# AI agent

An AI agent is software that pursues a goal through several steps, picks an allowed next action, uses tools, and responds to each result.

Source: https://rodamora.com/glossary/ai-agent
Updated: 2026-08-09

---

In the 150+ franchise units in the delivery data I review, roughly 80% still use chatbots, 15% use trigger-based workflows, and about 5% use agents started by conditions. The distinction is simple. A chatbot waits for a prompt. A workflow follows a written path. An agent reads the current state, chooses an allowed next action, uses a tool, and checks the result. Then it continues, stops, or asks for help. For a service firm, an agent is useful when the next action cannot be listed before the run, but the firm can still set a goal, permissions, completion check, and recovery owner. The word autonomous does not remove responsibility. The firm still owns the result.

## What does an AI agent do?

An AI agent repeats a small loop around a goal. It gathers information, decides what to do, acts through a tool, reads the result, and decides whether another step is needed. The loop ends when the job is complete, when a rule says to stop, or when a person must take over.

[OpenAI's practical guide to building agents](https://openai.com/business/guides-and-resources/a-practical-guide-to-building-ai-agents/) describes three basic parts: a model, tools, and instructions. The model reads the current state and selects an action. Tools let the software read or change another system. Instructions define the job, the allowed actions, and the limits.

Consider a new client request. A model can draft a reply from text that a person pastes into a chat. An agent can open the approved account record, check the current scope, find the last client commitment, choose a response type, draft the reply, create the next task, and route a pricing exception to a person. Each result changes the next choice.

The action matters as much as the words. Software that writes a polished email and leaves it in a chat window has made content. Software that updates the correct client record, creates the correct task, and leaves evidence of both has completed part of the job.

## How is an AI agent different from nearby terms?

These labels describe different parts of the work. Keeping the parts separate helps you buy and build the right thing.

| Term | What it names | Who chooses the next step | Example |
| --- | --- | --- | --- |
| Chatbot | A prompt-and-response interface | The person | A manager asks for a follow-up email |
| AI workflow | A fixed sequence with AI inside it | The written path | A trigger drafts, routes review, and files the result |
| AI agent | Software pursuing a goal | The model, inside limits | The agent checks an account and chooses an allowed next action |
| Agentic AI | Model-led choice and action | The model, inside the system | One or more agents adjust their process from what they find |
| AI employee | A software role defined by the firm | Depends on the role and its controls | A client-health role starts from conditions and owns a recorded result |

[Agentic AI](https://rodamora.com/glossary/agentic-ai) names the behavior of choosing and taking action. The AI agent is the software component doing the work. One agent can behave agentically. A system with several agents can still follow a mostly fixed path.

An [AI workflow](https://rodamora.com/glossary/ai-workflow) follows steps defined before the run. Anthropic's [engineering guidance](https://www.anthropic.com/engineering/building-effective-agents) draws the same boundary. Workflows use predefined code paths. Agents dynamically direct their process and tool use. A workflow can still contain one agent for a decision that cannot be mapped ahead of time.

An [AI employee](https://rodamora.com/blog/what-is-an-ai-employee) is the organizational role assigned to the software. It has a job, access, trigger, and way to validate the result. The agent is one way that role performs flexible work. [Orchestration](https://rodamora.com/glossary/orchestration) coordinates state, handoffs, retries, and stop conditions around the run.

## What does an AI agent look like in a service firm?

The first useful agent usually handles one narrow work loop. It should not try to run a whole department.

Take client health. The agent starts when an account meets a condition, such as four business days without a recorded client touchpoint. It reads the approved account history, checks for a scheduled meeting or delivery, and chooses among three actions. It can draft a check-in, create an internal task, or send the evidence to the account owner when judgment is needed.

The run ends in the system where the firm manages delivery. The task exists. The draft is attached to the correct account. The source records are preserved. A final message that says “done” proves none of these things.

Anthropic's [guide to agent evals](https://www.anthropic.com/engineering/demystifying-evals-for-ai-agents) makes this distinction clear. The outcome is the final state in the environment. The transcript is the record of steps the agent took. A booking agent can say it booked a flight while no reservation exists. A service agent can say it handled a request while the CRM points to the wrong client.

A long prompt that produces the same client-health draft can still help a manager. The person finds the account, collects the context, starts the work, and carries the answer into the CRM. That is assisted work. The agent starts to own the loop when a defined condition starts it and a checked business record finishes it.

## What authority should an AI agent have?

Tool access is authority. Reading the CRM, sending a client message, changing a price, and deleting a record have different consequences. They should not arrive as one permission called “CRM access.”

In February 2026, [NIST opened work on agent identity and authorization](https://www.nist.gov/news-events/news/2026/02/new-concept-paper-identity-and-authority-software-agents) because agents can reach data, tools, and applications. Its concept paper names identification, authorization, auditing, non-repudiation, and prompt-injection controls.

For a service firm, the operating rules are plain:

- give the agent its own identity;
- grant only the access the job needs;
- record reads, decisions, and writes;
- require approval before a hard-to-reverse action;
- name the person who receives a stopped or failed run.

Start with read access and a draft. Add a reversible write after the agent passes representative cases. Keep a person before money moves, scope changes, a client receives a promise, permissions change, or a record disappears.

Use this test for autonomy: can the firm [check the result and undo the action](https://rodamora.com/blog/delegate-the-inputs-own-the-outputs)? A better model can improve the answer. It does not make a bank transfer reversible or a client promise cheap to retract.

## When is a workflow better?

Use a workflow when the path can be drawn before the run starts. It will be easier to test, cheaper to operate, and more predictable when a step fails.

A proposal process often has a stable sequence. Collect the approved transcript, match the CRM record, extract requirements, draft, review the price, and file the final version. The model can write inside that path without deciding what the whole process should do next.

An agent is worth its extra cost when the next action depends on evidence found during the run. A support request may need a policy lookup, a refund check, a technical investigation, or a handoff. Writing every branch as ordinary rules may cost more to maintain than giving one bounded agent a small tool set and a clear decision standard.

Anthropic recommends starting with the simplest design that works because agents trade latency and cost for flexibility. A predictable path can be the more mature design. A fixed workflow fails along a known path. An agent may reach the same wrong outcome through several paths, so investigation becomes part of the operating cost.

One agent is usually enough. OpenAI recommends expanding a single agent's tools before splitting work across specialists. A second agent adds a handoff, another context boundary, and another place for responsibility to blur. Add one when the work has a real specialist boundary. Five boxes in a diagram do not show maturity.

## How should a service firm test its first agent?

Test the job before testing the demo. Write the goal, allowed tools, final business record, stop rule, and recovery owner before the first production run.

Build cases from real work. Include a clean request, missing information, conflicting client records, a tool timeout after a write, an instruction hidden inside an untrusted document, and a request the agent should refuse. Run several trials because model choices vary.

Score the final state with an [eval](https://rodamora.com/glossary/eval). Check that the right account changed, the right evidence was used, the action stayed inside permission, and the agent stopped where the rules required. A strong draft attached to the wrong client is a failed run.

Public evidence about agent testing remains thin. The [2025 AI Agent Index](https://aiagentindex.mit.edu/data/2025-AI-Agent-Index.pdf) documented 30 prominent deployed agents. Only four provided agent-specific safety evaluations. A product page cannot answer the firm's deployment questions. The firm needs its own cases, records, and recovery tests.

## Where does an AI agent sit on the Delivery Model Ladder?

An AI agent sits at Stage 2, Augmented, on the [Delivery Model Ladder](https://rodamora.com/delivery-model-ladder) when a condition starts a bounded job, the agent uses approved company systems, a check confirms the result, and a named person receives judgment calls or failures.

At Stage 1, a person starts the run, watches it, and carries the result into the next system. The model may make several choices, but the person still holds the workflow together.

Stage 3 requires a firm-level change in delivery. A complex group of agents can remain at Stage 1 if people repair every handoff. One plain agent with limited tools, a reliable check, and clean recovery can be solid Stage 2. The AI employee is the home for that operating role.

Use an agent where the next action cannot be fully mapped in advance and the firm can test, monitor, stop, and recover the work. Use fixed automation or an AI workflow everywhere else.