Docs / AMS / Start / What Mindset is
View as MarkdownWhat Mindset is
Learn what Mindset 4 does, who it suits and who it doesn't, and the pieces you build an agent from in Agent Management Studio.
After this page you can say what Mindset does, whether it fits your team, and what each piece of an agent is.
Mindset 4 is an agent framework with optimization at its core. You build agents in configuration, not code. Then you optimize them on three axes: cost, outcome and security.
You do both in Agent Management Studio (AMS), the web app where your org's admins build, test and watch agents.
Build
An agent is configuration you can read and change in AMS: its instructions, the model it runs on, the operations it may call and the other agents it may hand work to. Changing an agent doesn't mean deploying code.
| Piece | What it is |
|---|---|
| Agent | The thing that acts. A person in chat, a schedule, your own software or a vendor's webhook can start it. |
| Script | An ordered stretch of work, in phases, for steps that must wait for something or happen in a fixed order. Most agents don't need one. |
| Function | Fixed steps that give the same answer every time. The agent calls it like a tool. |
| Connection | A link to one outside system. It holds that system's credential, and the credential never reaches the model. |
| Operation | One named thing an agent may do on a connection, such as "get the purchase order for this invoice number". Each operation is a read or a write. |
| Run | One execution, recorded from what started it to how it ended. |
An agent reaches an outside system only through an operation on a connection. There is no other route out. The glossary defines every term.
Optimize
| Axis | What you do | Where to read more |
|---|---|---|
| Cost | See what each agent and function spends on models, then change the model or the work so it spends less. | See and control what it costs |
| Outcome | Run behavior tests and offline experiments to check whether a change made the agent better. | Test an agent's behavior |
| Security | Choose which operations exist, whether new write operations need a person's approval, and which environment an agent runs in. Read the audit record of what it did. | What Mindset enforces for you |
Who it is for
Mindset suits teams that:
- run a fleet of agents rather than one or two
- need agents that vary per customer
- want the people closest to customers to change agents without waiting for an engineer
- embed agents in their own product
Who it is not for
If you have one or two agents and one engineering team owns them, write them in code. Configuration, environments and approvals pay off when there are many agents and many people changing them.
How an agent reaches people
| Where | Who uses it |
|---|---|
| The Hub | People in your org, in a browser. |
| Your own product | Your customers, through the embedded element or the UI-less client. See How embedding works. |
| MCP | People working in Claude or another MCP client. See Use your agents from Claude. |
The example these pages use
Every AMS page uses the same agent: invoice exceptions. Each month, finance gets invoices that don't match the purchase order. The agent pulls the invoice and the purchase order, works out the difference, and checks whether the supplier contract allows it. Then it drafts a query to the supplier. Reading the invoice is a read. Sending the query changes another system, so it's a write.
Next, check whether your own automations fit: Will your automation work here?