What is an AI agent for an SME?
At Lindara, it is an assistant designed for a defined business task, with instructions, sources and an output format. It can use information you provide or connect to tools if agreed and validated. Having a chat with a prompt doesn't mean it can already act in your systems.
The starting point is the process: what goes in, what should come out, who reviews it and what it mustn't do. Technology and connections come afterwards.
When it makes sense
- The task repeats and you can describe how to solve it.
- You have examples and information you are allowed to use.
- Someone can check whether the result is correct.
- You want a defined solution, tested before use.
If you mainly want to learn the tools or haven't defined the process yet, a better fit may be individual AI mentoring session.
AI agent, chatbot or automation: what does your business need?
A chatbot is a conversational interface. An automation executes a sequence of steps. An agent can use an AI model to interpret a request and choose tools within defined permissions. These can be combined; what matters is the task they solve and how results are controlled.
| Need | First approach to consider |
|---|---|
| Learn to write or summarise better. | Mentoring and a reusable instruction template. |
| Copy data between applications using fixed rules. | Conventional automation; AI may not be needed. |
| Answer questions about approved documents. | An assistant with traceable sources that clearly states when it cannot find information. |
| Interpret requests and prepare work across several tools. | Consider a connected agent with limited permissions, testing and review. |
At Lindara, we start with a specific SME task. If a template or mentoring is enough, that can be the first step. If development is needed, the proposal specifies connections and expected behaviour.
What we define in the proposal
Goal, users, input data, expected output and excluded tasks.
Instructions, authorised sources and included tools or connectors.
Test cases, errors to detect and acceptance criteria.
How to use it, known limits, human review and maintenance terms.
The proposal specifies what is included. Interface development, integrations, API usage and maintenance aren't assumed to be part of the engagement.
Example: preparing a business proposal
ILLUSTRATIVE EXAMPLE · NOT A CLIENT RESULT
Input: meeting notes, an approved template and available service information. Output: an organised draft, open questions and sections needing confirmation.
The agent must distinguish agreements from missing information. It mustn't invent a price or promise a deadline. Someone reviews the proposal before sending it. CRM access or email drafting connections require specific scope and permissions.
A useful test is to provide incomplete notes: the correct response should flag missing information, not fill it with assumptions.
What to prepare before we talk
Bring a short task description, an anonymised input example and an example of the output you need. List your tools and who will review the work. Don't send passwords, confidential documents or clients' personal data in your first enquiry.
Permissions are limited to the agreed process. An agent that drafts messages doesn't need permission to send them. Integrations must handle errors, duplicates and unconfirmed actions. Validation uses the actual configuration, not a simulated demonstration.
See five questions for choosing your first AI use caseTo define acceptance criteria, use our editable AI test evaluation worksheet: record invented data, corrections and actual review time.
Questions about custom agents
How much does building an agent cost?
There is no fixed public tariff. Pricing depends on the process, data, connections, testing and handover. You'll receive a scope and price before accepting.
Can it send emails or change my CRM?
Only if that action is included, connected with appropriate permissions and tested. It can also be limited to suggested changes or drafts for review.
Is training included?
Handover is specified in the proposal. If you need individual practice to use the solution, we can include mentoring with an agreed duration and price.
Do you guarantee time savings or sales?
We don't promise results that haven't been measured in your process. We define usefulness criteria and review examples before expanding use.