Most businesses we talk to already pay for AI. Someone drafts emails with a chat assistant, the office suite has its AI features switched on, and the CRM offers to summarise a record. The question that follows is a fair one: do we need anything built, or are the tools we have enough?
Our answer is usually both, in a particular order. Get real value from the tools you already pay for first, and teach people to use them. Build an agent only for the work those tools cannot reach.
What each one is
An AI tool is something a person picks up. You open it, ask it something, read what comes back and decide what to do with it. A chat assistant, the writing help inside your email, the summary button in your CRM: each makes one person quicker at a task that person has already started.
A custom AI agent does a piece of work itself. It starts when the work arrives, not when someone remembers to open it. It works from material your business has approved, does a defined job inside the systems you already run, and leaves the result where your team will see it. A person stays in control of what goes out.
Put simply, a tool helps someone do the work. An agent does a recurring part of the work, and someone checks it.
Five differences that matter
- Who starts the work. A tool waits for a person to open it and type. An agent starts by itself when something arrives: a call, an email, a new ticket, a form.
- Where the answer comes from. A tool answers from its general knowledge and whatever it can see when the person asks, so the answer depends on who asked and how. An agent answers from the FAQ, documents and rules you signed off, and from nothing else.
- Where the result lands. With a tool, the result sits in a chat window until someone copies it somewhere useful. An agent writes to the mailbox, calendar, helpdesk or record your team already works from.
- Who checks it. With a tool, the only check is the person who asked, and only if they read carefully. With an agent, the checks are designed before it goes live: what a person approves, what it may answer, and who it hands to.
- Who owns it. A tool belongs to its vendor and changes when the vendor changes it. An agent we build sits inside your own accounts, and it is yours.
Signs you have outgrown tools alone
- The same work arrives every day. Not a task someone does once a month, but the same kind of call, email or request, many times a day, that somebody has to stop and deal with.
- The answer is already written down. You have an FAQ, a policy, a set of approved replies or a schedule, and people still answer from memory or go looking for the document each time.
- The result has to land in a system of record. Someone uses a tool to draft or summarise, then copies the output into the CRM, the schedule or the helpdesk by hand. The tool saved a minute and the copying took it back.
- Work waits for someone to be free. Calls go to voicemail while the team is out on a job, the inbox is read at the end of the day, tickets sit unassigned. A tool cannot help with work nobody has opened yet.
When two or more of these describe the same piece of work, an agent is worth a conversation.
Signs you have not
Do not build an agent for work that turns up a few times a week. A person with a good tool will handle it well, and an agent needs looking after whether it is busy or idle.
Do not build one when the answer is not written down. If the right reply lives in one colleague’s head, an agent will guess or hand everything back. Write the material first. That job is worth doing whatever you decide next.
Do not build one for something a tool you already pay for can do once it is switched on and set up. Look properly at the AI features in your office suite, email and CRM before you commission anything. Often the gap is setup and habit, not capability.
And do not build one because the tools feel disappointing. If nobody has shown your people how to use them on their own work, the problem is training, not software.
How the two work together
Tools and agents are not rivals. They cover different ground, and a business that uses both well tends to get there in the same order.
First, the team learns to use what it has. That is what our training is for. The AI skills programmes are hands-on sessions built from a department’s real files and daily tasks, and the team keeps rules for what AI may and may not do and a prompt and assistant library built from its own work. Leaders can begin with a strategy workshop to choose where AI goes first and what it may do alone.
Our training is not only about AI. We also build custom e-learning, compliance and safety courses, code of conduct training, live sessions and learning resources from your own people, documents and processes.
Then, once people use the tools well, the recurring work stands out. It is what still arrives every day, still has a written answer and still has to land in a system. That is where an agent fits: inbound call taking for the phone, inbox triage and reply drafts for email, a company knowledge assistant for the questions staff keep asking. Every agent we build is listed by team.
The agent does not replace the tools. Staff still open a chat assistant to think through a difficult email. The agent takes the routine ones that arrive all day.
How a person stays in control
Where an agent writes for you, a person approves anything that leaves the building. Reply drafts wait in your drafts folder, and nothing is sent until someone sends it. Where an agent answers customers directly, on the phone or in chat, no one signs off each reply as it happens. The rule there is different: It answers only from material you approved. Outside that, it takes a message or hands off.
Who owns it
Every agent we build is built inside your own accounts, on Microsoft Copilot Studio, n8n or custom code, so you own the agent and the intelligence in it: the rules, the prompts and the data. We build it and set it up for you, and answers land in the system of record you already use.
Which of those three it runs on is a separate decision that comes later. We wrote about how we choose where an agent lives.
Where to start
List the AI tools you already pay for and who really uses them. Then list the work that arrives every day. Where a tool sits unused, that is a training question. Where the work is repeated, written down and has to land in a system, that is a candidate for an agent. To choose which one goes first, read which AI agent a small business should start with.
Talk to sales. Tell us which tools you have and which work keeps arriving, and we will tell you whether you need training, an agent or neither yet.
