Can You Buy AI Agents? What Exists in 2026

This guide is for founders, CMOs and business leaders deciding whether to buy AI agents rather than build them. It is the fourth of five parts, and it describes the market as of October 2026.

AI Summary

Yes, you can buy AI agents, and for most first jobs you should. In 2026 they come in three forms: agents built into software you already pay for, such as Salesforce, HubSpot, Microsoft 365 and Google Workspace; specialist agents for one function, such as customer support, sales development, research or software; and general-purpose agents in the AI apps, such as Claude and ChatGPT. Agent marketplaces from Salesforce, Google, AWS and Microsoft make them easier to find and buy. Pricing is moving from seats to outcomes, such as paying per resolved conversation. Before you buy, test the vendor's agent against your own job description, agree what counts as an outcome, and make sure you can take your instructions and knowledge with you if you leave.

Background

In Part 2, we described a ladder: use an existing agent, configure a general one, or build your own. This part covers the first rung, using what already exists. Part 5 covers the platforms for building and running your own.

A word of warning before the list: this market changes monthly. In 2026 alone, Google renamed Vertex AI the Gemini Enterprise Agent Platform, Microsoft renamed Azure AI Foundry to Microsoft Foundry, Salesforce merged its app stores into AgentExchange, and OpenAI announced it would shut down its Agent Builder in November, only a year after launching it. Treat any list, including this one, as a starting point and check current terms.

Agents Built Into Software You Already Use

The easiest agents to buy are the ones your current vendors already sell. They know your data, run under your existing contract and permissions, and need little setup. Their limit is the product's walls: they are best at jobs that live inside that product.

• Salesforce Agentforce. Agents for sales, service and marketing inside Salesforce, with Slack now acting as an agent hub that other companies' agents can plug into.
• HubSpot Breeze. Agents for customer service, prospecting and data inside HubSpot. In 2026 HubSpot moved its Customer Agent to pricing per resolved conversation.
• Microsoft 365 Copilot and Copilot Studio. Agents in Outlook, Teams, Excel and SharePoint, plus a builder for your own. Microsoft's Agent 365 adds a registry to govern agents from any vendor.
• Google Gemini Enterprise. Google's workplace agent hub, formerly Agentspace, with agents across Google Workspace and a gallery of partner agents.
• Help desks. Intercom Fin, Zendesk and ServiceNow all sell agents that answer and resolve customer and employee requests, mostly priced per resolution or per assist.
• Other business software. Adobe sells agents for audiences, journeys and site optimization in its Experience Platform. Workday sells HR and finance agents. Notion sells custom agents that work across your workspace.

If your team lives in one of these products, start here. Ask your account manager which agents are already included in what you pay.

Specialist Agents for One Job

Specialist vendors build one agent and make it very good:

• Customer support. Sierra, Decagon and Intercom Fin build agents that resolve support conversations end to end, usually priced per resolution.
• Sales development. 11x and Artisan sell AI SDRs that research prospects and run outbound sequences, typically on monthly contracts.
• Research. ChatGPT, Claude and Perplexity all include research agents that read dozens of sources and write a cited report, included in paid plans.
• Software. Claude Code, OpenAI Codex, GitHub Copilot's coding agent, Cursor and Devin write, test and fix code. This is the most mature agent category, and the one where agents most often run for hours on their own.
• Browsing. ChatGPT Atlas, Perplexity Comet, Claude in Chrome and Claude Cowork's built-in browser can click, type and fill in forms on websites for a person.

Specialists usually beat built-in agents on their one job and lose on integration. They are worth it when the job is big enough to matter on its own, such as support at volume or outbound at scale.

General-Purpose Agents in the AI Apps

The AI apps have become agents too. Claude Cowork, generally available on paid Claude plans since April 2026, works through files, applications and a built-in browser, and can run scheduled tasks. ChatGPT Workspace Agents, in research preview, replace custom GPTs with agents that run on a schedule and work in Slack. Microsoft Copilot plays the same role inside Microsoft 365.

These are the most flexible agents you can buy, because you configure them with your own instructions, documents and connectors. They sit on the “configure” rung of the ladder from Part 2, and for many marketing teams they are the best first agent of all.

Agent Marketplaces

Agents now have stores, which makes buying them look more like buying apps:

• Salesforce AgentExchange, which since April 2026 combines AppExchange, the Slack Marketplace and the Agentforce ecosystem.
• Google Cloud Marketplace, where partners sell agents that plug into Gemini Enterprise over the A2A protocol.
• AWS Marketplace, with a category for AI agents and tools.
• Microsoft's Agent Store in Microsoft 365 Copilot.
• The ChatGPT app directory, where companies publish apps that ChatGPT can use, built on MCP.

A marketplace listing usually means the agent works with that platform's identity, permissions and billing, which saves your IT team real work.

How Agents Are Priced

Outcome pricing is the most attractive model on paper, because you pay only for work done. It also moves the hard question into the contract: what counts as an outcome? A “resolved” conversation might mean the customer said thanks, did not come back within a day, or confirmed the issue was fixed. Agree on the definition, and ask for a report you can audit.

Whatever the model, ask the vendor to run their agent on a sample of your real work and tell you what it would have cost.

How to Evaluate an Agent Before You Buy

Use the job description from Part 1 as your buying checklist:

• Does it do your job? Run it on twenty of your real cases, not the vendor's demo. Compare with your best person.
• What can it see and do? Which systems it reads, which it can change, and whether you can narrow that.
• When does it hand off? How it escalates to a person, and what context the person gets.
• Can you see its work? Logs of what it did and why, available to you, not just the vendor.
• What does an outcome mean? In writing, if you pay per outcome.
• Does it speak the standards? Support for MCP, to connect your tools, and A2A, to talk to other agents, keeps your options open.
• Can you leave? Can you export your instructions, knowledge, conversation history and test cases? The market is moving too fast to be locked in.

Buy or Build?

Buy when the job lives inside a product you already use, when a specialist has clearly solved it, or when speed matters more than difference. Build, or at least configure your own, when the job is how you win, when it crosses several systems, or when no vendor's agent passes your twenty-case test. Most companies end up doing both: bought agents for common jobs, and a few of their own where they are different. Part 5 covers the platforms for the second.

Conclusion

Can you buy AI agents in 2026? Yes: built into the software you already use, from specialists for single jobs, and as general-purpose agents in the AI apps, increasingly through marketplaces and increasingly priced per outcome. Buy for common jobs, test every agent on your own work before you sign, define outcomes in the contract, and keep your instructions and knowledge portable.

Next in the series: Platforms to Run Your Own AI Agents: Bring Your Own Agent.

Before you buy, test your own website. Send our A2A agent to it to see if vendors' and buyers' agents can talk to you, or run MachineReady. Get in touch and we can test a shortlist of agents against your own work.