MINT dentistry

Nils Madsen is the Vice President of Data and Analytics at MINT dentistry, a multi-state dental company headquartered in Dallas, TX. He oversees all aspects of the analytics lifecycle, including data architecture, data engineering, business intelligence and data science.

There has been a lot of excitement around modern generative AI and LLM-powered AI agents. These tools are already revolutionizing areas such as internet search, customer service communications and the generation of advertising copy.

As AI gets more advanced over time, it may eventually be able to perform every job better than a person. However, it isn’t clear that current LLM-based reasoning agents will be able to achieve this, because, as a recent Harvard and MIT study demonstrated, LLMs don’t form world models. In other words, they don’t understand the underlying concepts of their answers. This fundamental lack of understanding leads to the familiar issue of hallucinations—the confident statement of incorrect or entirely fabricated information.

Even if current AI approaches are limited, I believe they are already good enough to lead to a fundamental shift in how businesses are structured. Specifically, I think we will start to see a broader variety of departments run by a single person. Not just headed by a single leader, but where the entire department only employs a single, super-productive individual.

This is already feasible in my field of data analytics, and I’ve done it at MINT at various points in history. Database administration, data engineering, business intelligence and data science have enough conceptual overlap that a single individual with enough drive and enough time can learn all of them to a high degree of competence. Cloud services, modern data warehouses, and data transformation tools like dbt and DuckDB have abstracted away enough of the low-level coding and IT infrastructure work that a single person can be productive enough to design, build, extend, and maintain all the data infrastructure and business intelligence of an entire company.

None of the above currently requires AI, but AI might enable a similar thing to be done in other domains.

How AI Enables Single-Person departments

I see two requirements for single-person departments:

1. The person must have enough knowledge to cover all responsibilities of that department competently

2. The person must be productive enough to generate all the work output of the department

Deep research has dramatically reduced the amount of time required to learn a new skill or to gain new knowledge. These days, you can ask your favorite AI service to point you to credible resources to learn any new skill you want in the most time-efficient way possible, and it will list out those resources for you, along with convenient links.

AI is a fantastic tool to learn how to ask questions. When you are new to a field, you don’t yet have enough knowledge even to ask the right questions, which could be a time-consuming hurdle to overcome unless you have access to a human expert. However, with repeated AI prompts, you can rapidly refine your questions as you learn the concepts and terminology of the field.

Seeing how AI can enhance productivity is more obvious— the person can focus on the high-level decisions like tool selection, process design and prioritization, while farming out the lower-level work to AI agents, which would in turn build the components of traditional automated systems. Using agents to build traditional automation rather than doing all the work themselves would leverage the advantages of both kinds of tools, while avoiding the pitfalls of non-deterministic AI agent behavior.

Advantages

Single-person departments would enjoy several advantages over teams. The most obvious are cost efficiency and reduced liability—AI agents cost less, don’t call in sick, work 24 hours a day, and won’t sue the employer for wrongful termination.

Less obvious is the fact that systems designed and built by a single person can be set up and modified much faster than systems built by teams, since communication and task coordination (the job of project managers) are major bottlenecks of projects. This is a step that should have taken 10 minutes, instead takes days because Mike can’t make progress until he meets with Jessica, who is locked up working on a high-priority project for the executives. If Mike had instead set up everything himself, he would understand how it all worked and wouldn’t need clarification from anyone else, at least for topics internal to his own department.

Disadvantages

The major disadvantage of the single-person department is key person risk. The kind of person who can run an entire department alone, even with modern AI, is rare and expensive. And if Mike sets everything up by himself, no one else will know how to step in and pick up the reins if Mike gets hit by a bus. For this reason alone, I doubt larger companies will ever want to set up single-person departments; they will always want some degree of redundancy to reduce key person risk. However, these companies could still significantly reduce the size of departments by hiring a small number of Mikes with overlapping scope. For smaller companies, the cost savings may be worth taking on some key person risk, and this risk can be mitigated somewhat by using strategies like awarding an equity stake with a long vesting schedule.

In the Longer Term

As AI improves, the single-person department might eventually be replaced by the single-person company. Entrepreneurs who have started a company from scratch may laugh at this and say they’ve already done it. But I’m talking about big figure revenue companies run by a single super-engineer owner, with fully featured product, marketing, accounting and customer service departments, each headed up by an AI agent.

This would require AI agents that possess conceptual understanding and real-world contextual knowledge, which is well beyond the currently available state of the art.