"AI agent" now covers everything from a website chatbot to a promised autonomous employee. For an owner of an HVAC, plumbing, electrical or roofing company, that makes the term close to useless when a vendor says it on a sales call.
This guide gives you a working definition, shows where agents fit in a service business, explains how they connect to your files and software, and lists the guardrails to ask about. It also answers a question we get often: when is a plain automation enough?
What an AI agent is
The clearest definition we have found comes from Anthropic's engineering guide on building agents. It draws a line between two kinds of system:
- Workflows are systems where a language model and other tools are orchestrated through predefined code paths. The steps are fixed in advance.
- Agents are systems where the model dynamically directs its own process and tool use, deciding how to accomplish a task.
In plain terms: a workflow follows a script you wrote. An agent is given a goal and a set of tools, and it decides which tools to use and in what order.
Here is what that looks like in a trades company.
- Workflow: a form is submitted, the system texts the customer within a minute, asks three set questions, and creates a lead in your field service software. Same steps every time.
- Agent: a caller says the upstairs unit is leaking through the ceiling and they are not sure whether it is the AC or the water heater. The agent decides to ask about water color and smell, checks the schedule, offers the next emergency slot and flags the call for the on-call tech.
Most of what a contractor needs today is a workflow with a language model inside one or two steps. Anthropic's guide says as much: find the simplest solution that works, and add complexity only when needed. It also notes that agentic systems often trade latency and cost for better task performance, and that autonomy brings the risk of compounding errors, so they should be tested extensively with appropriate guardrails.
Where agents fit in a trades company
Four places show up again and again. None of them requires a science project.
1. Answering and qualifying inbound calls
Phones ring while the tech is on a roof and the CSR is on another line. A voice agent can answer, collect the service need, address and urgency, offer open appointment slots and escalate real emergencies. This is the most visible use and the most oversold. What to test before you sign is in our guide to AI receptionists for contractors.
2. Following up on leads and estimates
A lead goes quiet. An estimate sits unsigned. A follow-up sequence is mostly a workflow, but an agent can handle the replies: a homeowner who texts "can you do Thursday instead" should get a sensible answer, not a dead end.
3. Admin work around the job
Summarizing a messy service history before a tech arrives, drafting the invoice description from tech notes, sorting incoming emails into dispatch, billing and vendor piles. These are low-risk, internal tasks where a human reviews the output. Our AI operations and admin automation service covers this kind of work.
4. Reporting
Pulling booked jobs by branch, flagging the location whose response times slipped this week, and drafting the summary leadership reads on Monday. The agent reads data and writes a draft. It does not change anything.
How an agent reaches your data and software
An agent is only as useful as what it can see and do. Connecting a model to external systems used to mean custom code for each pairing. The Model Context Protocol (MCP) is an open standard for connecting AI applications to external systems such as data sources, tools and workflows. The MCP documentation describes it as a standard way to connect AI applications to external systems, like a USB-C port for AI applications. Assistants including Claude and ChatGPT support it.
Two practical points for an operator:
- A connection is not a strategy. Turning on a connector to your shared drive or CRM tells you nothing about whether the answers are right, who can see what, or what the agent is allowed to change. Decide the permissions first.
- Read access and write access are different decisions. An agent that can read a customer's service history is a different risk from one that can edit the job, change the price or send a text. Start read-only and add write access one action at a time.
For field service software, integration is the hard part of most projects. We cover it in our guide to AI integration with ServiceTitan, Housecall Pro and Jobber.
Guardrails to ask about
Whether you build or buy, ask these before an agent touches customers.
- What can it do without approval? List the actions, such as booking an appointment, quoting a price, issuing a refund or sending a text. Decide which need a person.
- What does it do when it is unsure? The right answer is a clean handoff to a human with the conversation attached, not a guess.
- Can you see what it did? You want a log of every conversation and every action, searchable by date and branch.
- Where does customer data go? Read the vendor's data terms: whether your data trains their models, how long it is retained and who can access it. Our AI compliance guide for small service businesses has a checklist.
- How will you know it is working? Measure booked jobs, not conversations handled.
When a plain automation is enough
If the steps never change, you do not need an agent. A missed-call text that sends the same message every time, a review request after a job closes, an appointment reminder: these are workflows. They are cheaper, more predictable and easier to debug. Use an agent where the situation varies in ways a script cannot anticipate, such as a conversation, and keep a person in the loop wherever a mistake is expensive.
If you do not know where your business sits, start with the leak. Where to start with AI as a contractor walks through how to pick the first project. ServiceTitan's April 2026 survey of 1,000 residential contractors found that 74% see AI as key to efficiency and that more than half do not know where to begin, so you are not behind.
How Di-Hy approaches it
Di-Hy sells no software product, so we pick tools to fit your stack. Our Booked-Jobs Recovery System is a fixed-scope engagement, live in 30 days and measured in booked jobs. It uses a mix of workflows and agents for missed-call recovery, after-hours qualification and follow-up, wired into your field service software. If you want help deciding what should be an agent and what should be a script, book a strategy call. Pricing is scoped on the call.
Frequently asked questions
What is the difference between an AI agent and a chatbot?
A chatbot answers questions in a conversation. An agent can also take actions with tools, such as checking a calendar or creating a record, and decides which actions to take. Many products called agents are workflows with a language model in one step.
Do I need an AI agent, or is automation enough?
If the steps are the same every time, automation is enough and is cheaper and easier to test. Agents fit situations that vary, such as open-ended conversations.
Is it safe to let an agent talk to my customers?
It can be, with limits. Define what it may do without approval, require a handoff when it is unsure, log everything and review the logs. Calls and texts also carry legal requirements, covered in our texting rules for contractors.