AgentExchange

AgentExchange is Salesforce's marketplace for AI agents, apps and add-on tools that run on Salesforce or Slack. It's where a Salesforce customer goes to find, try, buy and install software built by Salesforce and its partners.

The name has two chapters. Salesforce launched AgentExchange in March 2025 as a separate marketplace for Agentforce, its AI agent platform. It sold ready-made agent parts: actions (a single job an agent can do, like creating a case), prompt templates, topics (now called subagents) and full agent templates.

By September 2026, Salesforce had merged its older app store, AppExchange, and the Slack Marketplace into AgentExchange. The old appexchange.salesforce.com address still works and now opens AgentExchange. The combined marketplace lists traditional apps, consulting partners and Slack apps alongside agents. It also lists newer types of add-ons, such as skills, plugins and MCP servers. MCP (Model Context Protocol) is an open standard that lets an AI agent connect to outside data and tools.

Why it matters for lenders

For a lender on Salesforce, AgentExchange is now where both kinds of software live. These are the apps that manage loans and run checks, and the AI agents that can take actions on loan records. Many people will keep saying "AppExchange" for years, so it helps to know the two names mean the same place.

The trust signals carry over. Salesforce says listed solutions go through its security review, which tests how they handle data and how securely they're coded. Customer ratings and reviews are public. Many apps still install as managed packages, so the vendor handles upgrades.

An agent needs more scrutiny than an ordinary app, though. An app usually shows data to a person, while an agent can change records or contact borrowers. A security review checks the code. It doesn't check whether the agent's decisions fit a lender's credit policy or fair-lending duties. That review stays with the lender.

How it works in practice

A CDFI wants an agent that drafts reminder messages to borrowers who are a few days late. Its Salesforce administrator searches AgentExchange and shortlists two listings. For each one, the team asks four questions:

1. Which records can the agent read and change?
2. Does it send anything to borrowers on its own?
3. Which outside services does it call?
4. Does it log every action?

The team installs the stronger option in a sandbox, a test copy of the org, and runs it against past delinquent loans. They set it up so a collector approves each message before it goes out, a human-in-the-loop control. Only after compliance signs off on the vendor review and the approval step do they turn it on in production.