How we research and rank enterprise agent platforms
Last updated: November 9, 2026
What we include
A product qualifies if it lets an enterprise build AI agents, run them in production, and govern them with enterprise controls: identity and permissions, audit, monitoring or evaluation, and admin oversight. We included both general platforms and agent layers built into major business suites, because many buyers choose between those two kinds of product.
How we rank
Rankings are an editorial judgment based on these criteria, roughly in order of weight:
- Breadth across the agent lifecycle: build, deploy, orchestrate, evaluate, govern.
- Enterprise controls: identity, least-privilege tool access, audit logs, environment and lifecycle management, data residency or isolation.
- Reach into enterprise systems: connectors, MCP and A2A support, and native access to systems of record.
- Flexibility: model choice, framework support and deployment options.
- Pricing transparency: published list prices and trials rank above quote-only offerings.
We do not rank by market share, and we do not reproduce any analyst report. Where a vendor cites analyst recognition, we attribute it to the vendor.
How we verified facts
We read vendors' official product, documentation and pricing pages on November 9, 2026, and every claim links to those sources. We have not tested any product hands-on and do not publish ratings or reviews.
Most facts were read directly from the official page. A few official pages could not be read by our tools. In those cases we relied on a research summary of the vendor's official pages and say so here:
- ServiceNow: the product page returned HTTP 403 (access denied to automated readers), documentation pages loaded only a cookie banner or loader, and a pricing URL timed out. Product structure, tiers and assist metering come from a research summary of ServiceNow documentation.
- OpenAI Frontier: the Frontier page returned HTTP 403 behind a bot challenge, and OpenAI's business pricing page shows no prices. Frontier is described only briefly and listed as contact-sales. OpenAI's developer stack, API pricing and the Agent Builder shutdown date (November 30, 2026) were read directly from OpenAI's AgentKit announcement update and developer documentation.
- Oracle: AI Agent Studio capabilities were read directly from Oracle documentation. Custom AI Agents prices come from Oracle's price list and FAQ, which are PDFs our reader cannot open, so they rest on a research summary of those documents.
- Google Agent Runtime rates: Google's Agent Runtime documentation points to the Agent Platform pricing page. That page is about 120,000 characters long and our reader stops at 40,000, before the agents section. The $0.085 per vCPU-hour and $0.009 per GiB-hour rates come from a research summary of that page. Google's model token prices, Gemini Enterprise editions and the Business edition's $21/seat price were read directly.
- Kore.ai: no pricing is published; the pricing URL returned HTTP 404.
- AWS: we quote only AgentCore prices read directly (Web Search, Web Answer, Free Tier credits), not per-vCPU runtime rates.
What we do not do
- We do not claim hands-on testing.
- We do not publish user reviews or invented ratings.
- We do not accept payment for placement.
Updates and corrections
Product names and prices in this category change often; for example, Google renamed Vertex AI to Gemini Enterprise Agent Platform in 2026. We re-check official pages when we update the guide, and each page shows its update date. Prices are US-dollar list prices unless noted.