
Imagine you need to hire a new employee, and you have two great candidates sitting in front of you. The first one is unpredictable—they are sometimes late for work, they miss assignments and deadlines, and they often hand in subpar work that makes you wince. But every once in a while, they produce something extraordinary, a piece of work so brilliant it sits way above what anyone else in the building is doing. Then there is the second candidate. They are always on time, they never miss a deadline, and they produce solid, reliable work; but they never deliver anything above what is expected of them.
Which one would you hire? I would venture to say that, while there are always exceptions, the majority of hiring managers would pick the second candidate—unless the first one is very good at hiding their vices in the interview. The reason is not that business dislikes brilliance. It is that inconsistency carries a hidden, exhausting cost.
Think about what happens in the weeks between the first candidate’s flashes of genius. The manager has to pay extra attention to the quality of the work. They have to spend extra time reviewing every deliverable before it goes out the door. They often need to step in and rework the outputs themselves just to make them safe for a client to see. The brilliant employee might be doing the typing, but the manager is carrying the cognitive load of making sure the floor does not fall out. You are not just paying the employee’s salary—you are paying a tax in management attention, and that tax scales with every unpredictable person you add to the team.
We make this exact same calculation as consumers every day, often without realizing it. I continue shopping at Amazon, even though I know there are places on the internet with better deals, and often with a better customer experience. But I know that the package will definitely arrive, and arrive fairly quickly. I know that returning items is no-hassle. I know I do not have to talk to anybody to get it done. The consistency is the customer value. The relief of knowing what is going to happen is worth more to me than the occasional delight of a slightly better deal somewhere else.
When you bring artificial intelligence into your business for real work, you face the exact same dynamic. Without a contextual harness, a large language model behaves remarkably like that first candidate, the occasionally brilliant one. You ask it a question and it might hand you a profound strategic insight, and then the next afternoon it hallucinates a confident, perfectly formatted disaster. If you use it this way, you have not actually outsourced the work; you have transferred the review and rework onto yourself, and the tool that was supposed to reduce your busywork has added to it.
But with a contextual harness—the system around the model that I described a few weeks ago in Your Genius Employee Has His Limits, which surrounds it with your real business knowledge and explicit rules for how the work should be done—the AI behaves much more like the dependable employee, for the most part.
I saw this shift happen in my own practice. Not long ago, the quality of our presentations, white papers, websites, and infographics varied. It depended on who happened to be working on the project, and on whether I had enough time to sit down and review the final draft thoroughly. It was the definition of an inconsistent workflow. But by separating the knowledge from the output—building a maintained foundation of sourced knowledge that the system draws from every time, which we call the Knowledge Engine—we got a much more consistent quality of work. New deliverables no longer require us to rebuild the context from scratch, and the work depends far less on my memory or my availability to catch every omission.
These essays are a good example of how that works now. Each one starts as a conversation in which an AI agent interviews me and I supply the stories. A different agent writes the first draft; I mark it up, sometimes through as many as eight drafts before I am satisfied; another applies my edits and checks every link, number, and quote against its source; and a fresh session that never saw the draft being written reviews it against the same written standard, listing its findings below the text instead of quietly fixing them.
This leads to an implication that I believe changes the geometry of how businesses compete. As I see it, large companies gain momentum largely through consistent processes—whether in healthcare, customer experience, manufacturing, or the steadiness of their strategy and routines—and their scale is what lets them afford the infrastructure that keeps one person’s bad day from sinking a project. Think of the World Health Organization’s 19-item surgical safety checklist: when eight hospitals adopted it in a 2009 study published in the New England Journal of Medicine, death rates after major operations nearly halved and major complications fell by more than a third.
By that logic, smaller companies have had a harder time with this, often leaning instead on a few people’s heroic efforts, which can burn those people out. But well-organized, AI-supported workflows now allow smaller companies to create their own consistent, repeatable routines and, I believe, to match the institutional memory and dependable quality of a large enterprise without having to hire an army of middle managers to enforce it. The editorial routine behind these essays is one small firm’s version of that surgical checklist. I wrote about how we build routines like it—plain-English instructions for a single workflow, in which some steps are done by a person and some by an AI agent—in The Hidden Powers of Kick-Ass Companies.
Does this mean we should stamp out brilliance in the name of perfect predictability? Of course not. There are exceptions in the hiring example, and there are exceptions in how we use AI. When innovation is the actual product—when we need fresh, original ideas every once in a while—you want the brilliance. If the environment is highly complex and there are completely different problems to solve every single time, standardization will only get in your way.
If I need a brain surgeon for a complex procedure, I might choose the brilliant specialist regardless of whether he is late for work on a daily basis or fails to fill out his paperwork properly—although I would still want him operating in a hospital that runs the checklist. The novelty and exceptional judgment are the entire point. When you are trying to solve specific device antenna issues, or navigate complex battery problems, or figure out how to enable a new vector of growth, a combination of a brilliant human thinker with AI makes both unbeatable.
The same goes for creative work, and you can see what happens when people forget it. Standardize creativity and you get what is called neuroslop in modern culture: the same cartoonish illustrations, the same tidy three-part sentences, the same predictable formats that so many of us have learned to recognize as AI output on sight, and that drive just about everyone a little crazy. Put a pinch of human brilliance into the equation, and the end product gets much better. Take “Movin’ to the Sun,” which began as an AI-generated song by Imael Angel and ended up reaching the top of Shazam’s worldwide chart after human vocals were added.
The distinction is knowing which kind of work you are looking at. AI can help automate the routine tasks, freeing up the genius for the moments that actually require it. The goal is not to eliminate brilliant thinking, but to make dependable execution the default for everything else.
Look at the workflows in your own business that keep fluctuating based on who is working on them, or who happens to have time this week. Maybe it is your proposal writing, or your customer onboarding, or the way you handle frontline customer service. Ask yourself: what would change in my business if these workflows became consistent? How much time would it free up? How much would it improve customer satisfaction, or increase sales?
If you can answer that, you know exactly where to start. And does all this mean that an AI harness could make the sloppy candidate we didn’t hire back in the second paragraph more reliable? It just might.


