AI agent, chatbot, RAG assistant or rule-based automation? What each does, agentic vs generative AI, when each pays off, and UK costs from £20,000.
Short answer. A chatbot answers questions. A RAG assistant answers from your own documents and cites the source. An AI agent does the work: it plans the steps, calls your systems, checks the result and hands over to a person when it should. Rule-based automation follows fixed if-then steps between systems. Use rules when the steps never change, a RAG assistant when people need accurate answers from your content, and an agent when multi-step work varies from case to case. At Magora an agent for one workflow starts from £20,000, and a RAG assistant from £30,000.
This guide is for business owners and operations leaders who have been told they need "AI agents" and want to know what that means in practice. It explains the four options, the difference between agentic and generative AI, where each pays off, how to keep an agent safe and what it costs.
Last checked: 2 November 2026.
The four options compared
| Chatbot | RAG assistant | AI agent | Rule-based automation | |
|---|---|---|---|---|
| What it does | Answers questions from a script or the model's general knowledge | Answers from your own documents and cites the source | Plans the steps, calls your systems, checks the result, hands over to a person | Follows fixed if-then rules between systems |
| Acts in your systems? | No | No: it reads and answers | Yes, within the permissions you grant | Yes, only the steps written in advance |
| Best for | Simple FAQs and first-line questions | Accurate answers from private or changing content | Multi-step work that varies from case to case | Stable, predictable steps |
| Magora starting point | Scoped after the free prototype | From £30,000 | From £20,000 for one workflow | If rules are all you need, we say so |
Source: Magora AI agent development page, 2026 prices excluding VAT.
Agentic AI vs generative AI, in plain English
Generative AI produces content in response to a prompt: a draft email, a summary, an image, code. It is one step: you ask, it answers.
Agentic AI uses generative models as one part of a system that works towards a goal over several steps. It decides what to do next, uses tools such as your CRM's API or a search over your documents, keeps track of what it has done, checks the result and stops or asks for help when it should.
A practical example: a generative model can draft a reply to a customer asking to move a delivery. An agent reads the email, finds the order, checks the delivery slots, changes the booking, confirms by email and logs the change, and passes the case to a person if the order is already dispatched.
Where each one pays off
| Business function | Typical tasks | Usually the right tool |
|---|---|---|
| Finance and back office | Invoices, claims, contracts and forms read, checked against your rules, summarised and filed | Document agent |
| Customer service | Look up the order, change the booking, escalate to a person with the whole conversation attached | Customer service agent |
| Sales and operations | Reconcile the CRM with the ERP, raise the ticket, update the record, chase the missing field | Workflow agent, or rules if the steps never vary |
| People, IT and internal questions | Staff questions answered from policies, tickets and manuals, with citations | RAG assistant |
| Contact centres | Live prompts for people during a call | Real-time assistant, as in Empath AI inside Microsoft Teams |
| Website first line | Opening hours, prices, simple FAQs | Chatbot |
Based on the use cases on our AI agent and AI automation pages.
Five questions to decide
- Does the answer live in your own documents? If people mainly need accurate answers, start with a RAG assistant.
- Does the work involve taking actions in other systems? If yes, you need an agent or automation, not a chatbot.
- Do the steps vary from case to case? If they never vary, use rules. If they depend on reading free text or documents, an agent pays off.
- What does a mistake cost? The higher the cost, the more human sign-off and testing the agent needs.
- Is there enough volume? Agents pay back on repeated, high-volume work that people now do by reading and re-typing.
How to keep an AI agent safe
- Least permissions. The agent can only do what it needs to, in the systems it needs.
- Human sign-off on actions that matter, at least until the results prove reliable.
- Audit log of every tool call and decision, and alerts on every failure.
- Evaluations: a test set of real cases with known right answers, run automatically on every change, then a supervised pilot of two to three weeks in which your team approves or corrects the output.
- Data protection: a DPIA where personal data is involved, and clear rules on what data the model provider may see.
Not sure your organisation is ready? Use our checklist is your business ready for an internal AI agent?
We build AI assistants, automation and machine-learning features into real products, from proof of concept to production. First release in 6–8 weeks; developers from £49/hour.
What each option costs
| What you buy | Typical time | Magora price (GBP) |
|---|---|---|
| Free working prototype on a sample of your data | 5–10 working days | £0 |
| Discovery: data, approach, model choice, fixed estimate | 2 weeks | From £3,000 |
| AI proof of concept: a controlled pilot on real data, accuracy measured | 4–6 weeks | From £10,000 |
| AI MVP: a first release with AI at its core | 6–8 weeks after Discovery | From £10,000, most £10,000–£25,000 |
| AI agent automating one internal workflow | 6–8 weeks after Discovery | From £20,000 |
| Fine-tuned open-source model for your domain | Scoped in Discovery | From £25,000 |
| Knowledge assistant (RAG) over your documents, with citations | 6–8 weeks after Discovery | From £30,000 |
| Production LLM application, customer-facing or multi-agent | 3–6 months for multi-agent systems | From £40,000 |
| Adding AI to an existing application | Depends on scope | £20,000–£80,000 |
| Support after launch | Monthly | From £1,000 a month |
Magora 2026 prices, excluding VAT, as published on our AI and ML development, AI agent, RAG and AI proof of concept pages. A fixed price is quoted after Discovery.
For a fuller breakdown, including running costs, see AI development cost in the UK.
Not sure which one you need? Send us one workflow and a few real examples: we will build a free working prototype in 5–10 working days and tell you honestly whether it needs an agent, a RAG assistant or plain rules. More about our AI agent development.
Frequently asked questions
What is an AI agent?
Software that uses a language model to work towards a goal: it plans the steps, calls your systems through tools, checks the result and hands over to a person when it should. Unlike a chatbot, it acts in your systems within the permissions you grant.
What is the difference between an AI agent and a chatbot?
A chatbot answers questions. An AI agent does work: it looks up the order, changes the booking, updates the record or files the document, and checks that the step succeeded. A RAG assistant sits between them: it answers from your own documents and cites the source, but does not act.
What is agentic AI?
Agentic AI describes systems that pursue a goal over several steps with some autonomy, using tools and memory, rather than producing a single response to a single prompt.
What is the difference between agentic AI and generative AI?
Generative AI produces content such as text, images or code in response to a prompt. Agentic AI uses generative models as one component in a system that plans, takes actions in other software and checks the results.
When is rule-based automation better than AI?
When the steps are stable and predictable and the inputs are structured. Rules are cheaper to build and run, and fully predictable. AI is worth it when inputs vary, such as free-text emails or documents in different formats.
How much does an AI agent cost in the UK?
At Magora an AI agent automating one internal workflow starts from £20,000 and typically ships 6–8 weeks after a two-week Discovery. A knowledge assistant over your documents starts from £30,000. A free working prototype comes first.
Alexander Bessudnov
Alexander is an accomplished Chief Product Owner with a proven track record of leading successful product development initiatives at Magora. With a passion for innovation and a keen understanding of market trends, Alexander plays a pivotal role in shaping Magora's product strategy and ensuring the delivery of cutting-edge solutions to clients.
LinkedIn