The AI Rules People Quote, and Whether Any of Them Hold Up
Revised August 22, 2026
What is the 10/20-70 rule for AI?
It’s a resource-allocation principle from Boston Consulting Group: put roughly 10% of your effort into algorithms, 20% into technology and data, and 70% into people and process. BCG presents it as a pattern drawn from its own client engagements, not as a measured research result. That distinction is the whole ballgame, and it disappears the moment the number reaches social media.
Keep reading ↓Imagine it’s a Tuesday morning, the coffee is doing its work, and you’re half-scrolling LinkedIn before the first call of the day. A post slides by with fourteen hundred reactions. It says only 10% of AI success is the algorithm and 70% is your people, and it calls this the 10/20/70 rule. Two posts later somebody explains the 30% rule. Somebody else lays out the four stages of AI adoption. A commenter drops the three laws of AI underneath, and forty people nod along.
By the time you set the phone down you’ve absorbed four apparent laws of nature about a technology most of us have had for about three years. An office manager in Ladue forwards one of them to her boss with a note that says we should talk about this. A shop owner in Jennings screenshots another. A contractor in Highland reads a third at a red light and decides, quietly, that he’s behind.
None of that is anybody’s fault. This subject is genuinely confusing, the vendor noise is deafening, and a tidy number feels like a handrail in the dark. So here is each of those rules traced back as far as it actually goes: who said it first, what kind of claim it is, and in two cases, the plain admission that the trail runs cold.
What is the 10/20-70 rule for AI?
It’s a resource-allocation principle from Boston Consulting Group: put roughly 10% of your effort into algorithms, 20% into technology and data, and 70% into people and process. BCG presents it as a pattern drawn from its own client engagements, not as a measured research result. That distinction is the whole ballgame, and it disappears the moment the number reaches social media.
The framework runs through BCG’s generative AI writing, including work authored by Nicolas de Bellefonds, Sylvain Duranton and Vladimir Lukic, and it reappears in the firm’s 2025 report on the gap between AI spending and AI profit. Forbes contributor Joe McKendrick wrote it up on January 26, 2026 as a BCG principle, which is roughly the moment it started showing up in every third post on your feed.
The idea underneath it is worth hearing. Businesses that treat an AI rollout as a software purchase tend to end up with software nobody opens. The tool arrives, two people try it, the trial lapses, and everyone agrees AI didn’t work here. Anyone who has watched a point-of-sale migration go sideways already knows this in their bones. So the rule is pointing at something true.
It is still a rule of thumb wearing a lab coat.
Is the 10/20-70 split a research finding or a consulting framework?
A framework. BCG’s own framing is that the split reflects what it has seen across case work involving hundreds of companies — a consultancy describing the shape of its own engagements. Publishing that is perfectly legitimate and genuinely useful. It is not a study. Nobody assigned firms at random to a 10/20/70 budget, held everything else steady, and measured what came out the other end.
Why should you care about the difference? Because a finding tells you what is true and a framework tells you where to look. If your actual bottleneck is that your customer records live in three places and none of them agree, the 70% figure is wrong for your situation and following it will waste your year. Frameworks are lenses. Lenses are for pointing, not for obeying.
One more tangle worth undoing. There is an older 70-20-10 model in workplace learning — 70% of development from doing the job, 20% from other people, 10% from formal courses — developed in the 1980s by Morgan McCall, Michael Lombardo and Robert Eichinger in connection with the Center for Creative Leadership and popularized in their 1996 book The Career Architect Development Planner. Its evidence base has been argued over for years, because it grew out of roughly 200 managers recalling how they learned. Same digits, different field, unrelated claim. When you see the numbers reversed in an AI post, that is usually where the confusion started.
So where did the 30% rule come from?
I could not source it, and I want to be direct about that. There are at least two different rules sharing that number, neither carries a named author, a paper, a firm or a date, and the explainer articles circulating about it cannot source it either. This one is folklore, and it deserves to be labeled as folklore rather than quietly given a pedigree it never had.
Version one is a workflow split: let AI handle about 70% of a repetitive, data-heavy task and keep the remaining 30% — the judgment, the ethics, the creative call, the part where somebody is accountable — with a human. Version two comes out of schools, and says no more than about 30% of what you hand in should be AI-generated.
When BGR published an explainer on June 9, 2026, writer Joshua Hawkins reached for the same hedge anyone honest reaches for: according to some online reports, the idea may have originated in education, where Turnitin is used to assess how much of a submission looks AI-written. Note the hedging. Note also that Turnitin does not set a 30% threshold. The number looks like individual instructor practice that hardened, through repetition, into a rumored platform policy.
So the honest summary is this. The 30% rule circulates widely, it is quoted with real confidence, and I could not find its origin. That does not make the underlying advice bad — keep a person on the half that requires judgment is decent guidance in most trades. It means that if somebody quotes 30% at you as an established standard, they cannot tell you who established it, because nobody can.
Are the three laws of AI real laws?
No. They are Isaac Asimov’s Three Laws of Robotics, and they first appeared in full in a short story called Runaround, published in the March 1942 issue of Astounding Science Fiction. That is fiction, written by a young biochemistry student to give his robot stories something to break against. Asimov credited the magazine’s editor, John W. Campbell Jr., with helping formulate them during a conversation in December 1940; the first law alone had appeared in his story Liar! in May 1941.
They are elegant, and that elegance is exactly why they keep getting mistaken for policy. A robot may not harm a human or, through inaction, allow a human to come to harm. A robot must obey human orders unless doing so conflicts with the first law. A robot must protect itself unless that conflicts with the first two.
Here is the part almost nobody mentions when they quote them: most of Asimov’s robot stories are about the laws failing. The whole point of the device is that three tidy rules cannot survive contact with a messy situation. He built them so they would crack, then wrote the cracks.
No statute contains them. No AI product implements them. No regulator anywhere cites them. If you want to know what an AI system is actually forbidden from doing, you have to read law, and we get to that below.
What about the four stages of AI adoption?
There is no canonical four-stage model, and that is the answer rather than a dodge. Searching for the original turns up several unrelated four-stage lists published by different organizations with different stage names. Innovate UK Business Connect publishes an Adoption Stages Framework built on Strategy, Data, Build and Implement. Other consultancies publish four stages running from resistant through exploring and implementing to optimized. Others publish five stages and call those canonical too.
They are not versions of one underlying model. They are separate maps drawn by separate firms, usually to sit at the front of a service offering, and each one is arranged so that the reader lands somewhere in the middle and needs help getting to the end. That is not a scandal. It is just what a maturity model is for.
So when a post asks what the four stages of AI adoption are, the accurate response is: whose four? Any of these can be a useful way to organize a conversation. None of them is a measurement of where you actually are.
Wildly off-topic — but St. Louis antique dealers know about old rules.
Why does everyone say 95% of AI projects fail?
Because of one report. In August 2025, a group at the MIT Media Lab’s Project NANDA circulated a paper titled The GenAI Divide: State of AI in Business 2025, reporting that roughly 95% of enterprise generative AI pilots produced no measurable profit-and-loss impact. Fortune covered it on August 18, 2025, AI-exposed stocks wobbled, and the figure has been quoted daily ever since.
What it actually rests on: about 150 interviews with business leaders, roughly 350 employee survey responses, and analysis of around 300 publicly described deployments. It was circulated as a report rather than published through peer review. And the thing it counted was measurable impact on the profit line, not whether the software functioned. A pilot that worked fine and simply was never measured lands in the 95% alongside a pilot that collapsed.
It is also about enterprises. The subject was corporate pilots with real budgets, not a two-truck HVAC outfit using a chatbot to write follow-up emails. The report’s own conclusion was that the divide came down to approach rather than model quality — which, notice, is the same claim BCG’s 70% is making from a completely different direction. Two independent sources arriving at similar conclusions is the strongest thing in this whole article, and it is nobody’s favorite statistic.
Did Stephen Hawking really warn us about AI?
Yes, and unlike most of the material in this article, the quote is real and precisely sourceable. In a BBC interview published December 2, 2014, Stephen Hawking told technology correspondent Rory Cellan-Jones that the development of full artificial intelligence could spell the end of the human race. He went on to say such a system would redesign itself at an ever-increasing rate, while humans, limited by slow biological evolution, could not compete.
Two pieces of context change how that lands. The interview was pegged to a new communication system built for Hawking with help from Intel and a predictive-text company — he was, at that moment, an enthusiastic user of the technology he was warning about. And the phrase he used was full artificial intelligence: a hypothetical system that improves itself without us. That is a real debate among serious people.
It is not a comment on whether a salon in the metro should let software draft appointment reminders. When the quote gets pasted under a post about business tools, it borrows the weight of a very different argument. Hawking earned his warning. It just was not about your invoicing.
Which AI rules are actually written down?
The only place a genuine list of forbidden AI uses exists is statute, and the most complete one is European. Article 5 of the EU AI Act sets out eight prohibited practices — among them social scoring, untargeted scraping of facial images to build recognition databases, and emotion recognition in workplaces and schools. Those prohibitions became applicable February 2, 2025, and penalties became enforceable August 2, 2025, at up to 35 million euros or 7% of worldwide annual turnover, whichever is higher.
In the United States there is no comprehensive federal AI statute. So the rules that reach a business here are state rules, and in this metro they change when you cross the river.
Illinois has been writing AI law for years. The Artificial Intelligence Video Interview Act took effect January 1, 2020, and requires an employer to give notice, explain how the system works, and get consent before AI analyzes a candidate’s video interview. House Bill 3773 amends the Illinois Human Rights Act effective January 1, 2026, barring the use of AI that has a discriminatory effect in employment decisions, requiring notice, and specifically prohibiting zip code as a proxy for a protected class. And the Wellness and Oversight for Psychological Resources Act, announced by the Illinois Department of Financial and Professional Regulation on August 4, 2025, bars using AI to deliver therapy or make mental health treatment decisions, while still allowing licensed clinicians to use it for administrative work.
Missouri has no AI-specific statute. More than a dozen AI bills were filed in the 2026 legislative session and every one of them died before the session ended, as KFVS reported on May 25, 2026. The next chance for state lawmakers is the session that opens in January 2027.
The practical version: an employer in Highland screening applicants with an AI hiring tool carries notice and non-discrimination obligations that an employer twenty-five minutes away in Missouri does not. That is a real, checkable difference, and it is worth more to you than any of the numbered rules above.
What this looks like from behind the counter
If you run a small local service business, the constraint is not your software budget. It’s your own hours. You are the estimator, the dispatcher, the person who answers at 7:40 on a Wednesday night, and the one reconciling receipts on Sunday. Margin lives in scheduling density and in the callback you didn’t have to make, and a good chunk of the year gets decided in two or three brutal weeks when everybody calls at once. What owners complain to each other about is not falling behind on AI. It’s the estimate that sat in a truck three days and lost the job to whoever answered first.
That is why these rules land strangely. They were written for a company with a change-management line item. Your version of the 70% is you, on a Sunday, deciding whether a tool is worth learning. And that estimate only matters because somebody went looking for your trade in your town first — which is what being listed and complete on St Louis Near Me Directory is for.
What actually decides whether AI helps a small business
After taking apart the folklore, it would be cheap to hand you another invented rule. So here is only what the evidence and plain observation will support, offered as questions rather than laws.
Is it one named task? Not AI for the business — the follow-up email after an estimate, the weekly schedule, the first draft of a job description. Every source above that says anything credible says the same thing from a different angle: the approach decides the outcome, and a vague approach has no outcome to measure.
Does the output have a defined shape somebody checks before a customer sees it? This is the one that quietly matters most, because in a lot of small businesses the staff are already using these tools whether or not the owner knows, which is its own shadow AI problem worth understanding.
Does it give you back your hours specifically? Not a staff member’s slack time. Yours. And would you notice within a week if it stopped working? If you can’t say what breaks when you switch it off, it was never doing anything.
One last piece of reassurance, with real numbers attached. Census Bureau data published May 26, 2026, covering December 14, 2025 through May 3, 2026, found under 20% of firms with four or fewer employees using AI, compared with 32% of firms with 100 to 249 employees and 37% of those with 250 or more. A Federal Reserve note dated April 3, 2026 put overall use at about 18% of firms as of the end of 2025. Both are government figures, not vendor surveys, and they say the same thing: the smallest businesses are the least likely to have adopted anything.
One caution on those numbers. The Census Bureau changed its AI question in November 2025, from asking whether a firm uses AI to produce goods or services to asking whether it uses AI in any business function, and reported rates jumped between 47% and 159% depending on industry. Any trend line drawn across that date is measuring the question, not the behavior. If you want that pulled apart properly, the companion piece on what percent of small businesses actually use AI does it in detail, and the broader overview of AI for St. Louis businesses covers where to start.
If you would rather work through one real task with other owners in the room than read another numbered rule, that is more or less the premise of the AI workshops for metro business owners. And if you want to see how a specific tool behaves before you trust it with anything, start by reading up on Google AI Mode in plain language, since it is the one most of your customers will meet first.
Tired of being handed a rule instead of an answer? Look at the AI workshops for business owners across the St. Louis metro on St Louis Near Me Directory, then bring the single task that eats your Sunday evening and work on that one instead of on a framework.
Frequently asked questions
What is the 30% rule in AI?
Two different rules share that number and neither has a traceable origin. One says AI should handle about 70% of a repetitive task while a human keeps the 30% requiring judgment. The other, from education, says no more than about 30% of submitted work should be AI-generated. Turnitin does not set a 30% threshold, and a BGR explainer dated June 9, 2026 could only say the idea may have started in schools.
What are the four stages of AI adoption?
It depends whose four you mean, because there is no canonical model. Innovate UK Business Connect publishes an Adoption Stages Framework of Strategy, Data, Build and Implement. Other organizations publish four stages running from resistant through exploring and implementing to optimized, and others publish five. The phrase cannot be traced to a single originator, so treat any four-stage list as one firm’s map rather than a measurement.
What are the three rules of AI?
People usually mean Isaac Asimov’s Three Laws of Robotics, which appeared in full in his short story Runaround in the March 1942 issue of Astounding Science Fiction. They are fiction. No statute contains them, no AI product implements them, and Asimov wrote most of his robot stories about the ways three tidy rules come apart when a real situation gets complicated.
What are the 3 AI laws?
The same three, in Asimov’s wording: a robot may not injure a human or through inaction allow a human to come to harm; a robot must obey human orders except where that conflicts with the first law; a robot must protect its own existence except where that conflicts with the first two. Asimov credited editor John W. Campbell Jr. with helping formulate them in a December 1940 conversation. They are a storytelling device, not regulation.
What is AI not allowed to do?
In the United States there is no comprehensive federal AI statute, so it depends on your state. Illinois bars AI from delivering therapy or making mental health treatment decisions under a law announced August 4, 2025, and from January 1, 2026 bars AI with a discriminatory effect in employment decisions. Missouri has no AI-specific statute; more than a dozen bills died in the 2026 session.
What was Stephen Hawking’s warning about AI?
In a BBC interview published December 2, 2014, Hawking told correspondent Rory Cellan-Jones that the development of full artificial intelligence could spell the end of the human race, because such a system would redesign itself faster than slow biological evolution lets humans keep up. He was describing hypothetical self-improving general AI, not the business tools sold today, and he was using assistive technology at the time.
What jobs will be gone by 2030 due to AI?
Nobody knows, and any list presented as settled is a guess in list form. The closest official source is the US Bureau of Labor Statistics Employment Projections for 2024 to 2034, released August 28, 2025, which extrapolates occupational trends rather than forecasting AI effects and names no occupation as eliminated by AI. The 2030 date itself comes from headlines, not from a study.
Which 5 jobs will survive AI?
There is no authoritative list of five, and the ones circulating rarely name a source because there usually is not one. Work that requires showing up in person, holding a license, and being legally liable for the result is hardest to hand to software — but that is reasoning from how the work is structured, not a research finding, and it should be labeled that way.
What 5 jobs will remain after 2030?
No credible source counts jobs that way. Occupations are bundles of tasks, and AI reaches different tasks inside the same job unevenly, so a whole occupation rarely vanishes on a date. Government projections exist — the BLS Employment Projections released August 28, 2025 run through 2034 — but they are trend extrapolations, not AI forecasts, and they publish no five-item survivor list.
Written by a dedicated team of St. Louis locals who live, work, and play right here in the St. Louis metro. Founder Lane Forman and team are committed to building the region’s most trusted directory by verifying listings and connecting local businesses with loyal customers across Missouri and Illinois.