A key principle: AI is not an autopilot but a “co-pilot”. It speeds up the rough drafting, helps you structure things, suggests options and checks the logic. But you are the one responsible for accuracy, facts and the final tone.
The basic mindset: AI is not an autopilot but a “co-pilot”
AI works well where there is a clear goal and clear quality criteria. Give a model a “fuzzy” task (no context, no constraints, no format) and the answer will almost always be either generic or off the mark.
So it helps to think in three stages:
- Set the frame: the goal, the audience, the AI’s role, the output format, the constraints and the quality criteria.
- Produce a draft: structure, key points, alternative wordings, a list of questions, an action plan.
- Finish it by hand: check the facts, even out the style, remove the “AI-sounding” phrasing, check the risks and the context.
That way you gain speed without losing authorship.
Without this setup, everything misses the mark: your context and tone rules
Quality jumps sharply if you describe, once:
- who you are and what you do;
- what decisions you make;
- which constraints matter (deadlines, responsibility, confidentiality);
- what communication style you consider normal;
- how the AI should act when it lacks information.
You can keep this as a short “passport” and paste it at the start of complex tasks.
The “passport” template (paste it in and reuse it)
- Role and context: who I am, what my job is, what I am responsible for.
- Goal: what exactly I want to end up with.
- Constraints: deadlines, risks, budget, what must not be done or written.
- Tone: brief and to the point / friendly / strictly businesslike / no bureaucratic language.
- Clarification rule: if information is missing, first ask 3–7 clarifying questions.
- Facts rule: do not make up numbers or names; flag the places that need checking.
It also helps that many tasks can be “dictated”, especially when you need to gather your thoughts quickly and offload the context.
1) Email + attachments: “read it in a minute, reply in 10–30 minutes”
A classic case: a long email arrives with several attachments. Done by hand, this easily turns into an hour or two of reading, and sometimes a whole day if the subject is complex.
AI can be used as an accelerator:
- quickly pick out the gist;
- compile a list of actions;
- find deadlines and risks;
- suggest clarifying questions;
- prepare a draft reply.
The main thing is not to ask for “the perfect reply straight away”. First you need a diagnosis and a structure.
Prompt 1 — “Pull out the essentials”
Analyse the email and the attachments.
- The gist of the request in 2–3 sentences
- What is needed from me (as a list of actions)
- Deadlines/risks/dependencies
- What to clarify with the sender (3–7 questions)
- A draft of a short reply: acknowledgement + next step
Prompt 2 — “A reply in my own words” (so it doesn’t sound like AI)
Write a businesslike reply in plain, natural language.
Remove stock phrases and bureaucratic language.
Keep the meaning and add my points: (insert 3–5 points).
Practices that genuinely help
- First ask for the structure and the questions, then for the text of the email.
- Do the final edit by hand: 1–2 minutes to make the tone “yours”.
- If the email is confrontational, ask for 2–3 versions of the reply with different degrees of firmness.
2) A large document: a “2–3 page strategy” from dozens of pages of input
The situation: you have a lot of material — reports, research, notes, correspondence, internal documents. And you are expected to produce something short: a strategy, a concept, a 2–3 page memo.
A working approach:
- Gather the inputs: upload the materials and give the general context of the task.
- Ask for a structure: a 2–3 page outline, alternative structures.
- Write in blocks: one section at a time, so the logic holds.
- Bring in a “critic”: ask it to tear the document apart from the point of view of the person who will approve or fund it.
Prompt — “Strategy skeleton”
Goal of the strategy: …
Audience (who approves it): …
Constraints (deadline/budget/risks): …
Based on the attached materials:
- Propose a 2–3 page structure (sections + 1–2 key points per section)
- Flag what data is missing (as a list)
- Give 2 alternative structures: a conservative one and a bold one
Prompt — “Writing in blocks”
Write only section No. 1, “The problem”.
Use facts from the materials.
At the end: 5 questions that would strengthen the section.
Prompt — “Harsh critic mode”
Imagine you are the person deciding whether to fund the project.
Tear the document apart:
• what is unclear
• where it is not backed by data
• where the logical gaps are
• what will draw objections
Give a list of edits by priority: P0/P1/P2.
This step is especially useful: even a good draft often looks convincing “inside the team” but weak to an outside reader or a manager.
3) Presentations: AI helps you make strong material, but the structure is yours
“Slides from text” generators often produce odd results: the slides don’t hold the logic, add things nobody asked for and break the meaning.
The working principle: first the script and the structure, then the design.
How to put together a presentation with AI
- Formulate one main message (what the audience should understand or do).
- Ask for a structure of 8–12 slides.
- For each slide, get:
- a title;
- the key point;
- what to show (chart/table/diagram/example);
- speaker notes (what to say);
- the risk of misinterpretation (where the audience might “go off track”).
- Take a template and assemble it by hand — that way you control the style and the visual discipline.
Prompt — “Presentation plan”
Outline a 10-slide presentation.
For each slide: the goal, 1 key point, what to show (chart/table/diagram) and 2–3 points for the speaker.
Style: brief, no filler.
What improves the result
- Ask the AI not to embellish but to sharpen the meaning: where a number is needed, where an example, where a comparison.
- Ask it to play the “auditor”: which slides are weak, where there is no evidence, where the conclusion doesn’t follow from the data.
4) Repetitive processes: when it’s time to think about automation
Some tasks constantly “eat up” time:
- filling in standard forms;
- moving data between systems;
- gathering information and reporting;
- reconciling lists and spreadsheets;
- preparing similar messages.
If a scenario repeats, it can be partly or fully automated.
How to tell a task is ready for automation
- Repeatability: the same inputs give similar outputs.
- Checking rules: you can describe the criteria for “done/not done”.
- Control: at the last step there is a person who confirms the result.
- Risk level: a mistake does not lead to irreversible consequences.
5) Training materials and internal policies: you can speed them up, but not “get around” them
Formal courses and mandatory training often take up time, especially when they are bureaucratic. Here AI is useful in a legitimate role:
- make a summary;
- put together a checklist;
- explain the difficult parts in plain language;
- prepare practice questions for self-testing;
- help draw up a plan for building the requirements into everyday work.
Prompt — “Boil it down”
Write a summary of the material:
• 10 key rules
• 5 typical mistakes
• 5 “how to do it right” situations
Then create 12 practice questions with explanations.
That way you save time and actually understand the material, rather than just “getting through” it.
6) Confrontational correspondence: “cool down”, reply diplomatically, don’t wreck the relationship
One of the most practical scenarios: someone writes to you sharply, you get wound up and want to reply in kind — and the conflict escalates.
AI can be used as a “buffer”:
- turn the emotion into a task;
- clarify your goal;
- suggest wording that doesn’t attack;
- help you set boundaries and name the next step.
The key formula: “goal + boundaries + a concrete next step”.
Prompt — “A diplomatic reply”
Here is my colleague’s message: …
My context: …
My goal:
- set boundaries
- agree on the next step
- preserve the working relationship
Write 3 versions of the reply: soft / neutral / firm but polite.
Remove bureaucratic language and “AI-sounding” phrasing.
The final polish
- Remove words that sound “preachy”.
- Shorten long constructions.
- Add one human line (no sarcasm): “Got it”, “OK”, “Let’s do it this way”, “I agree with such-and-such”.
Final quality check: a short checklist before sending
The most common mistake is taking the first answer as final. Even if the text “looks good”, it may:
- get details wrong;
- promise too much;
- sound unnatural;
- ignore internal policy or risks.
Before sending:
- Check the facts, numbers, names and wording.
- Make sure there are no unnecessary promises.
- Bring the style in line with your own human tone.
- If the subject is sensitive, cut everything superfluous.
- If there is a risk, add a clarifying question instead of a statement.
A mini set of follow-ups for prompts
1) Quickly grasp the gist
Pull out the essentials from the text:
• 3 key ideas
• what needs to be done
• what to clarify
2) Make a plan
Draw up a 5–10 step action plan.
For each step: the result, the input data, the risk.
3) Give alternatives
Suggest 3 solutions: a quick one, a balanced one, a robust one.
List the pros and cons.
4) Check the logic
Find the weak spots: where the conclusions don’t follow from the facts, where data is missing, where there will be objections.
Summary
AI becomes genuinely useful when you:
- give it a clear frame;
- get a structure and questions;
- write in blocks;
- bring in a “critic”;
- always do the final edit.
That way you are not “handing the work to a machine” but speeding up your professional output — while keeping your style, responsibility and quality.
Useful articles and discussion here (chat in Russian)
Where delegation should stop
Drafts, searching for options and repetitive transformations are easy to hand over to AI. Decisions where a mistake is costly — money, access rights, legal wording and promises to clients — should stay with a human and be checked against the source data.
