# 10-80-10 — Full text for LLMs Site: 10-80-10 — Human · AI · Human URL: https://build108010.com Published by: Objektive Ventures (https://objektive.io) Series: Objektive Working Paper No. 1, in the Objektive Working Papers series (https://objektive.io), 2026 edition. This document is the full text of the paper in page order, followed by the Sources and the citation. Citation: Objektive Working Papers (2026). 10-80-10: Human · AI · Human. Working Paper No. 1, 2026 edition. https://build108010.com --- ## 1. Home ### Hero You are a builder. You can transform any industry or domain. **10 — 80 — 10** Human · AI · Human ### The 200 words Give AI the direction. Let it prototype. React. That is the whole method, and it is a ratio: ten from your instinct and your context, eighty from AI's analysis, processing and prototyping, ten from your judgement and your name. If you are doing most of the work, you are not using AI. If you are not having fun, AI is not working hard enough for you. So start now, three ways. Plan: ask for a plan with milestones small enough to hit this week, and hold it to helping you hit them. Ideate: ask for ideas around what you are working on, or hand over the idea you have and ask it to pressure-test it. Prototype: ask for a rough version of the thing and react to that. Chop, decide, go again. Speed is the point, and so is time. The faster and deeper you go, the more often AI hands back an idea built on yours — the moment you see where to go next. Then sleep on it. Let it percolate overnight and come back. The more rounds you give yourself, the prouder you will be of what you built. Finish it yourself. Make it specific, make it yours, sign it. ### Now / Tomorrow **NOW · PLAN** — "Build me a plan for [the thing] with milestones I can hit this week, one line each, and check in on each one with me." **NOW · IDEATE** — "Give me five ideas for [the thing] I have not considered — or here is my idea; pressure-test it and tell me what I am missing." **NOW · PROTOTYPE** — "Build the roughest version of [the thing] you can in one pass and label every guess. I will react." **TOMORROW** — Read it again. Take the one that surprised you. Go one round deeper. ### Formula strip **AIᴴ = Bⁿ** — AI to the power of human. Building at scale. --- ## 2. Method ### The first ten: direct Before AI touches it, decide two things. What would make this yours. The example only you would use. The number nobody else has. The position you hold that your category does not. The thing you would say in the room that the deck would not. Write those down; they are the brief. Where AI must stop and ask. The claims it cannot source. The conventions in your field it cannot be sure of. The numbers it would be guessing. Write those down too; they are the stop conditions. Quality goes in here, at the front, not at the end. Then ask AI to write the prompt it wants from you, and edit that. You will get a better brief in three minutes than you would write in thirty. ### The eighty: let it run, stay in the loop Hand it over. Analysis, processing, prototyping, the long middle — that is AI's, and it is getting longer: agents that write and run code, work a browser, carry a plan for hours while you are somewhere else. Let it. But do not disappear. Set the stop conditions so AI shows its work at the points you chose, asks the questions you told it to ask, and waits. Answer fast and let it continue. Interleaved beats hand-off: the builder who stays in the loop catches the wrong turn at minute ten, not hour four, and is still sharp when the last ten arrives. ### The last ten: finish and sign The last ten is a short checklist, not a long read. Reviews get skipped; checklists get done. Make it specific: swap every generic noun for the real one. Make it clean: strip the words and patterns AI puts in by default. Verify: every claim to a source you have actually read. Make it yours: read it aloud, or have someone who knows you read it — your own sense of your voice is the one instrument to distrust. Sign it: one line saying what AI did and what you did. Five tests. Pass them and it carries your name. Fail one and it is still AI's. ### Keep the tens Every correction you make in the last ten is something AI did not know. Keep it. Save the corrected version and the reason where the next run starts. Turn the corrections into standing rules — the words you never use, the examples you always use, the claims that need a source in your field — and open the next session with them. Your tens are how the loop learns; throw them away and every build starts from zero. ### Where the ratio does not hold 10-80-10 is for building. It is not for deciding between right answers, and it is not for checking a domain AI already handles better than you do. If the task is classification, verification or a decision with one correct answer, split the work rather than sandwich it. --- ## 3. Start Every item is a prompt to paste, then what to do with what comes back. Organised the way the Home page is: Plan, Ideate, Prototype, then Finish and Keep. ### Plan > I am working on [the thing]. Before you do any of it: tell me what you need from me to do it well, ask me the five questions whose answers would change the result most, and draft the prompt you want me to give you. Answer the five. Edit its prompt. Don't write your own. > Chop [the thing] into the smallest decisions I need to make, in the order I need to make them. One line each. I will answer them in a row. Answer them in a row. Fifteen small yeses beat one big maybe. > Build me a plan for [the thing] with milestones I can hit this week, one line each, and check in on each one with me. Cross out the milestone you would skip. Keep the rest. ### Ideate > Give me five ideas for [the thing] I have not considered. One line each. > Here is my idea for [the thing]: [idea]. Pressure-test it. What am I missing, what would a sceptic say, what would make it stronger? Take the objection that stung. Answer it in the build. > Based on everything I have told you so far, propose three directions I have not mentioned. Make one of them a stretch. Take the one that surprised you. That is where the exponent lives. ### Prototype > Build the roughest working version of [the thing] you can in one pass, and label every place you guessed. I want something to react to, not something finished. React to the guesses only. Leave the rest. > Draw the schematic: the parts of [the thing], how they connect, and the one part you are least sure about. Text or a table. I will correct it. Redraw the wrong connection. Name the missing part. > Here is the draft. Where are the three places someone who knows this subject would stop reading, and why? Fix those three. Run it again. Then stop for the day; read it tomorrow before you touch it. ### Set the stop conditions (before any long run) > Before you start, list the conditions under which you should stop and ask me rather than proceed: any claim you cannot source, any convention in this field you are not certain of, any number you would be estimating. I will approve the list. Add what it missed. These are your stop conditions for every run after this one. ### Borrow the other papers > **R2.** Read this through the seven R2 levers and tell me where it makes us more relevant, where more reputable, and where it does neither. — buildr2.com > **Q2.** List the sources this claim rests on, using the Q2 grid: asked, observed or assembled, and what n each would need before I could say it out loud. — scaleq2.com > **POET.** Which of Partnered, Organic, Earned or Targeted does this belong in, and what would a model need to see before it would cite it? — getpoet.ai > **P3.** Give me the five-word version, the two-hundred-word version and the outline of the eight-hundred-word version, each with the proof it needs. — provep3.com > **Per Mille.** Who are the Per Mille figures in this domain — the ones the tier below cites — and what would each say is wrong with this? — permilleeffect.com ### Finish > Strip every word on this list from the text: [the tells list]. Remove every sentence that restates the one before it and every closing sentence that summarises its paragraph. Return only the diff. > Replace every generic noun with the specific one. Where you do not know it, leave a bracket and I will fill it. > List every factual claim with the source it rests on. Mark the ones you could not source. Then the two things only you can do: read it aloud for voice; write the provenance line and sign it. ### Keep the tens > Here are my corrections to your last output, with reasons. Turn them into standing rules I can give you at the start of next time, in under fifteen lines. Save the rules. Open the next run with them. --- ## 4. Standard ### Formula block Compact: **AIᴴ = Bⁿ** Expanded: **(AI)ᴴᵘᵐᵃⁿ = Bⁿ** Terms: - **AI** — Artificial intelligence: the eighty; analysis, processing and prototyping - **H** — Human: the two tens; your instinct and context before, your judgement and name after - **B** — Building, in any industry or domain - **n** — Scale, set by a (authentic) and c (creative) - **a, c** — Authentic: recognisably and accountably yours · Creative: distinctive, specific, new Spelled out: Artificial intelligence, raised to the power of the human, produces building at scale that is defined by being authentic and creative. ### The four corners | | low c | high c | |---|---|---| | **high a** | Built, but ordinary | **Bⁿ** | | **low a** | Slop | Clever, but AI's | Bⁿ is the top-right corner. a and c decide the n. ### Five tests for the last ten 1. **Specific.** It names things: a company, a number, a date, a person, a place. Nothing that could be true of anyone in the category. (c) 2. **Clean.** The tells are gone: the default vocabulary, the triads, the "not A but B" reversals, the sentence that restates the one before it, the closing aphorism. (c) 3. **Verified.** Every claim traces to a source you have read, not a summary of one. (a) 4. **Voiced.** A reader who knows you would know it was you. Tested by a reader, not felt by you. (a) 5. **Signed.** One line: what AI did, what you did. If you would not sign it, it is not finished. (a) ### The tells Vocabulary: delve, landscape, leverage, robust, seamless, navigate, unlock, elevate, underscore, showcase, tapestry, journey, testament. Openers: in today's, it is important to note, crucially, ultimately, at its core, in essence. Structures: three near-synonyms in a row; "not X but Y" more than once in a passage; a rhetorical question answered in the next sentence; a paragraph that ends by summarising itself; a heading restated as the first sentence beneath it. Registers: a text describing its own structure; reassurance and caveats; encouragement. ### The provenance line One sentence where the work is published: "Drafted with [tool]; directed, verified and edited by [name]." Decide it before publishing, not after. --- ## 5. Long View ### A ratio that holds 10-80-10 began as a rule for leaders, not for software. John Maxwell wrote it down in 2007 and again in 2014: spend the first ten per cent casting the vision and lining up the resources, delegate the middle eighty, and come back for the last ten to put the finish on it. Dan Martell repeated it for founders in 2023. It held because it answers the question every leader actually has — how much of this should I do myself — with a number. AI made the number urgent. The eighty is no longer a team you have to hire; it is analysis, processing and prototyping available to anyone, at any hour, and now running as agents that write code, work a browser and carry a plan for hours. When the middle costs almost nothing, the only question left is what you put in front of it and what you do with what comes back. That is the ratio, and it is why the formula puts you in the exponent: AI to the power of human. ### What the tens are made of The first ten is instinct and context. Most of what you know about your field is not written down anywhere. It is in the customer-service call, the sales conversation, the regulator's aside, the colleague who says "that will not work here." No model has seen it. It enters the build through your brief and your stop conditions, and it is the single thing that makes your eighty different from anyone else's. The last ten is judgement and signature. AI's default is the average of everything it has read. Left alone, its output converges: measurably less varied across writers, measurably stripped of the small markers — contractions, first person, the specific noun — that readers use to recognise a person. The last ten puts them back and puts a name on the result. It is where distinctiveness is restored, discoverability is earned, and proof is attached so the work can travel. ### Why it works, in five findings The evidence is recent and it is consistent. When people use AI to write, their time moves to the ends on its own: drafting halves, editing doubles (Noy and Zhang, *Science*, 2023). Specific, task-level correction improves output; global end-of-process judgement often does not, and in a third of cases makes it worse (Kluger and DeNisi, 607 effect sizes). People who stay in the loop outperform those who hand off cleanly (Dell'Acqua et al., the BCG field experiment, 2023). A small, bounded adjustment at the end beats an open-ended review, both for adoption and for accuracy (Dietvorst, Simmons and Massey, 2018). And the combination of a person and AI beats the better of the two alone on creation tasks — building — while losing on classification and decision tasks (Vaccaro, Almaatouq and Malone, *Nature Human Behaviour*, 2024). The ratio is not a slogan. It is the shape the data takes. ### The two ways it fails The first failure is doing too much yourself. Most people, given AI, still draft first and ask second, and then use AI as a spell-checker on their own work. They get a faster version of what they already had. The second is doing too little at the end. In the largest writing experiment to date, 68% of participants submitted AI's draft unedited. Confidence in AI predicts less critical effort. Experienced builders reviewing AI's work on their own code were slower than without it and believed they were faster (METR, 2025). And a finishing pass recovers less of your voice than it feels like: people who edited AI drafts of their own vows and eulogies moved the text toward themselves, yet it stayed far closer to AI's, and they could not tell (Baumler et al., 2026). This is why the last ten is a checklist with five tests and a reader who knows you, not a feeling that it sounds right. ### The loop that learns Every correction in the last ten is expertise AI did not have. The builders who keep their corrections — as context the next run starts from, as standing rules, as the sequence of "no, like this" — have a loop that improves; the ones who discard them start every build from zero. The frontier of model training now rhymes with this: supervised reinforcement learning teaches a model step by step from expert trajectories rather than from final answers. Your tens are the trajectory. ### Where the other papers come from Build fast and you will need to know whether what you built is any good. That is R2: whether it is seen and believed, and the growth that produces. Build on what your field actually knows and you will need the sources that hold it: that is Q2, quantitative and qualitative, and why the asked and observed sources matter most. Build something distinctive and you will need to place it where people and models look, through the channels that decide: that is POET. Build it to be read and you will need it provocative, persuasive and positioned, with proof: that is P3. Build to reach the few who move the many: that is the Per Mille Effect. 10-80-10 is the first paper because it is how each of the others gets built. ### Provenance Human involvement earns trust when it is known and true — and it is now also law and infrastructure: disclosure rules for AI-generated content are enforceable, content credentials travel with files, and search systems reward demonstrated expertise. The provenance line in the last ten is your side of that bargain. Disclosure costs a little trust; being exposed costs more. --- ## 6. Sources Maxwell, J. C., "The 10-80-10 Principle," *Leadership Wired* (2014; first documented 2007). Martell, D., *Buy Back Your Time* (2023). Noy, S. and Zhang, W., *Science* 381 (2023). Kluger, A. and DeNisi, A., *Psychological Bulletin* 119 (1996). Dell'Acqua, F. et al., HBS Working Paper 24-013 (2023). Dietvorst, B., Simmons, J. and Massey, C., *Management Science* 64 (2018). Vaccaro, M., Almaatouq, A. and Malone, T., *Nature Human Behaviour* 8 (2024). Lee, H.-P. et al., *CHI* (2025). METR, experienced open-source developer study (2025). Baumler, C. et al., "Can You Make It Sound Like You?" (2026). Padmakumar, V. and He, H., *ICLR* (2024). van Nuenen, T., "Voice Under Revision" (2026). Kobak, D. et al., *Science Advances* (2025). Bainbridge, L., *Automatica* 19 (1983). Endsley, M. and Kiris, E., *Human Factors* 37 (1995). Deming, W. E., *Out of the Crisis* (1986). Simkute, A. et al., *Int. J. Human–Computer Interaction* (2024). Agarwal, N. et al., NBER w31422 (2023). Schilke, O. and Reimann, M., *OBHDP* 188 (2025). Proksch, S. et al., *Frontiers in AI* (2024). de Rooij, A., *Psychology of Aesthetics, Creativity, and the Arts* (2025). Google Research, "Supervised Reinforcement Learning: From Expert Trajectories to Step-wise Reasoning" (2025). Anthropic Economic Index, June 2026. --- Objektive Working Papers (2026). 10-80-10: Human · AI · Human. Working Paper No. 1, 2026 edition. https://build108010.com