Our Take
The story is accurate but treats capability as novelty: AI has done basic admin work for months; the real news is cost-per-task has crossed the threshold where it beats hiring, not that the skill exists.
Why it matters
Small businesses operate on thin margins and cannot afford specialist hires. When AI handles $15k-$30k worth of annual admin labor internally, the math changes overnight for payroll and cash flow.
Do this week
Finance owner: audit your current invoicing, expense categorization, and meeting-note workflows this week so you can measure what AI actually saves before upgrading tooling.
AI now handles routine admin tasks at small-business scale
Current large-language models can perform a range of administrative functions that small-business owners typically handle themselves or hire for: organizing notes, summarizing meetings, generating invoices, setting goals, and planning social media posts (per MIT Technology Review's Making AI Work newsletter). The capability is not new; what has shifted is usability and cost-per-task, making it economical for operations with single-digit staff counts.
The article notes that large companies have long offloaded these tasks to dedicated hires. Small businesses lack that luxury and have instead absorbed them as overhead or left them incomplete. Current models now substitute for that gap without requiring new headcount or lengthy implementation.
Cost per task, not capability, is the forcing function
A small business owner making $50k annually runs at roughly $25 per hour fully loaded. Meeting transcription, invoice categorization, and social post drafting are low-complexity tasks that consume 5–10 hours per week across the team. At $0.01 to $0.05 per task via API, the math is stark: AI costs 1–5% of what the human labor costs today.
This is not a feature release; it is a cost crossover. The moment that happens, adoption accelerates because the decision moves from "nice to have" to "leaving money on the table if we don't." Small-business finance and operations are now in that window.
Test on your highest-volume, lowest-value task first
Start with whatever your team does most often and least wants to do: meeting notes, expense reports, or social media scheduling. Run it through Claude or ChatGPT for two weeks, measure the time saved, and compare the API cost to the hourly rate of whoever was doing it. Document the error rate and whether you need a human to review the output. If the cost is under 10% of the labor it displaced and errors are below your tolerance, expand to the next task. Do not boil the ocean; small wins compound.