspace, Microsoft 365, Slack, Confluence, Salesforce and ServiceNow (Gemini Enterprise). A Deep Research agent runs “hundreds of searches” across the web and access-controlled enterprise data (Gemini Enterprise agents).

Governance and audit. Agents run under each person’s permissions with an audit trail, and a budget cap per project makes the agent pause instead of overspending (Gemini Enterprise). Admins get centralized visibility and control over agents from Google, partners or your own teams, including control over partner agents in the Agent Gallery (Gemini Enterprise agents).

Security statements. Customer data is never used to train Google models or models for any other customer, and Google never sells customer data (Gemini Enterprise). No certification list appeared on the pages opened.

Price. Editions are Business (1 to 500 users), Standard, Plus, Pay-as-you-go and Frontline (Gemini Enterprise editions). Per-seat prices: not stated. Google offers the Plus edition at no cost for 30 days without sales help (Gemini Enterprise).

Watch out for. Google’s page says “Not every feature shown on this page is available right now”, and lists background multi-step tasks, mobile and desktop, and third-party models as early access. The page title now reads “Gemini for Business” while the URL and docs say Gemini Enterprise.

4. Glean agents

Glean says its agents “reason through tasks, plan next steps, and take action using enterprise context” (Glean agents). It has the longest certification list of the group but no public price.

What it does. Glean lists an agent builder with event triggers and routing between agents, observability into adoption, error rates and ROI, and 275+ app connectors with permissions-enforced search (Glean agents). The agents page says 275+ connectors, while Glean’s navigation says more than 250.

Governance and audit. Admins can control how broadly agents are shared, and Glean says “Agents check permissions on every request” (Glean agent governance). The security page describes an end-to-end audit trail with SIEM streaming. It also marks some agent governance controls as beta or “coming soon” (Glean security).

Security statements. Listed certifications include SOC 2 Type II, ISO 27001, ISO 42001, HIPAA, GDPR and TX-RAMP Level 2 (Glean security). A data-training statement: not stated on the pages opened.

Price. Not stated. The pricing path redirects to the homepage, and the call to action is “Get a demo” (Glean).

Watch out for. Because no price or trial is public, you cannot size a budget before a sales call. Confirm which governance controls are generally available and which are beta.

5. IBM watsonx Orchestrate

IBM says you can “build and manage all your AI agents in one control plane” (IBM watsonx Orchestrate). Besides Microsoft and Salesforce, it is the only large platform here that lists monthly dollar prices.

What it does. IBM says it can scan for agents in other environments, such as Amazon Bedrock, and bring them under one control plane. It runs on multiple clouds or on premises, and connects through APIs and open standards such as A2A and MCP. Use cases include customer service, HR, procurement, finance, sales and IT (IBM watsonx Orchestrate).

Governance and audit. IBM lists consistent access, policy and oversight, controls for risks such as sensitive or personally identifiable information, and visibility into token usage and LLM calls (IBM watsonx Orchestrate). An audit-log statement: not stated on the pages opened.

Security statements. Premium adds data isolation and a HIPAA-ready option (IBM pricing). A certification list and a data-training statement: not stated on the pages opened.

Price. Essentials starts at $530 per month, with up to 4,000 monthly active users and up to 200,000 messages a month. Standard starts at $6,360 per month, with up to 40,000 monthly active users and up to 2,000,000 messages. Premium is contact us. There is a 30-day free trial (IBM pricing). Standard’s starting price is 12 times Essentials ($6,360 divided by $530), for 10 times the user and message limits.

Watch out for. The prices are “starting at” figures, and the security pages are thin on certifications, so ask for the compliance pack.

6. ServiceNow AI Agents

ServiceNow says its agents “act autonomously to get work done” (ServiceNow AI Agents). It has the most specific agent-control features on this list, and the least public information on price and security.

What it does. You build agents in natural language in AI Agent Studio or deploy prebuilt agents. Agent types include voice, web and desktop, and A2A and MCP connect third-party agents (ServiceNow AI Agents).

Governance and audit. Agents are governed by the ServiceNow AI Control Tower (the page is now titled Super Intelligence Control Tower), which lets you discover, secure, govern, observe and measure AI. ServiceNow says “Flip a kill switch to stop an AI agent.” Its AI Passport gives each agent a signed identity and permitted scope, enforced at runtime, and content packs cover the EU AI Act and NIST AI RMF (ServiceNow Control Tower).

Security statements. The trust page mentions annual third-party audits, regulated-market offerings and a white paper on AI security and data handling controls (ServiceNow trust). A data-training statement and a certification list: not stated on the pages opened.

Price. Not stated. The Control Tower page says “Contact Us for Pricing” (ServiceNow Control Tower).

Watch out for. No price, trial or training statement is public. ServiceNow’s pages were read in a browser because plain requests could not connect.

7. Dust

Dust describes itself as “Multiplayer AI for human-agent collaboration” (Dust). We include it as a smaller option for teams, not as a rival to the platforms above.

What it does. Dust lists connectors for Slack, Notion, GitHub and Drive, MCP servers, and scheduled and event-driven agents (Dust pricing).

Governance and security. Dust lists user, builder and admin roles, GDPR and SOC 2 Type II, AES-256 at rest, TLS in transit, hosting in the EU or US, and says data is never used to train models (Dust security). SCIM and audit logs are Enterprise features (Dust pricing).

Price. The Business plan is for teams up to 100 people. Prices are in euros, excluding VAT: a Free seat with 500 credits for life, Pro at 24 euros per seat per month billed yearly (30 monthly) with 8,000 credits a month, and Max at 120 euros yearly (150 monthly) with 40,000 credits a month. Enterprise is talk to sales (Dust pricing). The yearly price is 20% below monthly (24 divided by 30 is 0.8).

Watch out for. Dust’s connector count differs by page: 70+ on the home page and 20+ data sources on the pricing page. Credits reset monthly with no rollover.

OpenAssistant: an assistant per person, with Team and Enterprise admin controls

OpenAssistant is our recommendation for teams that want an assistant per person acting through email, text, calls and Slack. Its homepage says “Each assistant has their own computer, email address, and phone number.” (OpenAssistant). The big platforms win on deep system integration with the business systems they sit on. OpenAssistant is a general work assistant, not a specialist coding agent, SDR or support-desk product.

Where it works. The marketing page lists calls, iMessage, SMS, WhatsApp, email and the web app (OpenAssistant). The help center lists the dashboard, email, Slack, SMS and WhatsApp as request channels, and does not mention iMessage or calls (OpenAssistant help). Slack users DM @OpenAssistant or add it to channels, and it only responds when invoked (OpenAssistant Slack help).

Admin, SSO and audit. The Team plan adds an admin dashboard, SSO and SCIM, role-based access, shared billing, pooled team credits and “Granular IT controls”. Enterprise adds multi-team organizations, usage analytics, audit logs and data retention, custom integrations, private deployments, and dedicated and priority support with an SLA (OpenAssistant pricing). Admins can switch a connected service off for the whole organization, allow-list or block skills, and enable or disable MCP tool connections (OpenAssistant trust).

Approvals. Autonomy is set per action type. The levels are Manual, Confirmation Window and Autonomous, and a Confirmation Window is not the same as Manual approval (OpenAssistant help). An admin can lock those settings for everyone (OpenAssistant).

Security statements. The trust page states SOC 2 Type II audited annually by an independent CPA firm, and says “We never train on your data.” (OpenAssistant trust).

Price. The Starter plan is free: 1,000 credits at signup, then 100 bonus credits a day once used up, up to 1,000 a month. Team is $120 per user per month billed monthly or $100 billed annually at the 5,000-credit tier, and higher credit tiers cost more. Enterprise is custom through Contact Sales (OpenAssistant pricing; per-user pricing confirmed by 3Gen).

Watch out for. It does not join group texts or group WhatsApp chats, not every country is supported for texting (connecting a phone number), and it does not work in Slack Connect channels with outside companies (OpenAssistant Slack help). Plan credits reset each billing cycle and do not roll over (OpenAssistant pricing). If your core job is a customer-facing agent or a system-of-record workflow, a platform above fits better.

What to ask each vendor

• Can we export agent audit logs to our SIEM, and how long are they kept?
• Which agent actions can run without human approval, and can an admin lock that setting?
• How do agents resist prompt injection, where an email or web page tries to give the agent new instructions? What stops an agent from acting outside the user’s permissions?
• Is there a kill switch or account-wide stop for a misbehaving agent?
• What is the full price for our volume, including credits, conversations or messages, and what happens at the cap?
• Are customer prompts and data excluded from model training, in writing, including at the model provider?
• Which certifications apply to the agent product itself, not only the parent company?
• Which governance features are generally available today, and which are beta or early access?

What to keep for humans

• Approving payments, contracts, hiring decisions and anything sent to customers or regulators.
• Granting or widening agent permissions, and deciding which connectors an agent may use.
• Reviewing audit logs and agent errors on a schedule.
• Handling security incidents, including pulling an agent from service.
• Legal and compliance judgments. This is not legal advice.

Related reading

• Best AI customer support agents in 2026: prices per resolution and per seat: Seven support agents compared, with per-resolution and per-seat prices explained.
• Best AI agent builders and platforms in 2026: no-code to code: Ten agent builders, including n8n, Copilot Studio, Make, Zapier and CrewAI, with credits, tasks and executions explained.
• AI agent pricing in 2026: credits, seats and per-resolution fees explained: Credit plans, tasks, per-resolution fees, subscriptions and enterprise fees, with published prices and a worked example.
• Are AI agents safe? Prompt injection, permissions and approval settings: Real risks from prompt injection to excess access, the controls to look for and a 10-question checklist.
• How to choose an AI agent: 10 questions to ask before you buy: Ten questions on job fit, channels, approvals, price, data and exit, with a job-to-product table.

Frequently asked questions

What are the best enterprise AI agents in 2026?

On public governance, audit and price information, we rank Microsoft Copilot Studio with Microsoft 365 Copilot agents first, then Salesforce Agentforce, Google Gemini Enterprise, Glean, IBM watsonx Orchestrate and ServiceNow AI Agents. It is an editorial judgment from vendor pages, not a hands-on result.

How much do enterprise AI agents cost?

It varies by meter. Microsoft lists $200 per month for 25,000 credits, Salesforce lists $500 per 100,000 Flex Credits or $2 per conversation, and IBM lists plans from $530 and $6,360 per month. Google, Glean and ServiceNow state no price.

Which enterprise agent platforms publish no price?

Per-seat prices for Gemini Enterprise, Glean and ServiceNow AI Agents are not stated on the pages we read. IBM Premium, Dust Enterprise and OpenAssistant Enterprise are also quote-based.

Is OpenAssistant an enterprise platform?

It has Team plan admin controls (admin dashboard, SSO, SCIM, role-based access) and Enterprise audit logs, but it is a per-person assistant, not a system-of-record agent platform. Pick a large platform for deep integration with your core systems.

Try OpenAssistant

If you want an assistant per person that works through email, text, calls and Slack, try OpenAssistant. You can see plans and Team pricing on the OpenAssistant pricing page.