START HERE

What you will learn

  • Choose a sensible first task instead of starting with a tool.
  • Describe what success would look like before building anything.
  • Decide whether a trial is useful enough to continue.

Before you start: No coding knowledge required. Bring one task you already do yourself.

What is a workflow?

A workflow is the sequence of steps that takes a piece of work from its starting point to a finished result. For a research brief, the sequence might be: receive a question, collect sources, compare findings, write a recommendation and check the evidence. AI might help with one step; it does not have to take over the whole sequence.

Start by naming the finished result. “Use AI for research” is too broad to test. “Create a first-draft comparison of three suppliers using five documents I have already checked” tells you what goes in, what comes out and what you will need to inspect.

Which tasks are sensible starting points?

Look for work that repeats, uses information you are allowed to share and produces something you can review before it is used. Draft outlines, comparisons of approved documents and summaries of your own notes are possible candidates.

Avoid beginning with a task whose mistakes would be difficult to reverse, such as sending commitments to a customer or changing live infrastructure without review. You can still explore those areas later, after you understand the steps and controls they require. A first project should teach you something useful at a manageable cost.

Step-by-step process

  1. Write the task in one sentence

    Use this pattern: “When [trigger] happens, I use [input] to produce [output], which [person] checks.” If several unrelated outcomes appear in the sentence, split the work into smaller tasks.

  2. Record how you do it now

    Complete a few ordinary examples and note the time, corrections and quality. This is your baseline: the existing process you will compare against. Include the time spent finding information, not just writing.

  3. Choose the AI-assisted step

    Decide exactly where assistance fits. You might ask for a comparison table while keeping source selection and the final recommendation under your control.

  4. Test comparable examples

    Use the same questions and quality checks for the current and AI-assisted methods. Keep a note of missing facts, invented claims and corrections. A fast draft is useful only if the finished work is acceptable.

  5. Make a continue, change or stop decision

    Continue if the result improves something you care about. Revise the approach if the benefit is unclear. Stop if checking the output costs more than doing the task yourself or if the information cannot be used safely.

FOLLOW ONE EXAMPLE

Worked scenario

A stakeholder preparing a supplier comparison

Imagine you produce a short supplier brief each month. You already have approved product documents and a list of five questions. The following numbers are illustrative, not measured MintBios results.

  1. Your current method takes 50 minutes to draft and 10 minutes to check: 60 minutes in total.
  2. An AI-assisted trial takes 15 minutes to prepare the input, 10 minutes to create a draft and 20 minutes to check and correct it: 45 minutes in total.
  3. The finished brief still answers all five questions and contains no unsupported claims. You repeat the comparison on several other briefs before deciding whether the saving is dependable.

What this shows: In this example, the possible benefit is 15 minutes per brief, not the 40-minute difference between drafting times alone. The review effort is part of the work.

The process at a glance

Compare the complete task before deciding to continue. Checking the finished work comes before the decision, not after expansion.
ProcessDecisionHuman checkpoint
Read Mermaid diagram source
flowchart LR
 A[Choose one task] --> B[Record current effort]
 B --> C[Test one AI-assisted step]
 C --> D[Check the finished work]
 D --> E{Worth continuing?}
 E -->|Yes| F[Repeat and refine]
 E -->|No| G[Change or stop]

What you should have at the end

  • A one-sentence description of the task and its reviewer.
  • A small comparison showing time, corrections and finished quality.
  • A reasoned decision to continue, revise or stop the experiment.

Common mistakes and how to avoid them

Choosing a tool before a problem

A tool demonstration may look impressive without addressing work you actually need to do. Name the task first.

Ignoring correction time

Count preparation, checking and rework when comparing the two methods.

Expanding after one good answer

Try ordinary and awkward examples before relying on the result.

APPLY THE IDEA

Try it yourself

Your task: Write down three tasks you repeat. Choose the one with the clearest inputs and easiest-to-check result.

What to produce: A short task brief: trigger, input, output, reviewer, baseline and the improvement you want.

Check your understanding: Could another person understand what you will test and how you will judge it without seeing the AI tool?

Key terms in plain English

Baseline
A record of how the task performs before you change it.
Pilot
A small trial used to learn whether an approach is worth continuing.
Acceptance check
A specific condition the finished result must meet.