# Enterprise AI Platforms Compared: 10 Options by Operating Model > Published 2026-11-10 > Source: https://viktor.com/blog/enterprise-ai-platforms-compared The enterprise AI platforms worth a place on a late-2026 shortlist are Viktor, Microsoft Copilot Studio, [Salesforce Agentforce](https://viktor.com/blog/viktor-vs-salesforce-agentforce), IBM watsonx Orchestrate, UiPath, Automation Anywhere, Moveworks, C3 AI, Google’s Gemini Enterprise Agent Platform (formerly Vertex AI) and Amazon Bedrock. Their operating models differ more than their feature lists. Some sell model infrastructure that engineers build on. Others take a job and carry it out across the tools a company already runs. **The ten platforms in one pass: best use, governance and integration strength, main limit and pricing (checked November 2026).** Viktor is best for finished cross-tool work handed back in Slack or Teams, with 9,800+ tools and approval cards before sensitive actions; he has no role-based access control or SCIM yet, and the Team plan is $100 a month for 40,000 shared credits. Microsoft Copilot Studio is best for Microsoft 365 estates, with Entra ID, Purview audit trails and 1,500+ connectors; multistage approvals are still in preview, and credits cost $200 per 25,000 a month. Salesforce Agentforce is best for Salesforce CRM teams, with permission sets and the Einstein Trust Layer; outside systems need MuleSoft, and pricing is $500 per 100,000 Flex Credits or $2 per conversation. IBM watsonx Orchestrate is best for governing AI agents from many vendors, with audit logs, a kill switch and 80+ pre-integrated applications; it has no custom roles and starts at $578 a month. UiPath is best for existing RPA estates, with guardrails, human review and self-hosting; some features are cloud-only, and plans run from a free Community tier to quoted Enterprise. Automation Anywhere is best for back-office processes across SAP, Salesforce, ServiceNow and Workday, with SOC 2 Type 2 and ISO 42001; it publishes no price list. Moveworks is best for IT and HR service desks, with one-click approvals in chat; its MCP Workspace has no confirmation step, and pricing is by quote. C3 AI is best for industrial and government applications, with FedRAMP Moderate and air-gapped deployment; it needs dedicated data teams and starts with a paid pilot. Google’s Gemini Enterprise Agent Platform and Amazon Bedrock are best for engineers building their own AI products, with spend caps and tool-call policies; every workflow is the customer’s build, billed by usage. A Gartner survey of 1,303 respondents, run from January to April 2026, [found only 22% of organizations have scaled AI across multiple business units](https://www.gartner.com/en/newsroom/press-releases/gartner-survey-finds-only-22-percent-of-organizations-have-successfully-scaled-ai-across-multiple-business-units), and that stalled scale is the pressure to consolidate. In June 2025, Gartner [predicted](https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027) that over 40% of agentic AI projects will be canceled by the end of 2027, “due to escalating costs, unclear business value or inadequate risk controls.” Sprawl compounds the problem. FICO’s State of Responsible AI 2026, a vendor-run survey of more than 1,000 executives, found that only 8.2% of organizations have adopted a single shared AI deployment platform, [according to CIO Dive](https://www.ciodive.com/news/single-ai-deployment-platforms-rare-app-sprawl-grows/832443). Viktor is the [AI employee](https://viktor.com/ai-employee) who lives in Slack and Microsoft Teams, connects to 9,800+ tools, and does the work. He sits in the workflow-level execution layer, a different operating model from the RPA and model infrastructure vendors he shares a shortlist with. For a company that wants finished work handed back in chat, that layer is the right place to start. ## Quick reference: enterprise AI platforms compared Viktor and the nine other platforms offer different ways to get work done, from model infrastructure to workflow-level execution. Viktor comes first, and the other nine are grouped by operating model (pricing checked November 2026): ```json:object { "_key": "k42", "_type": "table", "rows": [ { "_key": "k31", "_type": "tableRow", "cells": [ "Product", "Operating model", "Best for", "Pricing model", "Key differentiator" ] }, { "_key": "k32", "_type": "tableRow", "cells": [ "Viktor", "Workflow-level execution (AI employee)", "Work handed off in plain language across the whole company in Slack, Microsoft Teams, the web app or Discord", "Usage credits shared across the workspace, free to start (Viktor pricing)", "Takes the job end to end across 9,800+ connected tools and returns finished files, pull requests and updated records" ] }, { "_key": "k33", "_type": "tableRow", "cells": [ "Gemini Enterprise Agent Platform (formerly Vertex AI)", "Model infrastructure", "Engineering teams building on Google Cloud", "Usage-based, billed per model, tool, storage and compute (Google Cloud pricing)", "200+ models through Model Garden" ] }, { "_key": "k34", "_type": "tableRow", "cells": [ "Amazon Bedrock", "Model infrastructure", "Engineering teams building on AWS", "Per-token by model across Standard, Flex, Priority and Reserved tiers (Bedrock pricing)", "100+ foundation models plus the AgentCore runtime" ] }, { "_key": "k35", "_type": "tableRow", "cells": [ "C3 AI", "Model infrastructure and data platform", "Industrial, energy and government applications on one data model", "C3 AI filings describe a paid pilot, then usage-based runtime fees", "Unified ontology graph of the enterprise" ] }, { "_key": "k36", "_type": "tableRow", "cells": [ "UiPath", "Automation and orchestration", "Organizations with an existing RPA estate", "Free Community plan up to sales-quoted Enterprise (UiPath pricing)", "Maestro BPMN processes that coordinate robots, AI and people" ] }, { "_key": "k37", "_type": "tableRow", "cells": [ "Automation Anywhere", "Automation and orchestration", "Process automation across ERP, CRM and legacy applications", "Free Cloud Community Edition; paid pricing not published on the retrieved official pages (Automation Anywhere products)", "Process Reasoning Engine" ] }, { "_key": "k38", "_type": "tableRow", "cells": [ "IBM watsonx Orchestrate", "Automation and orchestration", "Governing AI tools from many vendors in one control plane", "Monthly Essentials and Standard tiers, Premium by quote, 30-day trial (IBM pricing)", "Agent Identity GA with audit logs and a kill switch" ] }, { "_key": "k39", "_type": "tableRow", "cells": [ "Microsoft Copilot Studio", "Ecosystem builder", "Microsoft 365 and Power Platform estates", "Copilot Credit capacity packs, prepurchase or pay-as-you-go (Microsoft pricing)", "More than 1,500 prebuilt data connectors" ] }, { "_key": "k40", "_type": "tableRow", "cells": [ "Salesforce Agentforce", "Ecosystem builder", "Salesforce CRM and Data 360 customers", "Flex Credits, per-conversation or per-user licenses (Agentforce pricing)", "Einstein Trust Layer with zero data retention at third-party LLMs" ] }, { "_key": "k41", "_type": "tableRow", "cells": [ "Moveworks", "Employee support", "IT and HR service desks", "Custom quote (Moveworks pricing)", "One-click approvals in Slack or Teams; part of ServiceNow since December 15, 2025" ] } ] } ``` A shortlist built from this table should hold at most one product per operating model, unless two models answer clearly different jobs. ## The four operating models in the enterprise AI platform market An enterprise AI platform applies AI models to a company’s own data and systems under central controls for access, approval, audit and spend. The models below split on one question: who does the work once the platform is bought, the company’s engineers, a process designer, a vendor ecosystem, or the platform itself. ### Model infrastructure and data platforms Google renamed Vertex AI to [Gemini Enterprise Agent Platform on April 22, 2026](https://cloud.google.com/blog/products/ai-machine-learning/introducing-gemini-enterprise-agent-platform), and says all Vertex AI services and roadmap now ship through it. Amazon made [Bedrock AgentCore generally available on October 13, 2025](https://aws.amazon.com/about-aws/whats-new/2025/10/amazon-bedrock-agentcore-available) as a runtime layer on top of Bedrock’s model catalog. C3 AI sells the [C3 Agentic AI Platform](https://www.c3.ai/c3-agentic-ai-platform), which models the enterprise as an ontology graph and ships prebuilt applications, plus [C3 Code, generally available April 8, 2026](https://c3.ai/news/c3-ai-announces-c3-code). All three are build-your-own. A team picks models, writes the orchestration, wires the data and runs it. Choose this model when an engineering team already owns AI applications on that cloud and wants control over every layer. ### Automation and orchestration platforms UiPath extended RPA into orchestration with Maestro, generally available April 30, 2025, which runs long processes as BPMN diagrams that pass steps between robots, AI and people. Maestro Case followed on June 16, 2026 for judgment-heavy work such as claims. Automation Anywhere calls its platform the Agentic Process Automation System. Its Process Reasoning Engine, announced May 13, 2025, plans work across bots, automation and people. IBM positions watsonx Orchestrate as “an open, hybrid and trusted enterprise agent management platform” in its [TechXchange 2026 announcement](https://www.ibm.com/new/announcements/ibm-watsonx-orchestrate-advances-enterprise-ai-agent-management-with-new-capabilities-at-techxchange-2026), dated October 27, 2026. The emphasis is on discovering, governing and evaluating AI built on other platforms. These platforms are strongest where processes are already mapped. They assume someone designs and maintains the logic. ### Ecosystem builders and employee support platforms Microsoft Copilot Studio started at Ignite 2023 as the successor to Power Virtual Agents and builds on Power Platform connectors, Dataverse and Entra ID. Salesforce announced general availability of Agentforce 360 on October 13, 2025, grounded in CRM data and Data 360. Moveworks is now a ServiceNow product. ServiceNow announced the deal March 10, 2025, closed it December 15, 2025, and combined it with ServiceNow’s portal in EmployeeWorks on February 26, 2026. Each is strongest inside its own stack. The test is how much of the work a company runs sits in Microsoft, Salesforce or ServiceNow. ### Workflow-level execution: the AI employee layer Viktor works as a hire who takes the job end to end. Someone describes the work in Slack, Microsoft Teams, the web app at app.viktor.com or Discord, and he plans it, works on his own cloud computer and posts the finished deliverable in the thread. The homepage line is “Not a tool. A hire.” The difference from the other three models is who builds. No workflow canvas, no deployment and no infrastructure to maintain: a company onboards Viktor rather than building him. One Viktor serves the whole team, keeps workspace memory of decisions and preferences, and follows written skills the team can read and edit. The longer explainer covers [what an AI employee is](https://viktor.com/blog/what-is-an-ai-employee). Choose this layer when the backlog is cross-tool jobs, such as the Monday ad spend recap, the month-end numbers or bug triage into Jira, rather than one mapped process. ## How they stack up The comparison runs on four dimensions: integration depth, orchestration style, deployment and residency, and pricing with total cost. No single operating model wins all four, so the winner is named per dimension: ```json:object { "_key": "k110", "_type": "table", "rows": [ { "_key": "k105", "_type": "tableRow", "cells": [ "Dimension", "Strongest operating model", "Why" ] }, { "_key": "k106", "_type": "tableRow", "cells": [ "Integration with ITSM, HRIS, ERP and CRM", "Ecosystem builders for depth inside their own system; the AI employee layer for breadth", "Native objects and permissions on one side, 9,800+ tools on the other" ] }, { "_key": "k107", "_type": "tableRow", "cells": [ "Orchestration and cross-workflow execution", "Automation platforms for mapped, long-running processes; the AI employee layer for open-ended jobs that end in a deliverable", "BPMN and case models versus plain-language hand-off" ] }, { "_key": "k108", "_type": "tableRow", "cells": [ "Deployment and data residency", "Automation platforms and C3 AI", "Self-hosted, air-gapped and customer-cloud options" ] }, { "_key": "k109", "_type": "tableRow", "cells": [ "Pricing and total cost", "Usage-based platforms for low entry cost; quote-based suites for fixed enterprise budgets", "Entry cost versus predictability" ] } ] } ``` Read each row against the company’s own constraint: a regulated, on-premises estate settles the deployment row before anything else. ### Integration with ITSM, HRIS, ERP and CRM - Moveworks publishes plugin counts per system, including Workday (121), ServiceNow (103) and Salesforce (80), which reflects its IT and HR focus. - Microsoft cites more than 1,500 prebuilt connectors for Copilot Studio, including SAP, Workday and Salesforce. - Salesforce reaches outside systems through MuleSoft, whose Anypoint platform has [over 1,500 prebuilt connectors](https://www.mulesoft.com/platform/enterprise-integration). - UiPath’s Integration Service covers Salesforce, SAP, ServiceNow and Workday but publishes no total. - IBM says watsonx Orchestrate is [pre-integrated with 80+ enterprise applications](https://www.ibm.com/new/announcements/productivity-revolution-with-ai-agents-that-work-across-stack). - Automation Anywhere lists native connectors for SAP, Salesforce, ServiceNow and Workday. Viktor connects to [9,800+ tools](https://viktor.com/integrations), most through one-click OAuth, across CRM (HubSpot, Salesforce, Attio), ads, finance (QuickBooks, Xero, Stripe), project tracking (Jira, Linear) and engineering (GitHub). A tool with an API or MCP server can be added as a custom integration, and Viktor can usually build that connection himself. These counts measure different things (connectors, plugins, applications, tools), so they are not like-for-like. The useful test is a list of the company’s ten most-used systems checked against each vendor’s directory. ### Orchestration and cross-workflow execution RPA moves data along fixed paths. Copilots and chat tools hand a draft to one person. Orchestration platforms sit between: UiPath’s Maestro pauses a process at a user task until an assignee completes it, and Automation Anywhere’s Process Composer approval task routes a request to approvers before the next step. The output of those platforms is a process outcome inside an enterprise system, such as a record moved or a case closed. Salesforce’s output is usually a customer or employee interaction inside Salesforce. Viktor’s output is the finished deliverable: an Excel file with formulas, a PowerPoint deck, a PDF, an updated CRM record, or a pull request opened on a branch with context for review. He also keeps docs current from live data at the same link. A UiPath robot that moves records overnight and an AI employee who analyzes them and posts the summary answer different jobs. Map each candidate workflow to whether its end state is a moved record or a finished artifact, then match the operating model. ### Deployment options and data residency UiPath has the widest range. It offers Automation Cloud, a single-tenant Dedicated option, a FedRAMP Moderate Public Sector cloud and self-hosted Automation Suite. Automation Suite 2.2510.2, shipped April 15, 2026, brought agentic capabilities on-premises on AKS, EKS and OpenShift, though UiPath notes some capabilities remain Cloud-only. C3 AI runs [C3-managed, in a customer VPC, hybrid or on-premises](https://c3.ai/api/media/file/c3-ai-operational-controls-description.pdf). IBM offers watsonx Orchestrate [as SaaS on AWS and IBM Cloud or on-premises](https://www.ibm.com/docs/en/watsonx/watson-orchestrate/base?topic=notes-regional-availability-outbound-ip-addresses). Automation Anywhere supports cloud and on-premises from one codebase. Copilot Studio deploys to Azure datacenters across many geographies and to US government clouds; no on-premises option appears in its documentation. Agentforce runs on Hyperforce public cloud. Moveworks runs on AWS, with an on-prem connector that only opens outbound connections to reach internal systems. Viktor is a managed service, US-hosted by default. A company that needs self-hosting or air-gapped operation should shortlist UiPath, IBM or C3 AI for that workload. ### Pricing models and total cost of ownership Pricing splits three ways. Pay-as-you-go platforms bill usage: Bedrock charges per token by model, with [batch inference at a 50% lower price than on-demand](https://aws.amazon.com/bedrock/pricing), and Google charges [for tools, storage and compute used](https://cloud.google.com/products/gemini-enterprise-agent-platform). Credit and capacity platforms bill pools: Microsoft sells Copilot Credit packs of 25,000 at $200.00 per pack per month alongside pay-as-you-go. Quote-based platforms such as Moveworks price by custom quote (pricing checked November 2026). Typical entry cost by category, from the list prices above: ```json:object { "_key": "k184", "_type": "table", "rows": [ { "_key": "k178", "_type": "tableRow", "cells": [ "Category", "Entry cost on list prices", "What drives the bill up" ] }, { "_key": "k179", "_type": "tableRow", "cells": [ "AI employee (Viktor)", "Free to start; $100 a month for 40,000 credits shared by the workspace", "More work handed over means more credits" ] }, { "_key": "k180", "_type": "tableRow", "cells": [ "Credit and capacity platforms (Copilot Studio, Agentforce)", "$200 a month per 25,000 Copilot Credits; $500 per 100,000 Flex Credits or $2 per conversation", "Volume of conversations and actions, plus the makers who build each agent" ] }, { "_key": "k181", "_type": "tableRow", "cells": [ "Automation suites (watsonx Orchestrate, UiPath, Automation Anywhere)", "$578 to $6,360 a month on IBM’s listed tiers; UiPath Basic from $25 a month; enterprise tiers by quote", "Robots, platform units and process design work" ] }, { "_key": "k182", "_type": "tableRow", "cells": [ "Model infrastructure (Bedrock, Gemini Enterprise Agent Platform)", "No minimum; billed per token, tool, storage and compute", "Engineering time to build and run every workflow" ] }, { "_key": "k183", "_type": "tableRow", "cells": [ "Quote-based platforms (Moveworks, C3 AI)", "Custom quote; C3 AI starts with a paid pilot", "Scope, seats and implementation services" ] } ] } ``` Trials differ. IBM offers a [30-day free trial](https://www.ibm.com/products/watsonx-orchestrate/pricing), UiPath a 60-day Standard trial and a free Community plan, and Google $300 in free credits for new customers. A 2024 filing describes C3 AI’s pilot as a [paid phase of up to six months](https://ir.c3.ai/static-files/5fd6a2a7-9f9d-4e58-9b12-a95ec2c89393). [Total cost](https://viktor.com/blog/budget-for-ai-costs-forecasting-attribution) has four parts: - **Licensing:** Seats, editions or platform units bought up front. - **Usage:** Credits, tokens, tasks or conversations consumed. Gartner [predicts inference costs per agentic workflow will rise more than fivefold through 2028](https://www.gartner.com/en/newsroom/press-releases/2026-08-17-gartner-predicts-ai-inference-costs-per-agentic-workflow-will-increase-more-than-fivefold-through-2028). - **Integration build:** Connectors, data cleanup and process mapping before the first workflow runs. - **Run costs:** Monitoring, model upgrades and the people who maintain flows. Viktor runs on usage credits shared across the whole workspace, with no per-seat fee. He’s [free to start with up to $100 in credits and no credit card, and the Team plan is $100 a month for 40,000 credits for the whole workspace, not per seat](https://viktor.com/pricing), with unused Team credits rolling over. Customers don’t need to deploy Viktor or maintain his infrastructure. Budget the four parts separately for each finalist, using the company’s own volumes, before comparing totals. ## What to check before deploying an AI platform company-wide Security reviewers assess five controls before approving an AI platform: permissions and scoping, approval and human-in-the-loop controls, audit trail and spend controls, data residency and training on customer data, and certifications. Vendor vetting as a process is covered in [how to run security due diligence before hiring an AI employee](https://viktor.com/blog/how-to-run-security-due-diligence-on-an-ai-employee), and the access questions in the [security checklist for AI tool access](https://viktor.com/blog/security-checklist-for-ai-tool-access). The point of all five is the same failure mode. Gartner [predicts that by 2029](https://www.gartner.com/en/articles/agentic-ai-infrastructure-governance) at least 70% of organizations with production agentic AI in I and O will have a material service, security or cost incident linked in part to insufficient runtime controls. ### Permissions and scoping Copilot Studio manages access through Microsoft Entra ID groups and Dataverse security roles, and admins block unneeded connectors with Power Platform data policies. Salesforce assigns Agentforce access through permission set groups. UiPath uses Automation Ops policies to decide which connectors users can create connections for and offers SCIM user sync on its Enterprise plan. IBM documents [user and builder roles](https://www.ibm.com/docs/en/watsonx/watson-orchestrate/base?topic=users-roles-watsonx-orchestrate) and states the platform [does not currently support custom roles](https://www.ibm.com/docs/en/watsonx/watson-orchestrate/base?topic=users-managing-cloud). AWS lets admins define, through [AgentCore Policy](https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/what-is-bedrock-agentcore.html), which tools may be called and under what conditions, intercepting every tool call before it runs. Viktor works only with [the access a company grants](https://viktor.com/security). Connections can be shared with the workspace or kept private to the person who connected them and the people or groups they add. Admins decide who can connect or edit integrations and which tools need approval or are forbidden, and in Slack also who can use Viktor and who he can DM. Slack searches run with the asker’s own visibility. Viktor does not offer role-based access control or SCIM provisioning today. A company that requires either for every AI tool should treat that as a gating item and use admin tool controls and private connections as the scoping layer in a pilot. ### Approval and human-in-the-loop controls Viktor is an AI employee who waits for approval before he changes an account: before sensitive actions he posts approval cards in Slack or Microsoft Teams with Approve and Reject, and a rejected action never runs. [Viktor’s security page](https://viktor.com/security) describes explicit approval for money moves, code pushes and customer emails, and admin controls to require approval for any other integration tool. Joining channels, changing workspace settings, changing integration access, setting spend limits and installing skills always need approval. Competitor controls vary in maturity: - **UiPath:** Tool calls marked Require confirmation pause and let the user approve, edit inputs or reject, and guardrails can Block or Escalate. - **Microsoft Copilot Studio:** Multistage approvals are in preview, and the per-tool option to ask the end user before running is set to No by default. - **Salesforce Agentforce:** Salesforce recommends customers configure custom actions to require confirmation before sensitive actions. - **IBM watsonx Orchestrate:** Flows route tasks to a named user or role through [User Activities](https://developer.watson-orchestrate.ibm.com/agents/agent_design/agentic_workflow_decision_guide) and resume on their response. - **Moveworks:** Approvers approve or deny in chat with one click; its docs warn that MCP Workspace has no confirmation step before a tool call. In a trial, list the five most sensitive actions in the pilot workflow and confirm each one stops for approval by default, not after configuration. ### Audit trail and spend controls [Viktor Enterprise](https://viktor.com/enterprise) adds audit logs that include integration tool calls, SAML single sign-on and per-user spend caps. On every plan, the Activity view shows credit use by person and task, each scheduled task shows time saved and credit cost, and spend limits and auto top-ups run with a monthly cap. Copilot Studio sends interaction logs to Microsoft Purview and supports monthly consumption limits per agent; with pay-as-you-go billing, enforcement doesn’t apply. Salesforce offers Digital Wallet threshold alerts for Flex Credits. UiPath’s AI Trust Layer audit tab shows entries from the last 60 days, and tenant limits send notifications at 50%, 75% and 90%. Google lets a project budget for Vertex AI act as a [spend cap that stops paid usage](https://docs.cloud.google.com/gemini/enterprise/docs/manage-costs-overview). Bedrock logs API calls to [CloudTrail](https://docs.aws.amazon.com/bedrock/latest/userguide/logging-using-cloudtrail.html). IBM shipped [Agent Identity with audit logs and a kill switch](https://www.ibm.com/new/announcements/agent-identity-in-ibm-watsonx-orchestrate-is-now-generally-available) on November 3, 2026. Require a hard stop, not only an alert, for any platform billed on usage, and confirm which plan includes it. ### Data residency and training on customer data Customer data is never used to train models, by Viktor or by his model providers. API keys and tokens are added at run time by a gateway, so the model never reads credentials, and each workspace is isolated. Viktor is US-hosted by default. Major vendors make similar no-training commitments. Microsoft states that Copilot prompts and responses aren’t used to train foundation LLMs. Salesforce’s Trust Layer applies zero data retention with third-party LLMs, though pattern-based and field-based data masking is disabled for agents. Google [won’t use data to train or fine-tune models without permission](https://cloud.google.com/vertex-ai/generative-ai/docs/data-governance), and Bedrock [doesn’t use prompts and completions to train any models](https://docs.aws.amazon.com/nova/latest/userguide/responsible-use.html). UiPath states zero third-party training, Automation Anywhere says no customer data is used to train the models, and Moveworks says no customer data is used to train global generative models. C3 AI’s retrieved pages make no explicit no-training statement. Retention is the follow-up question. Bedrock retains inputs and outputs for certain models [for up to 30 days to detect abuse](https://docs.aws.amazon.com/ia_jp/AmazonCloudWatch/latest/monitoring/omni-opting-out-of-using-your-data-for-service-improvement.html), and OpenAI may retain API inputs and outputs for up to 30 days except under zero data retention. Ask each vendor for both answers in writing: who may train on the data, and how long each model provider keeps it. ### Security certifications side by side Status below is as each vendor’s pages stated it when checked; verify current status before procurement: ```json:object { "_key": "k342", "_type": "table", "rows": [ { "_key": "k331", "_type": "tableRow", "cells": [ "Vendor", "SOC 2", "ISO 27001", "Other attestations", "Residency", "Portability" ] }, { "_key": "k332", "_type": "tableRow", "cells": [ "Viktor", "Type 1; Type 2 in progress (as of October 2026)", "In progress", "CASA Tier 3; GDPR aligned, CCPA compliant", "US-hosted by default", "Public API and MCP server" ] }, { "_key": "k333", "_type": "tableRow", "cells": [ "Microsoft Copilot Studio", "Audited for SOC; type not stated", "ISO/IEC 27001:2013 listed", "In ISO 42001 portfolio; GCC meets FedRAMP High", "Azure geographies", "Solution export with limits" ] }, { "_key": "k334", "_type": "tableRow", "cells": [ "Salesforce Agentforce", "Report for Agentforce on Hyperforce", "ISO/IEC 27001:2022", "ISO 42001:2023; FedRAMP High", "Hyperforce countries", "Metadata moves between orgs via Salesforce DX" ] }, { "_key": "k335", "_type": "tableRow", "cells": [ "IBM watsonx Orchestrate", "Type 2 reports for listed IBM Cloud services; watsonx scope not confirmed", "IBM Cloud certificates", "FedRAMP for 11 solutions incl. watsonx (April 1, 2026)", "AWS and IBM Cloud regions, on-prem", "Not documented" ] }, { "_key": "k336", "_type": "tableRow", "cells": [ "UiPath", "Type 2", "ISO/IEC 27001:2022", "ISO 42001 (October 27, 2025); FedRAMP Moderate", "Regional clouds and self-hosted", "Solution packages via CLI" ] }, { "_key": "k337", "_type": "tableRow", "cells": [ "Automation Anywhere", "Type 2", "Certified", "ISO 42001; FedRAMP not listed", "Cloud regions incl. US, EU, India, Australia", "Bot export and import API (v11.3 docs)" ] }, { "_key": "k338", "_type": "tableRow", "cells": [ "Moveworks", "Type 2", "Certified", "ISO 42001 (May 27, 2026); FedRAMP Moderate", "AWS: US, EU, GovCloud, Canada, Australia, UK, Japan", "Not documented" ] }, { "_key": "k339", "_type": "tableRow", "cells": [ "C3 AI", "SOC 2 and SOC 3 (undated vendor statement)", "Not found", "FedRAMP Moderate (December 11, 2025)", "Customer-chosen cloud region, VPC or on-prem", "Not documented" ] }, { "_key": "k340", "_type": "tableRow", "cells": [ "Google (Agent Platform)", "SOC 1, 2 and 3 for Gemini", "Listed", "ISO 42001", "Regional endpoints", "Not documented for the Agent Platform" ] }, { "_key": "k341", "_type": "tableRow", "cells": [ "Amazon Bedrock", "In scope for SOC", "In scope for ISO", "AWS ISO 42001; FedRAMP High in GovCloud (US-West)", "Per-region and cross-region inference", "Not documented" ] } ] } ``` Viktor does not offer HIPAA support or BAAs. Regulated workloads that need FedRAMP, HIPAA or on-premises operation belong on UiPath, C3 AI or a public-sector cloud. [Review Viktor’s security details, including SOC 2 Type 1, data handling and approval controls →](https://viktor.com/security) ## Each platform in detail Vendor pricing models vary by operating model and plan; pricing checked November 2026. ### Viktor Viktor is an AI employee who works in Slack, Microsoft Teams (generally available since June 18, 2026), the web app at app.viktor.com and Discord. A team hands him a job in plain language, and he returns the finished work in the thread: a Word report, an Excel model with formulas, a deck, a PDF, a doc kept current from live data, or updates in the tools themselves. For engineering leaders, the boundary is review. Viktor clones GitHub or Bitbucket repositories, writes changes on a branch and opens a pull request with context; merges stay with the team, and code pushes wait for approval. Developers can also hand work to him through a public API, an OpenAI-compatible API and an MCP server at api.viktor.com/mcp, documented in the [Viktor developer docs](https://viktor.com/docs). Strengths for a company-wide rollout: - **Finished work:** Files, pull requests, CRM updates and Viktor Spaces (custom apps your AI employee builds and runs, such as internal dashboards) instead of drafts. - **Shared hire:** One Viktor serves the whole team with workspace memory and skills, billed on shared credits with no per-seat fee. - **Approvals:** Approval cards in Slack or Microsoft Teams, with admin-set needs-approval and forbidden tools. - **Model choice:** Presets (Ultra, Smart, Balanced, Cheap) or a specific model from Anthropic, OpenAI, xAI and other labs, plus an organization’s own OpenRouter key for 300+ more. - **Credential handling:** A gateway adds keys at run time, so the model never reads them. - **Recognition:** 4.9 on G2, named in the Forbes Agentic 20 (October 2026) and Sifted AI 100 2026 (#21). Limits to plan around: - **Identity:** No role-based access control or SCIM provisioning today; SAML SSO is on Enterprise. - **Compliance:** SOC 2 Type 1 only, with Type 2 and ISO 27001 in progress; no HIPAA or BAAs; no configurable data retention. - **Deployment:** Managed only, US-hosted by default, no self-hosting. - **Teams parity:** Teams lacks Slack canvases, Lists, Slack Connect, the !stop and !mute commands, the “Always approve this tool” button, and web-app chat for Teams-only workspaces. Best for: companies that want cross-tool jobs done end to end where people already talk, with approval and spend controls, rather than a platform to build on. [Start free with up to $100 in credits, no credit card →](https://app.viktor.com/signup) ### Microsoft Copilot Studio Microsoft Copilot Studio builds on Power Platform, Dataverse and Entra ID, and offers Anthropic models alongside OpenAI, which remains the default for new agents. It sells Copilot Credit packs of 25,000 at $200.00 per pack per month, a prepurchase plan, or pay-as-you-go through Azure. Strengths include: - **Connectors:** More than 1,500 prebuilt data connectors, including SAP, Workday and Salesforce. - **Governance:** Purview auditing, Entra ID groups and data policies. - **Government cloud:** A GCC environment that meets FedRAMP High. Weaknesses include: - **Preview controls:** Multistage approvals and request-for-information steps are in preview. - **Portability:** Exported solutions don’t include every component, and authentication must be reconfigured after import for some types. - **Deployment:** No on-premises option in the retrieved documentation. Best for: organizations whose data and identity live in Microsoft 365 and that have makers ready to build and maintain flows. ### Salesforce Agentforce Salesforce Agentforce grounds its work in CRM records and Data 360 and runs on the Agentforce 360 Platform. Pricing options include Flex Credits at $500 per 100,000 credits, Conversations at $2 per conversation, an Agentforce User License at $5 per user per month that requires Flex Credits, and add-ons at $125 per user per month. Strengths include: - **CRM grounding:** Work runs on the same records, objects and permission sets a Salesforce org already uses. - **Model choice:** Bring Your Own LLM and an LLM Open Connector for external models. - **Certifications:** FedRAMP High and ISO 42001:2023. Weaknesses include: - **Masking:** Data masking is disabled for agents. - **Billing structure:** Flex Credits and Conversations aren’t supported in the same org, and unused Flex Credits don’t roll over. - **Reach:** Systems outside Salesforce connect through MuleSoft, which is a separate integration project. Best for: companies where Salesforce is the system of record and an admin team owns the org. ### IBM watsonx Orchestrate IBM watsonx Orchestrate is built to govern AI across vendors, discovering and managing tools built on other platforms. Pricing lists Essentials starting at $578.00 per month and Standard starting at $6,360.00 per month, with Premium by contact and a 30-day free trial. Strengths include: - **Control plane:** Agent Identity, generally available November 3, 2026, with distinct identities, audit logs and a kill switch. - **Deployment:** [SaaS on AWS, AWS GovCloud and IBM Cloud, or on-premises](https://www.ibm.com/docs/en/watsonx/watson-orchestrate/base?topic=started-logging-in-watsonx-orchestrate). - **Analyst coverage:** IBM reports being named a [Leader in the inaugural 2026 Gartner Magic Quadrant for AI Agent Management Platforms](https://www.ibm.com/new/announcements/ibm-named-a-leader-in-the-2026-gartner-magic-quadrant-for-ai-agent-management-platforms). Weaknesses include: - **Roles:** [Custom roles are not supported](https://www.ibm.com/docs/en/watsonx/watson-orchestrate/base?topic=users-managing-cloud). - **On-premises lag:** Software Hub deployments [typically lag one release behind SaaS](https://www.ibm.com/docs/en/watsonx/watson-orchestrate/base?topic=notes-prebuilt-agent-feature-parity-across-deployments). Best for: large enterprises that already run AI from several vendors and need one place to govern it. ### UiPath UiPath extends RPA into agentic orchestration through Maestro, Agent Builder and the AI Trust Layer. Pricing lists a free Community plan, Basic starting at $25 per month, and Standard and Enterprise by sales contact, with a 60-day Standard trial. Strengths include: - **Hybrid execution:** Deterministic robots, AI steps and people in one BPMN process. - **Human review:** Action Center tasks, tool confirmations and guardrail escalations. - **Deployment range:** Cloud, Dedicated, Public Sector and self-hosted, including air-gapped. Weaknesses include: - **Cloud-first features:** Some capabilities are Cloud-only or have richer coverage in Cloud. Best for: companies with an existing RPA estate and process designers who want AI steps inside mapped processes. ### Automation Anywhere Automation Anywhere sells the Agentic Process Automation System, combining AI Agent Studio, RPA, document processing and the Mozart Orchestrator. It completed the Aisera acquisition on November 4, 2025 for IT service, and offers free Cloud Community Edition access; current paid prices were not published on the retrieved official pages. Strengths include: - **Process reasoning:** The Process Reasoning Engine plans across bots, AI and people. - **Certifications:** SOC 2 Type 2, ISO 27001, HITRUST and ISO 42001. - **Deployment:** Cloud and on-premises from one codebase. Weaknesses include: - **Licensing:** AI Governance requires an Enterprise Platform license. - **Pricing transparency:** No current public price list in the retrieved pages. Best for: RPA-heavy finance, procurement and IT operations that want reasoning layered on existing bots. ### Moveworks Moveworks handles IT and HR requests in chat and is now sold by ServiceNow, which renamed its combined AI experience ServiceNow Otto in August 2026. Pricing is by custom quote. Strengths include: - **Service-desk depth:** Plugins for Workday, ServiceNow and Salesforce and approvals mirrored into chat. - **Public sector:** FedRAMP Moderate, with standalone availability confirmed for public sector agencies. Weaknesses include: - **Write safety:** MCP Workspace has no confirmation step before a tool call. - **Roadmap items:** The Agentic Orchestrator is in Controlled Availability until a Q1 2027 upgrade. Best for: ServiceNow customers centralizing IT and HR requests behind one chat front door. ### C3 AI C3 AI builds enterprise AI applications on the C3 Agentic AI Platform and its ontology graph, with a strong public sector base. A 2024 filing describes a paid pilot of up to six months, then monthly fees plus vCPU and vGPU consumption; no current public price page was found. Strengths include: - **Deployment:** [C3-managed, customer VPC, hybrid, on-premises and air-gapped](https://c3.ai/api/media/file/c3-ai-operational-controls-description.pdf). - **Government:** [FedRAMP Moderate, building on IL5 and IL6](https://c3.ai/c3-ai-achieves-fedramp-authorization). Weaknesses include: - **Operational load:** Non-standard deployments [may need C3 AI’s Center of Excellence](https://docs.c3.ai/docs/platform/8.11/topic/admin-non-standard-overview), and on-premises LLM inference [needs added GPUs](https://docs.c3.ai/docs/platform/8.11/topic/admin-non-standard-requirements). - **Data use:** No explicit no-training statement in the retrieved pages. Best for: asset-heavy and government organizations building domain applications with dedicated data teams. ### Google Vertex AI and Amazon Bedrock Google’s Gemini Enterprise Agent Platform (formerly Vertex AI) and Amazon Bedrock are model infrastructure. Google bills Agent Platform tools, storage, compute and Cloud resources used; Bedrock bills per token by model across Standard, Flex, Priority and Reserved tiers. Strengths include: - **Model catalog:** [100+ foundation models on Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/what-is-bedrock.html) and 200+ models in Model Garden. - **Runtime controls:** [AgentCore Policy intercepts tool calls](https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/what-is-bedrock-agentcore.html); Google offers [project spend caps](https://docs.cloud.google.com/gemini/enterprise/docs/manage-costs-overview). Weaknesses include: - **Build effort:** Every workflow, connector and interface is the customer’s engineering work. - **Model lifecycle:** Bedrock legacy models reach [end of life and fail after that date with no automatic migration](https://docs.aws.amazon.com/bedrock/latest/userguide/model-lifecycle.html); Google’s [Agent Gateway is in Private Preview](https://docs.cloud.google.com/gemini-enterprise-agent-platform/release-notes). Best for: engineering teams building customer-facing or proprietary AI products on a cloud they already run. ## Build vs. buy, and avoiding vendor lock-in Engineering teams building on Bedrock or Google take full control and full maintenance responsibility. Companies buying a [ready AI employee](https://viktor.com/blog/build-ai-agent-or-hire-ai-employee) leave maintenance to the vendor. Gartner adds a third path to watch: it [predicts that by 2028, 70% of enterprises will abandon agentic AI built by vendor forward-deployed engineering](https://www.gartner.com/en/newsroom/press-releases/2026-09-29-gartner-predicts-70-percent-of-enterprises-will-abandon-agentic-ai-built-by-vendor-forward-deployed-engineering-by-2028), “trapped by soaring costs and unable to evolve it on their own.” Lock-in has three layers. Model lock-in eases when the platform supports several labs: Copilot Studio offers OpenAI and Anthropic models, Salesforce offers BYOLLM, UiPath supports bring-your-own-model configurations, and Viktor offers presets, models from several labs and an OpenRouter key. Data portability is the second layer. Export paths include Copilot Studio solutions, Salesforce DX metadata and UiPath solution packages; Viktor’s work is reachable through a public API and MCP server, and his outputs land as files and records in the company’s own tools. Process lock-in is the third. Skills that live as readable written procedures, as Viktor’s do, can be inspected and rewritten elsewhere. Logic buried in a proprietary canvas cannot. Consolidate point tools one workflow at a time. Retire a tool only after its replacement has run the same job for a full cycle, with the old tool’s outputs kept for comparison. ## How to spot agentwashing and LLM wrappers Gartner defines agent washing as rebranding existing products, such as chatbots and robotic process automation (RPA), without substantial agentic capabilities, and [estimates](https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027) that “only about 130 of the thousands of agentic AI vendors are real.” Four tests separate a working platform from a model wrapper: 1. **Executes across systems:** Ask the vendor to complete one job that reads from one system and writes to another during the demo. A wrapper returns text about the job. 2. **Holds context:** Repeat a request a week later and check whether the platform remembers the decision made the first time. 3. **Returns deliverables:** Ask for the end artifact (a file, a record update, a pull request) rather than a summary of what should happen. 4. **Respects approvals:** Trigger a sensitive action and confirm it stops for approval by default and does not run after a rejection. Preview labels are a fifth signal. Copilot Studio’s multistage approvals and Moveworks’ Agentic Orchestrator, both documented above, are not yet generally available. Score each finalist on all four tests in the trial, and drop any that fails two. ## How to choose an enterprise AI platform The right pick depends on where the company’s work already lives: - **Heavy Microsoft stack:** Copilot Studio fits when identity, data and makers sit in Microsoft 365 and Power Platform. - **Heavy Salesforce stack:** Agentforce fits when Salesforce is the system of record and an admin team owns it. - **ServiceNow IT and HR desk:** Moveworks fits when the job is employee requests routed into governed tickets. - **RPA estate:** UiPath or Automation Anywhere fits when robots and process designers already exist. - **Governing many AI vendors:** IBM watsonx Orchestrate fits as a control plane over other tools. - **Engineering-led build:** Bedrock, Google’s Agent Platform or C3 AI fits when the AI is the product. - **Company-wide execution in Slack or Teams:** Viktor fits when the backlog is cross-tool jobs and the team wants finished work without building flows. Readiness matters as much as the pick. Prepare documented workflows, updated knowledge bases, cleaned data and defined approval paths. A rollout that holds up follows four steps. The durations are planning ranges for a first workflow, not vendor commitments: 1. **Pick one pilot workflow (week 1).** Choose a recurring job with a clear owner and a finished artifact, such as the weekly pipeline report. 2. **Connect only what the job needs (days for a managed AI employee, weeks to months where connectors or processes must be built).** Grant the narrowest access, set sensitive tools to needs-approval, and confirm the audit trail captures tool calls. 3. **Plan change management (in parallel).** Name who hands work over, who approves, and where people see results; Gartner found vanguard CIOs, in a survey of 2,002 CIOs and technology executives from May to July 2026, [more likely to set stopping rules at pilot and production](https://www.gartner.com/en/newsroom/press-releases/2026-10-19-gartner-survey-finds-only-13-percent-of-cios-say-their-enterprise-has-realized-significant-or-transformational-value-from-ai). 4. **Set a production timeline (after one full cycle, usually four to six weeks).** Expand to more jobs only after the pilot has run a full cycle and passed its stopping rule. How long the whole path takes depends on the operating model. A managed AI employee such as Viktor is connected in days, because there is nothing to build. An ecosystem builder such as Copilot Studio or Agentforce usually needs a few weeks to a first production agent when identity and data already sit in that stack. Automation suites and model infrastructure are multi-month projects, because processes must be mapped or software written; C3 AI’s own filing describes a paid pilot of up to six months. Get each finalist’s timeline in writing, including UiPath and ServiceNow EmployeeWorks. To compare finalists without leaning on any vendor’s ranking, including this page, score them yourself. Give each criterion a weight that adds up to 100, score every platform from 1 to 5 on evidence from the trial, and multiply: ```json:object { "_key": "k724", "_type": "table", "rows": [ { "_key": "k718", "_type": "tableRow", "cells": [ "Criterion", "Suggested weight", "Evidence to score on" ] }, { "_key": "k719", "_type": "tableRow", "cells": [ "Fit with the systems you run", "25", "How many of your ten most-used systems connect without custom work" ] }, { "_key": "k720", "_type": "tableRow", "cells": [ "Governance", "25", "Scoped permissions, approval by default, audit trails that include tool calls" ] }, { "_key": "k721", "_type": "tableRow", "cells": [ "Time to first finished job", "20", "Days from contract to one recurring job running in production" ] }, { "_key": "k722", "_type": "tableRow", "cells": [ "Total cost at your volume", "20", "Licensing, usage, integration build and run costs for one year" ] }, { "_key": "k723", "_type": "tableRow", "cells": [ "Portability", "10", "Model choice, export path and readable process logic" ] } ] } ``` In Hampton’s case, Viktor [connected 24 tools](https://viktor.com/case-study), took over reporting, produced 12 dashboards and took on 26 recurring jobs a week within 44 days, and Hampton avoided $440K in planned hires. That is Hampton’s result, not a benchmark. [Read Hampton’s story to see how Viktor took on 26 recurring jobs a week in 44 days →](https://viktor.com/case-study) For a pilot that starts this week, hand Viktor the first recurring job. [Give Viktor his first job, free with up to $100 in credits, no credit card →](https://app.viktor.com/signup) ## FAQ A company-wide AI rollout needs scoped access, approval controls and a cost model matched to the work. The operating model determines who builds and maintains each workflow. ### What criteria should an IT leader use to evaluate enterprise AI platforms? Evaluate operating model first, then integration coverage of the company’s top systems, approval controls on sensitive actions, audit logs that include tool calls, data residency, training commitments and total cost. Score each finalist on whether it returns finished work across systems, not on feature counts. ### How are enterprise AI platforms priced? Model infrastructure such as Bedrock and Google’s Agent Platform bills usage per token or resource, Copilot Studio and Agentforce sell credit pools and per-user licenses, and Moveworks and enterprise tiers of UiPath and IBM are quoted by sales. Viktor is free to start with up to $100 in credits and no credit card, with usage credits shared across the workspace ([Viktor pricing](https://viktor.com/pricing)). ### How long does an enterprise AI platform take to deploy? Plan by operating model: days for a managed AI employee, a few weeks for a first agent on an ecosystem builder you already run, and several months for automation suites or a build on model infrastructure. Confirm implementation timelines for UiPath and ServiceNow EmployeeWorks during procurement. Viktor needs no build, and in [Hampton’s published case](https://viktor.com/case-study) he took on 26 recurring jobs a week within 44 days. ### What governance controls are required before a company-wide rollout? Require scoped permissions, approval before sensitive actions with rejection that blocks the action, audit logs that include tool calls, a hard spend cap, written no-training commitments for the vendor and its model providers, and current certifications. On Viktor, [Enterprise audit and access controls](https://viktor.com/security) include audit logs with integration tool calls, SAML SSO and per-user spend caps. ### Can a company replace point solutions without vendor lock-in? Yes, if it retires one tool at a time and picks a platform with multi-lab model choice, an export or API path and readable process logic. Run the replacement in parallel for a full cycle before switching off the old tool. Planning a company-wide rollout? [Talk to the Viktor team about onboarding, security review and spend controls →](https://viktor.com/contact-sales) [Hire your first AI employee: start free with up to $100 in credits, no credit card →](https://app.viktor.com/signup)