Question 1

What are AI agents?

Accepted Answer

AI agents are software systems built on a large language model that can take a goal, plan the steps and carry them out using tools: reading emails, looking up data, updating a CRM or drafting a document. Unlike a chatbot, an agent acts rather than just replies, working within instructions, permissions and approval rules that you set.

Question 2

How is an AI agent different from a chatbot?

Accepted Answer

A chatbot answers questions in a chat window and waits for the next one. An AI agent can also take action: it carries out tasks in your systems, can start by itself when something happens, and strings several steps together to finish a job, such as replying to a lead, logging it in the CRM and booking a call.

Question 3

How are AI agents different from Zapier or n8n automations?

Accepted Answer

Rule-based automation in tools like Zapier, Make or n8n follows fixed steps and stops when the input doesn't match. An AI agent reads and understands the content, copes with variation and exceptions and decides which steps are needed. In practice we often combine them: fixed automation for predictable steps, an agent where judgement is needed.

Question 4

How do AI agents actually work?

Accepted Answer

The agent is given a goal and a set of tools, such as access to an inbox, a CRM or a database. The language model chooses a tool, reads the result and decides the next step, repeating until the job is done. Instructions, business rules and approval steps limit what it is allowed to do at each point.

Question 5

What can AI agents do for a small or medium business?

Accepted Answer

Typical jobs include replying to and qualifying leads, processing supplier invoices, compiling management reports, chasing unanswered quotes, preparing briefs before sales calls and sorting incoming requests. The best candidates are tasks that repeat every week, follow a recognisable pattern and need a little judgement that fixed rules can't provide.

Question 6

What are autonomous agents?

Accepted Answer

Autonomous agents work through a task from start to finish without a person steering each step. Full autonomy is rarely the right choice for a business. We build agents that run on their own for routine work but stop and ask whenever a decision is important, unusual or the agent isn't confident in the answer.

Question 7

Can several AI agents work together?

Accepted Answer

Yes. In a multi-agent setup each agent has its own role, for example one that triages incoming requests and another that drafts replies, and they hand work to each other. Splitting a large workflow into smaller agents makes it easier to build, test and fix, because you can see exactly which step went wrong.

Question 8

Which AI models do your agents use?

Accepted Answer

We choose the model for the task, typically from OpenAI's GPT models, Anthropic's Claude, Google's Gemini or Mistral. Different steps sometimes use different models to balance quality, speed and cost. Because the agent's logic sits outside the model, we can switch to a better or cheaper model later without rebuilding.

Question 9

Can an AI agent use our own documents and data?

Accepted Answer

Yes. Using retrieval-augmented generation (RAG), the agent searches your documents, contracts, price lists and past cases and uses what it finds when it replies or decides. It only retrieves what the task needs, and your documents are not used to train the underlying model.

Question 10

Is an AI agent the same as ChatGPT?

Accepted Answer

No. ChatGPT is a tool where a person types a question and gets an answer. An AI agent can run on the same kind of model, but it works inside your systems, starts by itself when an email or form arrives and completes the whole job, rather than handing a person a suggestion to copy and paste.

Question 11

What does agentic AI mean?

Accepted Answer

Agentic AI is the umbrella term for AI that doesn't just respond but acts: it plans, uses tools and takes several steps towards a goal. AI agents are the practical way to put agentic AI to work in a business, applied to specific jobs with clear limits, logging and a person approving the decisions that matter.

Question 12

What's the difference between an AI assistant and an AI agent?

Accepted Answer

An AI assistant such as Copilot or ChatGPT helps a person while they work and waits for the next prompt. An AI agent is given a job and does it by itself, even when nobody is at a computer, then reports back when it's finished or needs help. Many businesses use both side by side.

Question 13

What types of AI agents are there?

Accepted Answer

In business use, the common types are intake agents that handle incoming requests, document agents that prepare quotes and reports, follow-up agents that chase leads and clients, and integration agents that move information between systems. Several can work together in one workflow, each with a narrow, well-defined role.

Question 14

Are AI agents regulated in the UK?

Accepted Answer

The UK has no single AI Act. Instead, existing regulators apply existing law, so an AI agent handling personal data must comply with UK GDPR, and the ICO publishes guidance on AI and data protection. In practice that means a lawful basis for processing, data minimisation, transparency and human review of significant decisions, all of which we build in.

Question 15

Can an AI agent make decisions by itself?

Accepted Answer

Within limits you set, yes. It can decide whether an email is a new enquiry or an invoice and choose the next step accordingly. Decisions with bigger consequences, such as sending a quote, issuing a refund or making a payment, we always configure to need a person's approval first.

Question 16

What is MCP and why does it matter for AI agents?

Accepted Answer

MCP, the Model Context Protocol, is an open standard for connecting AI agents to tools and data sources. A growing number of software vendors offer ready-made MCP connectors, which makes it quicker and safer to give an agent access to exactly the systems and actions it needs, and nothing more.

Question 17

Are AI agents safe for business use?

Accepted Answer

They are when they're built with the right limits. We give each agent access only to what its job requires, require approval for important actions and log every run, so nothing happens that can't be traced afterwards. We also test the agent on real examples before it touches live customer data.

Question 18

What happens when an AI agent gets something wrong?

Accepted Answer

Agents make mistakes, just as people do. That's why we test on real cases before go-live, set the agent to stop and ask when it's unsure, and review runs afterwards. Each correction becomes a clearer instruction or rule, so the error rate falls over time instead of repeating.

Question 19

Where is our data processed?

Accepted Answer

We choose providers and settings so data is processed in line with UK GDPR, keeping it in the UK or EU where that is required. We use business tiers of AI services where your data isn't used to train their models, and we document the data flows so you can answer questions from clients or auditors.

Question 20

Can we control exactly what the agent is allowed to do?

Accepted Answer

Yes. You decide which systems the agent can read and which it can change, what it may do alone and what always needs sign-off. Permissions are set per action, not just per system, and you can tighten or loosen them at any time without rebuilding the agent.

Question 21

Will AI agents replace our staff?

Accepted Answer

Rarely. They take over the repetitive steps, such as sorting, compiling and chasing, which frees people for customers and work that needs judgement. Most businesses we speak to use agents to handle growth without hiring more admin staff, rather than to cut existing roles.

Question 22

Who is accountable for what the agent does?

Accepted Answer

You own the process and set the rules; we are responsible for building and testing the agent to follow them. Every action is logged with what the agent saw and why it acted, so if a question comes up later, from a customer or a regulator, you can see exactly what happened.

Question 23

Does an AI agent need to be trained on our data?

Accepted Answer

No, not in the sense of training your own model. The agent uses an existing language model and is given instructions, examples and access to your documents when it needs them. That makes it quick to get started and easy to change its behaviour by editing the instructions rather than retraining anything.

Question 24

Which tasks are a poor fit for AI agents?

Accepted Answer

Tasks that rarely happen, depend on personal relationships, or where any mistake is very expensive are weaker candidates. In those cases it's usually better for the agent to do the groundwork, such as gathering information and drafting, while a person makes the final call.

Question 25

How much does AI agent development cost?

Accepted Answer

It depends on how many steps and systems the agent covers. A single well-defined job is far simpler than a whole workflow across several systems. After a free review you get a fixed price proposal, and the ongoing cost of the AI model itself is usually small compared with the time saved.

Question 26

How long does it take to build an AI agent?

Accepted Answer

A first agent for a well-defined job is often live within a few weeks. Larger workflows covering several systems take longer, so we build them in stages and put each part into use as soon as it works, rather than waiting for everything to be finished.

Question 27

Which systems can an AI agent connect to?

Accepted Answer

In principle any system with an API, including Gmail, Outlook, HubSpot, Salesforce, Xero, QuickBooks, Teams and Slack, plus many industry-specific platforms. Where there's no ready-made connector, we build one. We check the API's limits and authentication early, so there are no surprises during the build.

Question 28

Can an AI agent use systems that have no API?

Accepted Answer

Often, yes. The agent can operate a web browser, logging in, clicking and filling in forms the way a person would. It is slower and more fragile than an API connection, so we only use it where no better option exists and monitor those steps more closely.

Question 29

Does our data need to be perfect before we start?

Accepted Answer

No. We start with a job where the data is already good enough, and the agent often helps tidy up the rest as it works, for example by flagging duplicate contacts or missing fields. During discovery we point out anything that needs fixing first.

Question 30

Can we start with just one AI agent?

Accepted Answer

Yes, and we recommend it. One agent doing one clear job quickly shows what it's worth in hours saved, and it builds trust in your team. Once that is running smoothly, we add more jobs to the same agent or build further agents alongside it.

Question 31

How do we know an AI agent is paying off?

Accepted Answer

We agree what to measure before we start, such as time per case, time to first reply or cases handled per week, and record a baseline. The agent's run reports then show how those numbers change, so the return is measured rather than assumed.

Question 32

What happens after the agent goes live?

Accepted Answer

We monitor the agent's runs, refine its instructions when something could be better and update its connections when your systems change. You get regular reports on what the agent has done, how often it asked for help and how much time it has saved.

Question 33

Can we build AI agents ourselves?

Accepted Answer

Simple agents can be built in-house with platforms such as n8n or Make, and it's a good way to learn. When an agent has to work across several systems, handle personal data or run reliably every day, it usually pays to bring in people who build and maintain them for a living.

Question 34

Can AI agents work around the clock?

Accepted Answer

Yes. An agent can respond to a new email or web form in the middle of the night and have replied before the office opens. If you prefer, certain steps, such as sending messages to customers, can be held until working hours.

Question 35

What does an AI agent cost to run each month?

Accepted Answer

Running costs are the AI model usage, which scales with the number of cases the agent handles, plus any ongoing support and improvement from us. We estimate both with you before you decide, so you can compare the monthly cost directly with the hours the agent will save.

Question 36

How do we get started with AI agents?

Accepted Answer

Start with a free 30-minute review. We talk through how your team works today, point out one or two jobs where an AI agent would make the biggest difference and outline what a first agent could look like, including the systems it would need to connect to.

Question 37

Can AI agents work alongside Microsoft 365 and Copilot?

Accepted Answer

Yes. Agents can work in Outlook, Teams and SharePoint and complement Copilot rather than compete with it. Copilot helps people while they work; an agent handles the jobs that should happen automatically in the background, such as triaging a shared inbox overnight.

	CaptivateClickAI agents built for you	DIY with ChatGPTYou prompt and paste yourself	Off-the-shelf AI toolsStandard flows in an app	Traditional IT consultancyLarger systems projects
Built around your own workflows	Yes	Partly	No	Yes
Works in your systems, like Xero, HubSpot and Outlook	Yes	No	Partly

AI agents that get work done, not just answer

How AI agents differ from chatbots and normal automation

We find the jobs worth handing over

It plugs into the tools you already use

The agent does the job and shows its working

You approve what matters

Then it runs by itself and reports back

replying to new enquiries

preparing draft quotes

chasing sent quotes

reading supplier invoices

chasing clients for records

writing monthly reports

triaging repair requests

answering tenant questions

writing up inspections

answering order questions

handling returns

sending a daily sales report

preparing for sales calls

logging calls in the CRM

following up new leads

Email and chat

CRM and finance

AI and platforms

AI agents you stay fully in control of

Discovery

Prototype

Build & connect

Running & improvement

An AI agent in production, not a demo

For businesses that want AI to actually do things

You repeat the same work every day

You've tried ChatGPT and want more

You're growing and don't want to hire for admin

Which job would you hand over first?

AI strategy for business: how to go from experiments to results

Xero API: how to connect Xero to your other systems

ChatGPT Enterprise or Business? ChatGPT for business in the UK

What is n8n? A plain-English guide, with n8n vs Zapier, Make and Power Automate

AI Automation

Ready to grow?

AI agents: your questions answered