1. Which task do you want to solve exactly?
“I want to use AI in my business” is a useful starting interest, but not yet a brief. Choose a task whose start and finish you can describe: summarising meeting notes, drafting a proposal or finding an answer in internal documents.
Write this sentence: “When I receive [input], I need [output], reviewed by [person].” If you can't complete it, organise the process before automating it.
A bounded first task lets you check behaviour through examples. Avoid starting with irreversible decisions or an agent expected to manage the entire business without criteria.
2. Is the input information available and usable?
Gather an input example and a correct output example. Remove unnecessary names, private amounts and client data. Check which documents are current and who may access them.
An agent can't fix contradictory records on its own. If two documents give different instructions, first decide which is valid. If information sits in several applications, connecting them is part of the technical scope, not something to assume.
A process description and anonymised examples are enough for your first conversation with Lindara. You don't need to share passwords or a full export of your systems.
3. How will you know the result is correct?
Define the review before building. A summary should preserve decisions and separate open questions. A proposal mustn't invent prices or commitments. A document answer should identify its source and acknowledge missing information.
Test incomplete inputs too. A useful system should flag gaps. Fluent or convincing wording alone doesn't prove that the facts are correct.
Decide who reviews results and which actions need confirmation. Drafting, sending and modifying a record are different actions with different permissions.
4. Would mentoring or an agent suit you better?
| Your situation | Starting point |
|---|---|
| You want to learn a tool and review its responses. | 1:1 mentoring with practice on a task. |
| The task is defined, repeats and needs reusable configuration. | Consider an agent with a defined scope and tests. |
| You need a solution and want to learn how to use it. | Agent and mentoring, each with a clear scope. |
| You don't know the input, output or reviewer. | Clarify the process first. Don't start with an integration. |
Mentoring can help you make a useful template without development. A connected agent may fit when a task needs specific sources or tools. Choose based on your process, not on which option is most complex.
5. What would justify continuing?
Agree a small test and what to evaluate: output quality, corrections, review time and limits. If comparing times, measure the same task and include preparation and review. Don't assume apparent time savings are a real improvement without checking.
Before commissioning, clarify the goal, deliverable, tools, price, external costs and maintenance. If the test fails the criteria, correct or narrow the scope rather than expanding automation by default.
Your brief in five lines
The task I repeat is…
The information I use is…
A good result would include…
It would be reviewed by…
I want to learn to do it / have an agent / both.
Explore what's included in AI agents for businesses or how our sessions work with individual mentoring at Lindara.
TRY A SMALL TASK
Practical example: turn meeting notes into clear tasks
This fictional exercise lets you practise a useful skill before connecting any tool: distinguishing confirmed information from open points. Try it in your AI tool and compare the response with the criteria below.
1. Start with these notes
Preparing a proposal to renew the catalogue is approved. Marta will write the draft. No delivery date has been agreed. Luis suggests publishing on Friday, but nobody confirms the date. The budget will be decided after reviewing the draft.
2. Give a specific instruction
Ask it to separate confirmed decisions, pending proposals and tasks. Each task should show an owner and date, use “to be confirmed” for missing data and end with open questions. It must use only the notes and take no external actions.
Download the exercise and instructions3. Check the result, not just the wording
| Item | Expected result |
|---|---|
| Decision | Prepare a proposal to renew the catalogue. |
| Confirmed task | Marta writes the draft; date to be confirmed. |
| Publication on Friday | A proposal awaiting confirmation, not a commitment. |
| Budget | To be decided after reviewing the draft; no amount given. |
Try a variation: remove Marta's name. The response should mark the owner as “to be confirmed”. If it keeps the name or assigns Luis, it has added information missing from the notes.
This exercise doesn't prove a tool's reliability for every case. It teaches how to define a task and spot a specific error. In an individual AI mentoring session we can adapt the practice to a task in your business.
How to check whether a first AI test is worthwhile
Record examples of varying difficulty, including missing or contradictory data. Before testing, define errors that would make the output unusable. Then note what it preserved, what needed correcting and how long review took.
For a time comparison, add preparation, generation, review and corrections. Compare with the same task without AI. A faster draft that takes longer to correct hasn't yet demonstrated improvement.
Download evaluation worksheetThe worksheet downloads as editable text and sends no data to Lindara. If the process repeats and criteria are clear, use it to prepare an enquiry about a custom agent and its acceptance tests.