# AI-Powered Founder A working manual for founders. Build a personal AI operating system for research, decisions, writing, meetings, planning, delegation, and automation without expanding your team. ## Course format 9 chapters, 12 lessons, and a final working project. ## FAQ ### Who is AI-Powered Founder for? It is for founders who want to use AI in the work they already own: research, decisions, writing, meetings, planning, delegation, and recurring workflows. ### What will I build in this founder AI course? You will assemble a founder AI operating system: reusable workflows, a context pack, an AI-assisted decision process, a weekly review, recurring automations, and clear control rules. ### Do I need to code to take this course? No. The manual is written for founders and uses practical prompts and working exercises instead of code. ## Course lessons A working manual for founders ## AI-Powered Founder Build a personal AI operating system that helps you think better, decide faster, communicate clearly, and execute more without expanding your team. Course outcome Founders leave with AI integrated into their daily work, reusable workflows, and a weekly operating system they can continue using. How to read it Nine chapters, one scroll. Each ends with a prompt you can copy straight into your own tools and one thing to do today. Start chapter 01 See all chapters Scroll, or press 1–9 to jump. Your place is saved. 01 ### Stop Using AI Like a Search Box 12 min read · one prompt to keep Most founders ask AI questions. The ones who get real leverage give it a role, a brief, and a standard, then judge the work like a manager rather than a user. In this chapter Use AI as a researcher, analyst, critic, coach, and execution partner Define the goal, context, constraints, and quality standard Choose when AI should advise, create, challenge, or act A question gets you an answer. A brief gets you work you can use. The difference is four sentences of setup. #### Five roles, not one assistant Name the role before you type. Each one wants a different brief and produces a different kind of output. Researcher Finds and sources. Asks for citations, flags what it could not verify. Analyst Turns material into a conclusion with the reasoning shown. Critic Attacks your plan. Only useful if you ask it to be specific and unkind. Coach Asks you questions instead of answering. Best before a hard conversation. Partner Executes: drafts, builds, files, follows up. Needs your standard in writing. #### The four-part brief Goal What decision or artifact this produces. Not the topic, the outcome. Context Company, stage, customer, what you already tried, what you have decided. Constraints Budget, headcount, timeline, tone, what is off the table. Standard What "good" looks like, in a sentence you would use with a new hire. #### Advise, create, challenge, or act Pick one per request. Asking for advice and a finished draft in the same message gets you a confident draft built on advice you never evaluated. When you want to be challenged, say so explicitly, because the default behaviour of every model is to agree with you. Prompt to keep · the founder brief Copy prompt Act as my {researcher | analyst | critic | coach | execution partner}. GOAL — the decision or artifact I need: ... CONTEXT — company, stage, customer, what I already tried: ... CONSTRAINTS — budget, time, tone, what is off the table: ... STANDARD — good looks like: ... Mode: {advise | create | challenge | act} — do only this. Before you start, ask me the 3 questions that would most change your answer. Do today Take the last three things you asked AI for. Rewrite one as a four-part brief and compare the output. Keep the better version in a notes file called Briefs . 02 ### Build Your Founder Context System 14 min read · builds your context pack The reason AI output feels generic is that you keep starting from zero. A context system is the set of documents that make every future request short. In this chapter Organize your company, customers, strategy, preferences, and previous decisions Give AI the right documents and background before asking for work Create reusable context instead of explaining everything repeatedly Write it once, at the quality you would give an incoming chief of staff. Seven files. A weekend to draft, ten minutes a week to maintain. #### The seven files File Holds Refresh Company What you sell, to whom, pricing, stage, numbers Monthly Customer ICP, jobs to be done, objections, real quotes Quarterly Strategy This year's bets, what you are deliberately not doing Quarterly Voice How you write, three samples you are proud of, words you never use Rarely Decisions What you chose, why, what you would revisit Weekly Standards What good looks like per artifact: email, deck, spec, post Rarely Open loops Live questions, blockers, who owes whom what Weekly #### Rules that keep it usable One page each If a file grows past a page, you are keeping history, not context. Move the history elsewhere. Facts with dates Numbers age. Stamp them, so the model can tell you when its input is stale. No secrets in the pack Keep personal data, legal terms, and credentials out. Chapter 09 covers what stays offline. Prompt to keep · draft the pack from what exists Copy prompt Here are raw materials about my company: {paste site copy, deck, notes, 3 customer emails}. Draft my context pack as 7 separate one-page files: Company, Customer, Strategy, Voice, Decisions, Standards, Open loops. Rules: facts only from what I gave you; mark anything you inferred as [INFERRED]; stamp every number with a date; list at the end the 10 questions you need answered to remove the inferences. Do today Write the Voice file first. Paste three things you wrote yourself, then ask AI to describe your rules back to you. Correct it once, and keep that description. 03 ### Research and Learn Faster 13 min read · one research protocol Research is where AI is fastest and least trustworthy at the same time. The protocol below is designed so speed does not cost you accuracy. In this chapter Research markets, competitors, customers, technologies, and opportunities Turn large amounts of information into useful conclusions Verify sources and separate evidence from AI assumptions Ask for the finding, the source, and the confidence. Three columns, every time. Unsourced claims are hypotheses, and should be labelled as such. #### A four-pass protocol Pass 1 Map the question Ask AI what would have to be true for your assumption to hold, and what evidence would settle it. You are buying a search plan, not an answer. Pass 2 Gather with sources Findings in a table: claim, source, date, confidence. Anything without a source moves to a separate list called Assumptions. Pass 3 Synthesise against a decision "Given this, what would you do and what would change your mind?" A summary you cannot act on is entertainment. Pass 4 Spot-check the load-bearing facts Verify the two or three facts the decision rests on, by hand, at the source. Not all of them. The ones that carry weight. #### Where research goes wrong Harvard's field-experimental work on the "jagged frontier" found that AI can substantially improve knowledge work when the task fits its capabilities, and can reduce performance when users trust it outside those boundaries. Research sits on both sides of that line: synthesis is inside it, obscure private-market facts are usually outside. Inside the frontier Summarising documents you supply, structuring interviews, comparing public positioning, drafting frameworks, translating jargon, generating hypotheses to test. Outside it Private company numbers, recent funding, niche pricing, regulation specifics, anything where being 90% right is the same as being wrong. Prompt to keep · four-pass research Copy prompt Research question: ... Decision it feeds: ... Pass 1: tell me what would have to be true for my assumption to hold, and what evidence would settle it. Pass 2: give findings as a table — claim | source + date | confidence high/med/low. Anything unsourced goes in a separate list titled ASSUMPTIONS. Pass 3: given the findings, what would you do, and what would change your mind? Pass 4: name the 3 facts my decision rests on, so I can verify them myself. Do today Run the protocol on a question you have been avoiding because it felt like a week of work. Stop at pass 3 and see whether you can already decide. 04 ### Make Better Decisions 15 min read · your decision process AI is at its most valuable before the decision, not at it. Used well it widens your options and finds the flaw you were motivated to miss. It never takes the accountability. In this chapter Generate options instead of accepting the first recommendation Challenge assumptions and expose blind spots Run scenarios, pre-mortems, risk analysis, and decision reviews Keep final judgment and accountability with the founder Never ask "what should I do?" first. Ask for five options you have not considered, then argue with them. The first recommendation is an anchor, and anchors are expensive. #### Five moves, in order Widen Ask for six distinct options, including one that costs nothing, one that is reversible, and one you would be embarrassed to propose. Stress Have it list your unstated assumptions, then rank them by how much damage a wrong one does. Pre-mortem "It is twelve months out and this failed. Write the post-mortem." Then fix the two causes you can cheaply remove now. Decide You choose, in writing: the call, the reason, the reversal cost, and the date you will review it. Review On that date, feed the original entry back in and ask what you got wrong and why. This is how the Decisions file earns its keep. #### One-way and two-way doors Two-way door Cheap to reverse: pricing tests, channel experiments, copy, tooling. Decide in an hour with AI as your analyst, and move. One-way door Hard to undo: hires, fundraising terms, category choice, key partnerships. AI prepares the case; you sleep on it and sign it. Prompt to keep · decision partner Copy prompt Decision: ... Context files attached: Company, Strategy, Decisions. 1. WIDEN — give me 6 genuinely distinct options, including one that costs nothing, one fully reversible, and one I'd be embarrassed to propose. 2. STRESS — list the assumptions I am making without saying them, ranked by damage if wrong. 3. PRE-MORTEM — it's 12 months on and this failed: write the post-mortem. 4. Tell me whether this is a one-way or two-way door, and what it would cost to reverse. Do not tell me what to do. I decide, then I'll ask you to review it on a date I set. Do today Write a pre-mortem for the biggest thing on your plate this quarter. Ten minutes. Then remove one cause of failure before lunch. 05 ### Create and Communicate 13 min read · your voice at volume The goal is not to sound like AI wrote it faster. It is to get what is already in your head out at the quality you would reach on a good day, every day. In this chapter Draft emails, proposals, presentations, strategies, and reports Preserve your own voice, judgment, and standards Turn rough thoughts, conversations, or voice notes into finished work Talk first, edit second. Dictate the mess, then ask for structure, never for opinions you did not have. Blank-page prompting is what produces generic writing. #### The dictate-shape-cut loop Step 1 · dictate Two minutes of unedited voice note. Everything you know, in the wrong order. This is your content. Step 2 · shape Ask for structure and gaps against your Voice and Standards files. Same words, better order. Step 3 · cut Ask what can be deleted without loss, then delete more than it suggests. You send it. #### Tells that give you away Machine tell What you do instead Three-item lists everywhere Two items, or seven. Real thinking is uneven. "It's not just X, it's Y" Say the thing once, plainly. Confident claims with no number Add the number, or drop the claim. Praise before the answer Open with the ask or the decision. Prompt to keep · voice note to finished draft Copy prompt Attached: my raw voice note / rough notes. Also attached: my Voice file and Standards file. STEP 1 — restructure MY words into {email | proposal | update | post}. Add no new claims, no new opinions. STEP 2 — list what's missing or unclear that only I can answer. STEP 3 — mark anything that reads like generic AI writing and rewrite it in my voice. STEP 4 — show me a version 30% shorter. Never add praise, hedging, or 'it's not just X, it's Y' constructions. Do today Pick the message you have been putting off writing. Dictate it badly, run the loop, send it within twenty minutes. 06 ### Run Meetings and Relationships 12 min read · briefing and debrief pair Founder meetings fail at the edges: nobody prepared, and nobody wrote down what was agreed. Two small habits, one before and one after, fix most of it. In this chapter Prepare agendas, questions, and background briefings Capture decisions, commitments, and follow-ups Prepare for sales calls, negotiations, investor meetings, and difficult conversations Walk in with one page: who they are, what you want, the three questions, and your walk-away line. Ten minutes of prep, and the meeting is half won. #### The one-page briefing Who Background, incentives, what they are measured on, last interaction. Outcome The one sentence you want true when the call ends. Questions Three, ordered so the answers build. Written, not improvised. Objections Their likely three, and your honest answer to each. Limits What you will not agree to today, decided before you are in the room. #### Rehearse the hard ones For a negotiation, a difficult performance conversation, or a tense investor update, have AI play the other side at their most reasonable, then at their most difficult. Two rounds. You will find the sentence you cannot yet say out loud, which is exactly the one to prepare. Prompt to keep · briefing and debrief Copy prompt BEFORE — brief me on one page for a meeting with {name, role, company}: who they are and what they're measured on · the one outcome I want · 3 questions in build order · their likely 3 objections with honest answers · what I will not agree to today. Then role-play them: first at their most reasonable, then at their most difficult. AFTER — here is my 3-minute debrief: {paste}. Extract: decisions made · commitments (mine / theirs, with dates) · open questions · the follow-up email draft. Do today Brief your next external meeting on one page. After it, dictate three minutes of debrief and let AI extract decisions, commitments, and owners. 07 ### Plan, Execute, and Delegate 14 min read · your weekly review This is where the manual becomes an operating system: one recurring hour that converts goals into a week, and a delegation habit that works the same whether the worker is human or not. In this chapter Turn goals into projects, priorities, and weekly actions Delegate work clearly to humans and AI Use AI to monitor progress, identify blockers, and prepare the next decision A handoff is complete when the receiver could finish it without asking you anything. Same test for a person and for a model. #### The Friday hour 15 min · look back Dump the week into AI. Ask for what moved, what slipped, and what you avoided. 15 min · decide Pick the three outcomes for next week. Not tasks. Outcomes with a finish line. 20 min · delegate Split each into human work, AI work, and yours. Write the briefs now, not Monday. 10 min · clear Update Decisions and Open loops. The context system stays alive here or nowhere. #### Who gets the work Goes to When You still owe them You Judgment, relationships, one-way doors Uninterrupted time A person Ownership, taste, accountability over time Context and a decision right AI Repeatable, reviewable, reversible A brief and a standard Prompt to keep · weekly founder review Copy prompt Here is my week: {calendar + notes + what I shipped + what slipped}. Context files attached: Strategy, Decisions, Open loops. 1. What actually moved, what slipped, and what did I avoid? Be blunt. 2. Given my strategy, what are the 3 outcomes for next week — outcomes with finish lines, not tasks. 3. Split each into: mine (judgment) / a person / AI. Write the brief for each AI and human handoff. 4. Draft the updates to my Decisions and Open loops files. Do today Put the Friday hour in your calendar as a recurring block, and run the first one this week even if the week was a mess. 08 ### Automate Your Founder Workflow 13 min read · three automations Anything you have done the same way three times should not start from a blank prompt again. A randomized field experiment run through NBER points the same way: the gains come from embedding AI in the everyday email and document flow, not from treating it as a separate occasional tool. In this chapter Identify recurring work that should no longer begin manually Create routines for research, reporting, follow-ups, planning, and monitoring Move from occasional AI use to repeatable systems Three automations, not thirty. Chosen because they run whether or not you are having a good week. Every one needs a trigger, an owner, and a place the output lands. #### Start with these three Monday brief Pipeline, product, cash, and open loops in one page waiting for you before your first meeting. Trigger: calendar. Lands in: your inbox. Follow-up sweep Every commitment you made or received, with drafts ready for the ones that are overdue. Trigger: daily. Lands in: a draft folder you approve. Watchlist Competitors, customers, and hiring signals, filtered to what changes a decision. Trigger: weekly. Lands in: one note, five bullets, sources attached. #### The four-line spec Write this before you build anything, in any tool. If you cannot fill line four, the automation will quietly produce output nobody reads. Trigger Time, event, or inbox rule. Inputs Which context files and systems it may read. Output Format, length, and where it lands. Use The moment in your week when you actually read it. Prompt to keep · automation audit Copy prompt Here are the things I did more than twice this month: {list}. For each, tell me: should this still start manually? If not, write a 4-line spec — TRIGGER (time/event/inbox rule) · INPUTS (which context files and systems it may read) · OUTPUT (format, length, where it lands) · USE (the moment in my week I read it). Then pick the 3 with the best ratio of time returned to setup effort, and tell me which one to build first and why. Do today Build the Monday brief. Manually the first week, so you learn what you actually want in it, then automate exactly that. 09 ### Stay in Control 11 min read · your rules page Control is not caution. It is knowing precisely where the tool is strong, where it fails quietly, and which of your decisions never leave your desk. In this chapter Understand where AI performs well and where it becomes unreliable Verify important outputs and protect sensitive information Define which decisions require human review The failure mode is not a wrong answer. It is a fluent one you did not check. Confidence in the output tells you nothing about its accuracy. #### Three rules to write down Verification Anything that goes to a customer, an investor, or a regulator gets checked at the source. Name the two facts per document you always verify yourself. Privacy Decide once what never gets pasted: employee data, customer personal data, signed terms, credentials, unannounced numbers. Keep a redacted version of anything you need often. Approval List the decisions that are yours alone regardless of how good the draft is: pricing commitments, hiring and exits, fundraising, legal positions, anything that moves money. #### Where the value actually sits Microsoft's work on advanced AI use describes the same shift this manual is built on: AI increasingly supports analysis, problem-solving, evaluation, and decision-making, while human value moves toward intent, judgment, taste, and the design of the work itself. That is the job you are keeping. Everything in chapters 1 to 8 exists to give you more room to do it. Prompt to keep · your control page Copy prompt Help me write my control page, one page, plain language: 1. VERIFY — for my business, which outputs must be checked at the source before they leave, and which 2 facts per document I should always verify myself. 2. NEVER PASTE — list the categories of information I should keep out of AI tools entirely, and how to redact the ones I need often. 3. MY CALL ONLY — the decisions that stay with me regardless of how good the draft is. 4. Then tell me where, for MY kind of work, you are most likely to be confidently wrong. Do today Write the privacy list. One page, plain language, and share it with anyone else in the company who uses AI on your behalf. Final project ### Your Founder AI Operating System Seven pieces. Each one is the output of a chapter you have already read, which means you are assembling, not starting. 01 A map of where AI belongs in your personal work Role by role, mode by mode. From chapter 01. 02 Ten reusable founder workflows The prompts from this manual, edited into your own words. Chapters 01–08. 03 A company and personal context system The seven one-page files, with a refresh cadence. From chapter 02. 04 An AI-assisted decision process Widen, stress, pre-mortem, decide, review. From chapter 04. 05 A weekly founder review The Friday hour, in your calendar with the prompt attached. From chapter 07. 06 Three recurring automations Each with a trigger, inputs, output, and the moment you read it. From chapter 08. 07 Clear privacy, verification, and approval rules One page you would be comfortable showing your board, and that anyone working for you can follow. From chapter 09. Done means You used at least five of the ten workflows in the last week without thinking about it. Someone else could run your weekly review from the written prompt. Your three automations ran without you starting them. You can name the last decision you refused to delegate, and why. Evidence ### Sources Three findings shape the structure of this manual. Microsoft, 2026 Work Trend Index Advanced AI use increasingly supports analysis, problem-solving, evaluation, and decision-making, while human value shifts toward intent, judgment, taste, and work design. microsoft.com — Agents, human agency, and opportunity Harvard Business School — Navigating the Jagged Technological Frontier Peer-reviewed field experimental evidence: AI can substantially improve knowledge work when the task fits its capabilities, but can reduce performance when users trust it outside those boundaries. hbs.edu — Dell'Acqua et al. (PDF) NBER randomized field experiment Supports embedding AI directly into everyday email and document workflows rather than treating it as a separate occasional tool. Back to the top Sista AI · AI-Powered Founder