What Should a Solo Operator Running on Agents Be Unusually Good At?

What Is a Solo Operator Actually Unusually Good At?

It is Tuesday and the support agent has been quoting the wrong refund window for nine days. Nobody complained loudly enough to notice. The operator who catches that reads a sample of outbound messages every week, because they assume drift the way a mechanic assumes wear. That habit, repeated across five functions, is most of the job.

Triage is the half that is hardest to buy: knowing which function is currently the problem when the only visible symptom is a flat week. Revenue is soft. Is it top of funnel, a broken checkout, an agent answering pre-sales questions wrongly, or a delivery backlog that quietly stopped renewals? A capable operator narrows that in a day with three cheap checks. A weaker one starts rebuilding the marketing automation because that is the part they enjoy.

The tell that separates real from performed is what they check and how often. Ask what broke in the last month and how they found out. The real answer names a detection path: a weekly read of the last fifty support replies, which is how a wrong refund window surfaces on day nine rather than day ninety; a spot check of invoices against the bank feed; a customer call that gave it away. The performed answer names a tool, then a diagram, then hours saved. Hours saved is the least interesting number in a small business, because an agent that produces wrong output faster has saved nothing.

The same person can say plainly which parts of their own operation they understand well enough to fix and which parts they only understand well enough to notice. Holding a whole business at shallow depth without pretending to depth is rarer than it sounds, and that sentence is worth listening for when it arrives unprompted.

Then there is the line they draw and keep. One-person companies run on agents hit real limits exactly where relationships, accountability and judgment are required ³. An operator who has already drawn that line, and can tell you where it sits and why, has run something. Pricing exceptions, firing a customer, anything that involves an apology, and anything a regulator might read later tend to end up on the human side of it.

Look Past the Resume Screen to the Charge Nurse and the Restaurant GM

A charge nurse would have found the refund window on day two. Noticing that a standing instruction has quietly gone wrong, under incomplete information, while four other things are also happening, is the entire shape of the shift. That is the capability this hire is for, and it is filtered out of your pipeline by a keyword screen before a human ever reads it.

Restaurant general managers run inventory, staffing, marketing, cash and complaints inside a single shift, and they triage by reflex. Film and event line producers hold a budget, a schedule and forty vendors with no authority over any of them. Charge nurses do continuous triage under incomplete information and are trained to escalate rather than guess. Independent insurance adjusters investigate, document and defend a judgment call all day. None of these people will have an AI title, and several will be better at the actual work than someone who does.

The conventional feeders are real too, and they arrive by four routes. Agency account managers carried five clients across every function. Operations leads at seed-stage startups worked where nobody had a title yet. Ecommerce owners ran their own fulfillment and their own ads. Technical support leads wrote both the macros and the runbooks. Each learned range against a real deadline, which is the part no course produces.

What they bring differs in useful ways. The agency person is fluent at context switching and at explaining a miss to someone paying for it. The startup ops lead is fluent at building a process that is exactly as heavy as it needs to be. The ecommerce owner has felt the margin consequence of a bad decision in their own bank account, which changes how they treat an agent's confident recommendation. The support lead knows what a customer sounds like two messages before they churn.

What transfers less well than expected: deep single-function specialists who have never owned an outcome outside their function, and candidates whose entire preparation is prompt technique. The ability to get a good first draft out of a model is now common. If what you actually need is someone to build durable instruction and retrieval scaffolding for the agents themselves, that is a different hire, closer to a context engineer.

Ask How This Operator Got Good With Their Own Agents

The strongest candidates got good by running an operation on agents and watching it fail in specific ways. Ask what they delegated to a model, what it got wrong, and what changed afterward. The answer is either a concrete story with a failure and a fix, or it is nothing. Vague enthusiasm about productivity is a signal, and not the one you want.

Listen for the crossing pattern. In OpenAI's economic research, reported by Axios, 43.5 percent of occupation-specific work messages to ChatGPT concerned tasks associated with a different occupation ². That is the mechanism behind the whole operating model: the operator does bookkeeping-shaped work on Monday and copywriting-shaped work on Tuesday without being a bookkeeper or a copywriter. Solo founders in 2026 describe an org chart of one human over agents handling research, content, code, sales and support ¹. The risk that comes with it is obvious once said out loud. Nobody in the building is qualified to catch the mistakes, which is how a refund window stays wrong for nine days.

So screen for the catching. A working screen takes an hour, not a take-home week. Give the candidate a messy real artifact from your own business, redacted: a month of support transcripts, an ad account with a step change in it, or a reconciliation that does not tie. Give them an assistant. Ask for a written account of what is wrong, what they checked, and which of their conclusions is soft. What you are reading is whether they verified anything against something outside the conversation, and whether they labeled their own uncertainty without being asked.

Two things not to screen for. The ability to tell whether an application was written with a model cannot be done reliably, and it predicts nothing about operating judgment. And volume of tools: a candidate running eleven automations is not ahead of one running three that they check. If a specific function in your business is the one that keeps failing, hire into it directly rather than hoping range covers it, the way a support-heavy operation eventually needs an AI voice support operations specialist.

Find Them Where They Already Run Something in Public

They are not on job boards, because most of them already work for themselves. They are in the places where people run small businesses in public: Indie Hackers, MicroConf and its bootstrapped-founder circles, Product Hunt maker profiles, and the subreddits where owners argue about fulfillment and ad spend. The strongest signal is a public trail of an operation they ran, including the parts that went badly.

Adjacent titles that already hold most of the skill: agency account director, head of operations at a company under thirty people, ecommerce store owner, technical program manager at a small firm, and freelance producer. A person who spent a year running their own thing and is now open to a role is often available for a reason worth asking about plainly, and the reason is usually cash flow or isolation rather than failure.

What they care about, in the order it comes up: scope that stays wide, authority over their own stack, and a clear picture of the numbers. Someone who has run a business alone is used to seeing revenue, costs and churn every day, and being cut off from that view reads as a demotion even when the pay goes up. They also care about who owns the outcome. A person accustomed to being the last line of defense wants to know which decisions remain theirs.

The offer usually dies in one of three ways. Boxing them into one function after hiring them for range, which is the most common failure and usually happens by accident in the first month. Requiring approval for every tool purchase, which turns a self-directed operator into a ticket filer. And leaving accountability vague, so the operator carries the blame for outcomes they were not allowed to decide. Put the wide scope in writing, name the two or three decisions that need a second signature, and say what the tooling budget is before the offer goes out. Reliability ownership deserves the same explicitness, which is why a growing agent stack eventually justifies an AI SRE rather than another generalist.

Expect to Price Against a Senior Operations Manager, Then Add Upside

No published compensation series exists for this role, because the title is not one a survey has counted yet, and every point estimate circulating for it is somebody's guess dressed as data. This piece will not offer one. Build the band from comparables you can actually verify: what you pay a senior operations manager in your market, and what the candidate currently earns from their own operation, which they will usually tell you if asked directly.

Two forces pull in opposite directions and you should price with both in view. The range is genuinely rare and the person is replacing more than one function, which argues up. The candidate is often coming from unpredictable self-employment income and values stability, which argues down and is exactly the reason to be careful about lowballing. An operator who takes a number below what the role is worth tends to leave in the year, and the cost of that is the entire operating context they were holding, including the weekly habits nobody wrote down.

Equity and variable pay matter more here than in most roles. Someone who ran a business alone is used to owning the upside, and a purely salaried offer with no participation in what they build reads as a step backward. If equity is not available, say so early rather than at the end, and compensate with scope and autonomy instead.

On location: this work is remote by default and mostly async, because the artifacts are documents, dashboards, agent configurations and customer conversations. Three things pull it onsite. Physical delivery or inventory, where somebody has to look at the shelf. Regulated or restricted data that cannot leave a controlled environment, which changes both the tooling and the audit trail. And cash handling or anything with a signature requirement. If none of those apply to your business, requiring an office is an unforced constraint on a candidate pool that is already thin, and remote-native operators will simply keep working for themselves.

See a sample report >