You founded it, and the shape of your week still shows that. There is a job title on the website and there are three jobs in your diary, and the third one exists because nobody else has ever done it. Fifty people now. A finance lead who is good, an operations lead who is better, and a set of client relationships that come to you by name because clients remember who answered them in the first year.
Somebody told you recently that this class of software is really for larger companies. Somebody else, possibly the same person over the same lunch, said the work your firm does is too specialised for it. You did not argue with either, partly because you had nothing to argue with and partly because both sounded right.
One of those two is a fair read of your position. The other is how a founder talks themselves out of a day a week, and it is worth twenty minutes to know which is which.
Specific work, ordinary hours around it
Start with the objection that sounds strongest, because it is the one you believe. Your work is too specific. It probably is. A firm that has reached fifty people has done it by knowing something a general practitioner does not, that knowledge is written down nowhere, and it is not in a training set.
Now look at where it sits in a week. The specialised judgement in most professional work is a small number of decisions made quickly by people who have made them many times before. Reading the survey and knowing the finding is wrong. Pricing the awkward job. Deciding this client gets the exception and that one does not. Those decisions are fast. Everything arranged around them is slow.
Around the judgement sits the chasing: the third email asking for the document promised on Monday. The assembling: pulling six figures out of four systems into the shape somebody downstream needs. The correspondence: the reply that says yes with conditions, written for the fourth time this month in slightly different words. The reporting: the same weekly summary nobody would miss for a fortnight if it stopped. None of that is specific to your firm. It is the same work in a specialist insurer, a translation bureau and a recruitment business, and it is the majority of the hours.
Which changes the question you should be asking. Not whether a machine can do your specialised work, but how many hours a week sit either side of the specialised part, and who is currently spending them. At fifty people the founder usually is, because the hours either side were never given to anybody, and you are the one who notices when they go undone.
There is a second objection under the first, and you were right about that one. In the Cabinet, ArkOne’s reference design for an executive agent, the room consults ten specialists, one for each standing concern of a company, and every one ships at propose and wait: the work is drafted in full and stops until a person answers. Authority is one word against each specialist, act or propose or escalate, written by a person on a date the Register, the standing page of every setting in the design, records. Ten is the roster the design offers. A firm of fifty should use one or two, and the reason is in the fourth row of the second table below.
Your week, and the same job at two sizes
Suppose your own last full week, reconstructed honestly rather than from the calendar, which flatters everybody. Four categories of hour. What matters below is which column an hour falls into rather than the totals, which are yours to count.
| The hours | What they are | Would a design take them | Why |
|---|---|---|---|
| The judgement | The four or five calls only you can make: the awkward price, the exception, the finding you know is wrong | No | It is the thing your firm sells, it is unwritten, and it is fast |
| Before the judgement | Chasing documents, assembling the figures, reading the file so the judgement can be made in ten minutes | Mostly | Ordinary work on your own material, checkable by you in a glance |
| After the judgement | Writing the reply, telling three people, updating the record, sending the weekly summary | Mostly, as a draft you approve | The decision is already made; what remains is carrying it |
| The hours that are not work | The meeting you attend because you always have, the call you take because they asked for you | No | A design cannot decline a meeting on your behalf, and should not |
Exhibit 2. Illustrative. One founder's week split by what the hours actually are, and the same recurring job at two company sizes.
The first row is the one to protect and the third row is the one to look at first. The judgement needs everything you know about the client and the trade, and it takes ten minutes. Carrying the judgement needs almost nothing you know and takes an afternoon, because it is four messages, two updates and a summary, each one easy and none of them finished.
The second row is where a founder loses the most and admits it least. Reading a file for twenty minutes so a decision can be made in ten is work whose output is a précis. A design that assembles that précis from your own documents, with the sources named, is doing the ordinary half of your specialised job rather than making the call.
One limit belongs in the same breath. A design assembling that précis discards a weak match rather than using it, because a passage that is merely nearby is worse than an empty answer, and the reference design draws that line at a similarity distance of 0.55, recorded on the Register. The consequence is unwelcome and direct: a design searching a thin corpus returns more empty answers, and a small firm’s corpus is often thinner than it feels, because the part that matters was never typed anywhere. A design that filled the gap by inventing a plausible passage would be worse, and inventing is what a design does when nobody set a floor.
The same job, at fifty people and at five hundred
Suppose a marine survey and cargo inspection practice, invented for this exercise, with one port office and surveyors who travel. Every week it issues condition reports to insurers and cargo owners, and each follows the same course: the notes arrive, somebody assembles them against the policy and the previous inspection, a senior surveyor makes the finding, and the report goes out. At fifty people the founder is the senior surveyor and also the person chasing the notes. Now suppose the same practice at five hundred.
| The step | At fifty people | At five hundred | Does size change it |
|---|---|---|---|
| Chasing the notes | The founder, on a Thursday, from memory of who is late | A coordinator whose job it is | No, the work is identical; only who does it changes |
| Assembling against the policy and the last inspection | The founder, twenty minutes a report | An analyst, to a template | No |
| The finding | The founder, ten minutes, unwritten expertise | Two senior surveyors, same ten minutes | No, and this is the part neither size delegates |
| Keeping the design current when the work changes | Nobody, or the founder at the weekend | A named person with it in their objectives | Yes, and it is the only row that does |
Exhibit 3.
Three of the four rows are the same job at both sizes, which is what the objection about larger companies does not survive. Chasing notes does not become a different activity at five hundred people. It becomes somebody’s job description.
The fourth row is where the difference is real, and it is a genuine disadvantage rather than a smaller version of an advantage. When your work changes, and it will, somebody has to notice that the design is now wrong about something and correct it. A large firm puts that in a person’s objectives. At fifty, that person is you or nobody, and you already have three jobs. So the honest recommendation for a firm your size is narrower rather than broader: one recurring piece of work, one specialist out of the ten the design offers, and everything drafted and waiting for a person.
Your firm does have the advantage that makes a narrow scope work. In a company of fifty, how the work is actually done sits in three or four heads and you own one of them. Ask a five-hundred-person practice how a condition report gets written and you get four answers and a process document describing none of them. Ask yours and you get one answer, from you, in an afternoon. The large firm has the person to write it down; you have something worth writing.
What to ask, and what both answers hide
You will end up in front of somebody selling this, and your objection will be the one you opened with: our work is too specific. Two replies are common, and each is suspicious in its own way.
The first reply is agreement. Yes, your sector is unusual, and that is why the work has to be configured to your process. Take this seriously, because it is often sincere, then ask the question that tests it: which parts of the job I have just described are the same as the last three companies you sold to. A seller who has built this before names them at once, because the chasing and the assembling and the reporting really are the same and the specific part really is small. A seller who cannot name a single shared part is either new to the work or is agreeing with you in order to price a bespoke build, and a bespoke build at your size is the version nobody looks after.
The second reply is denial. Every business says that, and the system handles specialist domains. This is the more dangerous one, because it is a category statement with nothing behind it and it is sometimes true. Ask to watch a search for a document your firm has never held. An empty answer is the finding you want. A confident paragraph about a file that does not exist tells you the system will fill every gap in your specialised knowledge with something plausible, and as far as it is concerned your specialised knowledge is mostly gaps.
Then two questions that are yours rather than anybody’s to sell you.
Who looks after it in a year, by name. Not a team and not a supplier. A person in your building, with hours in their week for it, and a stated answer for when that person is on leave. If the honest answer is you, the scope has to be small enough that looking after it costs an hour a month rather than an afternoon a week, and that is your decision rather than the seller’s.
What may it do without a person, on the first day and in the third month. Every specialist in the reference design ships at propose and wait, and for a firm your size that should hold for longer than a vendor will suggest. Say the cost of it out loud in the meeting: a draft that waits is a draft somebody reads, and at fifty people that somebody is you, so propose and wait gives you both the safety and the bottleneck. It is the right trade for a first quarter and it is the reason to start with one job rather than four.
Ask, last, where your own material would sit relative to the vendor’s. The reference design keeps four shelves apart, house knowledge and your documents and recorded failures and recent research, which the twelfth paper sets out and the same question put to a seller tests in the room.
One page, in your handwriting
Before you take another meeting about this, do the thing that costs twenty minutes and settles most of it. Take a sheet of paper. Write down, in order, every step of one recurring job you are personally the bottleneck in, from the moment it arrives to the moment it leaves the building. Put a mark beside each step only you can do.
Count the marks. Whatever the number is, it will be smaller than the number of steps, and the gap is the point. The unmarked steps are the answer to whether your work is too specific, written in your own hand rather than taken from a seller or from a page like this one, and the sheet is the only document a sensible scoping conversation needs.
Then leave it a week and see whether the marks are still in the same places. The ones that move were never the specialised part. If somebody inside the firm then offers to build the thing rather than buy it, the outline they will send you wants reading for coverage rather than for competence, and that is a different set of questions again.
component: null
Asked plainly
Are AI agents only worth it for large companies?
Size changes the honest scope rather than the answer. A large firm can put a named person on keeping an agent current, and a firm of fifty usually cannot, so the same design should be bought narrower: one recurring piece of work, one specialist, and everything waiting for a person before it goes out. What a small firm gains is that its own way of working genuinely sits in a few heads and can be written down in an afternoon. What it lacks is anyone whose job is to look after the result.
Our work is too specialised for AI. Is that a reason not to look?
It is usually an accurate statement about the work and the wrong conclusion from it. The specialised part of a job is normally a small number of judgements made by a small number of people. Around those judgements sit the chasing, the correspondence, the assembling and the reporting, and those are the same in every firm. The question worth asking is how many hours a week sit either side of the specialised part, and who is currently spending them, because in a founder-led company the answer is often the founder.
What should a fifty-person company automate first?
One recurring piece of work that happens weekly, that somebody already owns, and whose output a person checks before it leaves the building. Not the most valuable process and not the most interesting one. Pick the job you are personally the bottleneck in, run it for a month with every output waiting for a human, and add nothing else until a month has passed in which you did not have to think about it.