AI agents vs hiring employees cost comparison: the line-item worksheet.

• 1.1

  Salary is not the cost.

  Treat a salary as the total and the comparison is wrong before it starts: payroll taxes, benefits, equipment, software seats, recruiting time, and a ramp of months all sit on top. The honest unit is loaded cost — everything a hire costs, not what they're paid. Pages that compare a salary against a subscription are comparing one line against four.

• 1.2

  Monthly is not annual.

  Comparison pages flip between a monthly figure and an annual one whenever the smaller number flatters their side. Pick one period, convert once, label it. A stack quoted at $400 a month is $4,800 a year — and it should never reappear as "$400 a year" or "$4,800 a month" further down the page, in either direction.

• 1.3

  Vendor math ends in the vendor's pricing table.

  Nearly every page ranking for this comparison is selling agents — their loaded costs, their stack prices, and their ROI all route to one destination: their plans page. A neutral worksheet has no plans page. Its only product is the course that teaches you to run the math yourself, and it says so up front.

The worksheet contract, stated once: figures that come from reporting carry their source inline; figures that depend on your business appear as blanks for you to fill. Nothing in between.

• 2.1

  Line 1 — the salary.

  The number in the offer letter. It anchors every other line: payroll taxes are a percentage of it, benefits are quoted against it, and the ramp months bill it in full while output is still partial. Write yours on the worksheet; nothing below works without it.

• 2.2

  Line 2 — the statutory lines.

  Payroll taxes and employer insurance load on top of gross pay at rates that vary by state and situation — which is exactly why this worksheet asks for your rate instead of quoting an average that isn't yours. Your payroll provider or accountant can state the number to the decimal; an honest comparison page can't.

• 2.3

  Line 3 — benefits, equipment, and seats.

  Health coverage, retirement match, the laptop, and the software licenses a human needs that an agent doesn't. Recurring, mostly monthly, always per head — and the reason two hires at the same salary can carry meaningfully different totals.

• 2.4

  Line 4 — recruiting and ramp.

  The hire costs money before they produce: posting, screening, interviewing, onboarding, and months of ramp to full output. Amortize that one-time line over the months you expect to keep the role — that's the honest monthly figure, not a sunk cost the offer letter hides.

• 2.5

  The reporting scale, for orientation.

  Per Forbes reporting from August 2026, a comparable human team at salary scale runs $80,000–$120,000 per month, against a complete working AI stack of roughly $3,000–$12,000 a year. That is the ten-person company's arithmetic — orientation, not your worksheet. Your Side A is one role's lines, and only your numbers finish it.

• 2.6

  Your number here: L.

  Sum the lines: salary, plus statutory lines, plus benefits and seats, plus amortized recruiting and ramp. That monthly total — call it L — is Side A of the worksheet. A comparison page that skips straight to a total without these lines isn't doing math; it's hanging a price tag.

Sources cited inline: Forbes. Published comparisons disagree wildly on loaded cost because the lines vary by state, role, and benefits package — which is exactly why this worksheet computes yours instead of quoting theirs.

• 3.1

  Line 1 — usage.

  Model calls and automation runs meter with the work: more drafts, more triage, more outreach, more tokens. Usage is the healthiest line on this side of the ledger — it scales with value delivered — but it still goes in the total.

• 3.2

  Line 2 — subscriptions.

  Chatbot seats, automation platforms, outreach tools. Independent reporting puts the full working stack at about $300–$500 a month (per aibusiness.vc's one-person-company field report), and Forbes reporting puts a complete stack at $3,000–$12,000 a year. Below the full stack sit two lighter tiers you assemble yourself: $0 on free plans, or about $50 a month focused. Pick the tier the task list needs; the tiers are inputs here, not a shopping list.

• 3.3

  Line 3 — one-time setup.

  Wiring the workflow: the prompts, the connections, the first supervised runs. It's a real cost with a real payoff schedule, so amortize it over the months you expect the workflow to run — the same treatment Line 4 on Side A got.

• 3.4

  Line 4 — supervision, the line vendors skip.

  Every agent output gets read before it ships. Price those checking hours at what your hour is worth and put them on the stack's side of the ledger — a stack that saves an hour but costs ninety minutes of reviewing is a loss with a subscription fee. Vendors omit this line because it's the one that sometimes flips the answer.

• 3.5

  Your number here: S.

  Usage, plus subscriptions, plus amortized setup, plus priced supervision — per month. That's S, Side B of the worksheet, and unlike Side A it is mostly knowable before you spend: free tiers and starter plans publish their prices.

Both sides now exist as line items, which is more than any page this one competes with can say. The comparison happens in the next section, in four steps.

1. 4.1 — Gap

   Compute the monthly gap

   Monthly net saving = L − S, where L is Side A's monthly cost for the hours the agent absorbs and S is Side B's true monthly cost. If the agent absorbs only part of the role, L is the loaded cost of the absorbed hours — not the whole role's price tag.

2. 4.2 — Payback

   Compute the payback period

   Payback period = one-time setup cost ÷ monthly net saving. A workflow that nets $200 a month against $600 of wiring pays back in three months. The same workflow netting $40 a month never really does — the formula's job is to tell you which one you built.

3. 4.3 — Period

   Respect the period

   Compare monthly with monthly, or annual with annual — convert once, label it, and don't let the page you're reading do the converting for you (that's trap 1.2). The stack's $400 a month and the hire's annualized lines meet only if you put them in the same period on purpose.

4. 4.4 — Verify

   Count only real hours

   Run the workflow supervised for a week and log every miss before its savings enter the worksheet. Never annualize a good month: hours times 52 is the honest annual number only after the workflow has survived the log. The math is only as honest as the output it counts.

The boring baseline still holds, and it is the only ROI claim on this page: a workflow that reliably saves two hours a week has already paid for the $30 course that teaches you to build it.

• 5.1

  Example 1 — the contractor's ten hours.

  Say a contractor covers your admin at ten hours a week and $25 an hour: $250 a week, $13,000 a year, about $1,083 a month of loaded cost to hand over. A free-tier or single-tool stack covers the tooling for that task list, and even after you price your own supervision hour honestly, the annual gap remains the contractor's invoice minus a stack bill measured in hundreds. Payback on a weekend of wiring: weeks.

• 5.2

  Example 2 — the first hire, deferred.

  The cleanest win in this math is the invoice that never happens: the task list a first hire would have owned, absorbed by a stack instead. It's also where one-person companies actually form — the Census Bureau counted 29.8 million US businesses with zero employees in 2022, the majority of all US establishments, and the Nasdaq Economic Institute found new one-person business applications up more than 20% since early 2025, with AI the named driver. Your worksheet runs the same decision in miniature: if Side B covers the task list at a fraction of Side A, the hire waits.

• 5.3

  Example 3 — capacity, not replacement.

  The third example breaks the framing: nobody gets replaced, because there was no one to replace — the stack does work you'd never have hired for. The US Chamber of Commerce found 82% of small businesses using AI increased their workforce over the past year; adoption is showing up as leverage on people, not a swap for them. Worksheet version: the outreach workflow costs S and returns replies — the average cold email earns a 3.43% reply rate and the top senders clear 10.7%+ (Instantly's 2026 benchmark), while Backlinko's 12-million-email study puts overall response at 8.5% — so the revenue side of the math is sourced too.

Every rate above is yours to replace: the contractor's hours, the tier, the supervision price. The sources — Census Bureau, Nasdaq Economic Institute, US Chamber of Commerce, Instantly, Backlinko — are the only fixed points, and they're cited where they appear.

• 6.1

  Cost is one input.

  A stack can clear the break-even math and still be the wrong call. Whether agents should take over a role at all — which parts of the work are actually theirs to take, and which stay human no matter what the worksheet says — is the question our second field guide answers: the replace-or-not decision guide. Run both pages before you decide; this one prices the move, that one judges it.

• 6.2

  Even hyperscale miscounts this math.

  Per Ars Technica's August 2026 reporting on Meta's scrapped AI-native plan, some teams were slated for cuts of up to 60%, while agents were reportedly making what the report called "large-scale, disruptive" actions. Two sentences, one lesson: the supervision line you priced in section 03 is not optional.

Already priced it and ready to build? The five-step first deployment — with the weekly supervision ritual — is the subject of our first field guide: the full deployment guide.