Guide
Which tasks to automate, and which not to build an agent for
For this guide and its prompts we recommend:
Offers checked 21 Oct 2026
You can now sort tasks: automate, template, keep human.
A scored list of twenty tasks showing which to automate, which to turn into a template and which to keep human.
What you need
- A time tracker, or a notebook, for one week
- A spreadsheet and any AI chat assistant
- A document for the templates
- Gamma
Recommended AI tools
GammaA laid-out deck or page from an outlineFree plan
Steps
Track one real week
Log every task for a week: what, how long, how often. Guessing from memory misses the small tasks that add up.
Score twenty tasks on four questions
Score each task 1 to 3: does it repeat, does it follow fixed rules, how cheap is a mistake, how much time does it take. High repeat and fixed rules with a cheap mistake points to automation.
ExampleCopying web orders into the order sheet: repeats daily, fixed rules, a mistake is easy to spot. Automate. Replying to a complaint: needs judgment, a mistake is costly. Keep human.
Choose a template before an agent
Many tasks need a person but follow the same pattern: a quote, a reply to a common question, a weekly update. A good template with blanks saves much of the work and needs no agent to look after.
WeakBuild an agent that writes all my quotes.
BetterA quote template with blanks for client, item, quantity and dates, plus a prompt that fills it from my notes for me to check.
Keep some tasks human on purpose
Anything with money, legal or health weight, sensitive client data or a relationship at stake stays with a person, even if a tool could do it. Write down why for each one.
Show the plan to the team
Turn the scored list into a few slides: automate, template, keep human, and who owns each. Agree on one task to start with. For building a first small agent, see /guides/first-ai-agent-for-one-repetitive-task/.
The full instruction
Copy it into any AI chat assistant and change only the first line.
Change only this line: BUSINESS = [a bike repair shop] Act as an honest operations adviser who is not selling automation. My business is my BUSINESS. Below I paste tasks from one tracked week: task, minutes, how often. Do three things: 1. Score each task 1-3 on repeat, fixed rules, cost of a mistake and time, and sort them into three groups: automate, template, keep human. 2. For the template group, say what each template would contain. 3. For the keep-human group, give the reason in one line. Then argue against your own sorting: name two tasks you may have put in the wrong group and why. Use only my tasks and times. Start with the scoring table. Paste your material below this line:
What you end up with
Twenty tasks sorted on evidence from a real week: a few to automate, several to turn into templates and the ones that stay human for a written reason, shared with the team.
Common mistakes
- Building an agent for a task a simple template would cover.
- Scoring tasks from memory instead of a tracked week.
- Automating tasks with money or sensitive data because the tool allows it.
Processes that need more than one tool
Each process below uses several tools, each for the job it does best. Any AI chat assistant works where one is named.
List a week of tasks
Tools, in order: a time tracker → a spreadsheet
- Track every task for one week in a time tracker with short, consistent names.
- Export the week to a spreadsheet and merge tasks that have the same name.
The prompt for this process
Help me clean up one week of tracked tasks before I score them. Export from my time tracker: [paste: task name, date, minutes] What my business does: [one sentence] Merge entries that are the same task under different names and give each task one clear name. Return a table: task, times per week, total minutes, average minutes each time. Flag entries too vague to score, such as "admin", and ask me what they were. Keep the 20 tasks with the most total minutes. Start with the merged table.
Score them
Tools, in order: a spreadsheet → any AI chat assistant
- Paste the cleaned task table from the spreadsheet into the chat assistant with what a mistake would cost.
- Copy the scores and groups back into the sheet and change any score you disagree with.
The prompt for this process
Act as a strict scorer. Score my tasks for automation. Tasks: [paste table: task, times per week, total minutes] What a mistake in each task would cost: [notes] Score each task 1 to 3 on: repeats, fixed rules, cheap mistake, time used. Add up the scores and sort into automate (high total and low risk), template (needs a person but follows a pattern) and keep human (money, legal, sensitive data or a relationship at stake, whatever the score). Then review your list once and move any task whose risk you underrated. Start with the scores.
Make a template instead of an agent
Tools, in order: any AI chat assistant → a document
- Give the chat assistant three past examples of one repeated task.
- Save the template and its fill-in prompt in a document and use it on the next real case.
The prompt for this process
Turn one of my repeated tasks into a reusable template instead of an agent. Task: [e.g. writing a quote after a call] Three past examples: [paste, without client personal data] Find what stays the same and what changes. Write a template with a [blank] for each changing part and short guidance under each blank. Then write a one-paragraph prompt I can reuse to fill the template from my notes, which always leaves the result for me to check. Format both for a document. Start with the template.
Show the plan to the team
Tools, in order: a spreadsheet → Gamma
- Copy the three groups from the spreadsheet.
- Build a short deck in Gamma and agree on the first task with your team.
The prompt for this process
Help me present my automation plan to my team in a few Gamma slides. Scored list: [paste the three groups from my sheet] Team: [who, and what they do] First task we will start with: [task] Write at most 6 slides: why we looked at this, automate, template, keep human and why, who owns each item, and the first step with how we will check it after a month. Use only my list. No promises of time saved that the list does not show. Start with slide 1.
For this result we recommend these tools
This guide describes a result. These tools are one way to get there faster. Whether one pays for itself depends on your work, and we promise no result.

From Ema’s story: Ema designs invitations that spell correctly
ToolGamma
AI tool for creating presentations, documents and simple websites from prompts.
From $9 per seat per month, $108 per seat billed annually (Plus) (source, checked 2026-10-07)
Gamma is an AI tool that builds slide decks, one-page documents and simple web pages from an outline. In this guide it is used in “Show the plan to the team”. People use it for proposals, lesson slides and reports, and it stands out because the result is shared as a link that reads well on a phone.
Free plan (source, checked 21 Oct 2026)
We recommend a specialised tool for each process rather than one tool that does everything. Tools that try to do everything rarely do any one job really well, because no single function gets perfected. It is the same in life, in business and in relationships: spread your attention over ten places and nothing gets done properly.
Forward, without brakes
Now that the Kailiniai Heroes exist, you can have your own. Create one hero or several, and use them for goals that used to need a whole team or a big budget.
