At company scale, you don’t deploy an enterprise AI platform. You staff digital employees into seats, one job per seat, and you keep four kinds of work human. That’s the whole answer. Most of what’s sold as “enterprise AI agents” is a platform conversation — logins, integrations, governance slide decks — and platforms don’t do work. Jobs do. The companies that get value at scale are the ones that stop shopping for software and start drawing an org chart.
Some people call them agents. We call them digital employees, because at your size the word decides the whole project. “Agent” puts the conversation in IT: which of two hundred tools it connects to, what the usage caps are, where the audit log lives. “Digital employee” puts the conversation where the budget already is: what job, what number, who it reports to, what happens when it’s wrong. Same technology. Different verb, different meeting, different price.
This piece answers one buyer’s question fully: what to deploy first at company scale, the four tests to run on every vendor demo, what a deployment costs, and which work stays human no matter how big the rollout gets.
Why enterprise deployments stall
The pattern is consistent across every stalled rollout we’ve seen: someone buys a platform, gives it to a team, and measures nothing. Six months later the usage dashboard shows a spike for two weeks after the training and a flat line after. The team went back to what they were doing, because nothing in their job description changed. A platform makes people faster at tasks they still own. Speed without a handover just means the same human does the same work, slightly faster, forever.
The fix is the unit. Agents do tasks. Digital employees do jobs. A job has a scope, a trigger that starts it, a standard it finishes to, one number it’s accountable for, and a person it reports to. Take any one away and you have activity. When you staff a seat with a job, the Friday question changes from “is the team using the tool” to “did the number move.” That’s the only question a scaling deployment should be able to ask.
Scale makes this easier, not harder. A ten-person company has maybe two seats worth automating. A three-hundred-person company has thirty — prospecting, follow-up, bookkeeping, reporting, monitoring, intake, diligence — and each one has a number someone already owns. The enterprise buyer’s advantage is that the jobs already exist on paper, with managers and KPIs attached. You’re not inventing work for AI. You’re filling seats that are already scoped.
The deployment sequence that works at scale
- Map before you buy. One org chart, every seat, and the number each seat owns. Mark the seats nobody has filled — the follow-up that never happens, the reporting that arrives late, the books that get reconciled quarterly in a panic. Those are your first deployments, because they pay back the day they start.
- One job per seat. Not a bundle of ten capabilities. One outcome, one number, one reporting line. Go finer as you scale, not broader — more digital employees doing narrower jobs beats one doing everything badly. A single-purpose hire is easy to review, easy to fix, easy to expand.
- Scope, access, review — before it touches anything. Scope: what job, specifically. Access: the smallest set of permissions that lets it do the job, because removing a permission is a hard guardrail and an instruction is only a request. Review: who checks its work, how often, and what happens when it’s wrong — decided before it is. If you have a security review process, this is what it should be reviewing, seat by seat, not platform by platform.
- Probation, then scope growth. Onboard it like a hire: written job description, narrow first task, a two-week period where it does the full job but nothing leaves the building without a human saying yes. Daily review at first, less once it has earned it. More responsibility only after the number has moved. You already know how to do this. You have been managing people for years.
The four tests for every vendor demo
Run every enterprise “agent” pitch through these. They measure whether your team can walk away:
- Does it start itself? The trigger is part of the job — a new lead, a new recording, the clock hitting eight. If a human has to open it, prompt it, or feed it, you bought a task with a human at the front of it.
- Does it finish? To the standard you’d accept from a person you pay. Not a draft for your team to check and carry. Eighty percent and a handover is not finishing.
- Does it have one number? “It does lots of things” is what a platform says. A seat in a company has one outcome a manager can look at on a Friday and say yes or no.
- Does it report to a person? When it can’t hit the number, a person hears about it — from the employee itself, with a reason and a proposal. If it fails quietly, it fails, and at your scale it fails in front of customers.
Most enterprise demos fail the first test on stage and nobody notices, because the demo is a human driving. Ask the vendor to leave the room and let it run. (The full framework — role, goal, trigger, reporting line — is in What Is an AI Employee?, the hub for this subject.)
What to deploy first
The seats that pay back fastest at scale all touch revenue, retention, or risk, and they share one shape: repeatable work with a number.
- Follow-up and chasing. Every lead contacted within the hour, every quiet thread chased, every conversation moved toward a booked call. The leads are already paid for; sitting leads are the most expensive inventory in the building.
- Reporting and monitoring. The numbers pulled on schedule, the anomalies flagged, the report that arrives instead of the dashboard someone has to remember to open.
- Bookkeeping and reconciliation. Watching the books full time — the work that currently happens quarterly in a panic.
- Intake and diligence. The reading and prep work that makes your senior people’s first calls second calls.
The seat-by-seat versions of this, in depth: AI Agents for Business for the map across the business, AI Sales Agent and AI Customer Service Agent for the two seats most companies fill first. If you’re scaling a team of knowledge workers rather than seats, AI Workers covers that side.
What it costs at scale
Two numbers per seat.
What it costs to run: about two dollars a day per digital employee on the recommended open-source frontier models. At fifty seats that’s a rounding error against the payroll it sits next to. Discount it to zero in your head — what you’re paying for is the expertise in the job description, not the electricity.
What it costs to buy: what you’d pay a human for the month, you pay the digital employee for the year. Take the fully loaded monthly cost of the role the seat replaces — salary, taxes, benefits, overhead — and that monthly number is the annual fee. A bookkeeper at $4,375 loaded a month is $4,375 a year. An SDR seat at $6,250 a month is $6,250 a year. Year two is the same fee — it’s on salary, and it gets better every year as the models move. Model and tool bills are on top, at cost, paid by you directly. Minimum $3,600 a year per seat, and one number all in: design, build, deployment, support, and a year of running.
That pricing rule comes from our own program — we train and certify the people who build these, and the deployment program is here if you’d rather hand the whole thing over.
What stays human — the part scale can’t change
There’s a version of this pitch that ends with “and then you don’t need the people.” Reject it. Not for sentiment — for arithmetic. AI deployed into a broken process amplifies the breakage. Work nobody can call good or bad cannot be managed. Four kinds of work stay human, at any headcount:
- Where the relationship is the product. The rescue call for a top account. The apology that has to be meant. Billing conversations for customers worth losing. The digital employee’s job is to make sure the human walks in prepared, and then to hand it over.
- Where a wrong answer can’t be reviewed before it costs you. Regulated, financial, medical, reputational. Either build the review queue or keep the human in the seat. Never deploy where “I screwed up” arrives after the damage.
- The moment of trust. Every AI output that matters is checked by a human, and you should say so out loud, in the deployment docs, on day one. These are AI employees; they can be wrong confidently.
- The owner’s job. What the business is, the offer, who it’s for, what to refuse. AI produces the average of everything ever said. The idea that makes the year is not in the pile.
The market context, briefly: 92% of companies plan to increase AI investment, and about 1% of leaders call their company mature at deploying it (McKinsey, “Superagency in the Workplace,” January 2025). IDC counts 2.5 billion active agents used by organisations by 2030, nearly 80× the 2025 figure. The gap between those numbers is the whole opportunity — and the gap doesn’t close with a platform purchase. It closes seat by seat, with a number on each one.
The Monday move
Take one unfilled seat, one number, one reporting line. Deploy it on probation. Review daily for two weeks, then let go of the review. When the number moves, staff the next seat. That’s how a deployment scales: not a rollout, a hiring plan.
Want this deployed in your business instead of built by your team? We diagnose the jobs, scope the seats, build the digital employees and hand them over. See how deployment works →