Five ways to get AI work done.
Who each one is really for.
Four of the five are better than us for at least one kind of company. This says which, and names the trade-off in every column including our own.
There are five ways to get AI work done inside a small company. Four of them are better than us for at least one kind of company, and this guide says which.
Most comparison pages are written to make one column win. This one isn't, because the version that makes us win every row is useless to you and you can tell within about nine seconds.
So here's the honest shape of it. The criteria first, then the five options, then who each one is genuinely the right call for. If the answer isn't us, that's a fine outcome and you saved a call.
Six criteria that actually decide it
Not features. These are the six things that determine whether the work survives contact with a real business.
Does it fit the systems you already run?
Almost nothing in this category is a greenfield build. The work is connecting a CRM, an inbox, a calendar and two spreadsheets that were never designed to talk to each other.
Count the systems. That number drives cost, timeline and risk more than any other input, in every column below.
Who is accountable on the Tuesday after it breaks?
Name a person. Then check whether it's already their job or an eighth hat they're wearing.
An unowned system is the most expensive outcome available. You pay for it twice. Once to build it, and again to work around it.
Does it survive the person who built it?
Ask about a specific person, not in the abstract. If the honest answer is that the work stops when they're out for two weeks, you don't have a system. You have an arrangement.
How long until it's actually running?
Not until it's bought. Until output lands somewhere a person reads it, on a schedule, without anyone starting it by hand.
Is the hard part the knowing or the wiring?
The judgment is usually the easy half, and it's the half that's public. Anyone can learn what to check.
What costs weeks is connecting systems that were never designed to talk, and keeping them connected when one of them quietly changes a field name.
Mostly judgment points inward. You already have the judgment, and it transfers badly from outside. Mostly wiring points outward.
What does a month without it cost you?
A rough number is fine. An order of magnitude is fine.
If you can't produce one, don't start. A build with no cost of delay attached loses to every urgent thing for as long as it exists. That's true whoever does the work.
The five options, side by side
Read the trade-off column. It's the only one that tells you anything a sales page wouldn't.
| Option | Best when | Time to running | The real trade-off |
|---|---|---|---|
| Build it in-house | Someone already built a working version that a second person has used | Depends entirely on whose week it is | It stops when they leave, unless you fund it as a real job with a review schedule |
| A SaaS tool | Your problem is genuinely the same as everyone else's | Days, if your process already matches the tool | You adapt to the tool. Fine for email. Expensive for anything shaped like your business |
| A full-time hire | This is a permanent function, not a project | Three to six month ramp | Highest fixed cost and highest ceiling. Wrong if you can't describe the job past year one |
| An agency | You need breadth, volume, or several disciplines at once | Varies with their queue, not your urgency | You get an account manager, not the person who built it. Depth suffers |
| A freelancer | The scope is clear, bounded, and ends | Fast, often the fastest | Nobody owns it after handover. Excellent for a build, poor for a system |
| MarqueeOS | Multiple systems, nobody internal to own it, and it has to keep running | Two weeks to a written scope, then weeks not quarters | One person, so there's a ceiling on volume. We're the wrong call for breadth |
So which one
If someone in your building has already built a working version, fund them. This is the strongest signal on the list and it points away from hiring anyone, us included. Give them the time formally, put a review schedule on the output, and write down how it works so it survives them. That's a better position than most companies are in and almost nobody recognises it as one.
If your process is genuinely standard, buy the tool. Payroll, email, scheduling, invoicing. If a category exists and thousands of companies run the same shape of it, paying someone to build you a custom version is a way of setting money on fire slowly.
If this is a permanent function you can describe past year one, hire. The ramp is real and the fixed cost is real, but the ceiling is higher than any outside arrangement and it compounds. The failure mode is hiring for a job you can't describe yet, then asking that person to invent it while also doing it.
If you need six things at once across several disciplines, use an agency. Breadth is what they're built for and it isn't what a single operator can do. Ask who is actually doing the work and how much of their week you get. Then ask again in month three.
If the scope is clear and it ends, use a freelancer. Fastest route to a finished thing. Just be honest about whether it ends. Most of these don't, and a build handed over with nobody accountable degrades on a schedule you don't control.
If it touches four systems, nobody internal can own it, and it has to still be running in six months, that's the case we're built for. That's a narrow description on purpose. It excludes most companies.
Question one loses us more deals than every other factor combined. When there's already a builder inside the company, the internal version wins, and it usually should.
We ask it in the first conversation now, because finding out in month two is expensive for both of us. If you're reading this and the answer is yes, the guide you actually want is the seven questions on building in-house.
Two things that are true of every option
Price it per connected system, not as one number. Cost scales close to linearly with how many systems the work has to read from or write to. A quote given without asking how many systems are involved hasn't been scoped, and that's true of an agency, a freelancer and us. Ask for the breakdown. Watch what happens.
Ask what happens when a connected system changes. Fields get renamed. API versions retire. Permission scopes narrow. Anything that ends at handover is a build that degrades on someone else's schedule, and that failure is silent. It keeps producing output. The output just stops being true.
More on that in the four ways an AI system fails without telling you and what an agent actually costs to install.
Keep reading
The 6-Point Check For AI Work Before It Leaves Your Company
Most AI failures don't look like errors. Here's how to grade output you didn't build, and turn it into an error rate you can actually act on.
Read the guide → Operations · 6 min readThe 30-Minute AI Inventory: What Is Running In Your Business And Who Owns It
Four questions, four groups, one specific order. Surfaces the undocumented setup carrying real volume before the person running it resigns.
Read the guide →Still not sure which
column you're in?
Thirty minutes, no deck. Bring the process and the list of systems it touches. If the answer is one of the other five, we'll say so on the call.
Book a 30-minute call →