Uber burned through its entire 2026 AI coding budget in four months. The number that should worry a small business owner isn't Uber's — it's the line right after it in Forbes' teardown of the story: "Enterprise budgets contain a cushion no small business has. If the AI costs for small businesses were treated similarly to Uber's, they would create payroll pressures for a 12-employee firm" (Forbes).
A big company absorbs a bad AI month with a policy memo. A 12-person mortgage brokerage, real estate agency, or aged care provider feels it straight against payroll. And these are exactly the businesses moving fastest — NAB's 2026 research shows AU firms in finance, property and business services are adopting AI at 2 to 3 times the rate of other sectors (NAB). MIT's GenAI Divide study, meanwhile, found 95% of enterprise AI pilots deliver zero measurable financial return (Fortune). If businesses with finance teams and dedicated budgets get this wrong that often, a small firm running its first agent off a personal login has even less room to.
Three ways this goes wrong
Token spirals
Agents remove the one spending brake chat-based AI had — a human could only type so fast. An agent can retry a failed task all night, burning money the whole time. One team watched an agent burn $2,847 in four hours stuck in a loop, while every monitoring signal looked normal: API calls returning 200 OK, latency fine, only the token counter telling the real story (n1n.ai). For Uber, that's noise. For a five-person agency, it's the month's AI budget gone in an afternoon.
Subscription creep with no "decision moment"
"Subscriptions build upon each other... no single 'I'm making that decision' moment" (Forbes). Around 64–65% of employees admit to using AI tools their employer hasn't approved, often feeding in client files or loan documents through free tools with no governance (WatchGuard). In a small business, the owner is the IT department — five staff quietly expensing personal subscriptions can cost as much as one governed account, minus the audit trail.
Nobody's watching the meter
One AI consultant told Axios a client burned roughly half a billion dollars in a month simply because nobody set a usage limit (Forbes). Drop a few zeros and it's the same mechanism at SMB scale: most small businesses review the books monthly at best, giving a runaway agent weeks to compound before anyone notices.
Three fixes that fit a small business
Ask if you need an agent at all
Not every task needs autonomous AI. Extracting three fields from a PDF, sending a scheduled reminder, or routing a form submission is often better handled by a simple script or a rules-based workflow tool — it costs fractions of a cent per run, never loops, and doesn't hallucinate. Save agents for work that genuinely needs judgment or handles messy, unstructured input. And where AI is the right call, match the model to the task: a routine summary doesn't need your most expensive model — save that for work that actually pays for itself. Our guide to what AI automation costs breaks down the price gap between a simple rules-based automation and a full agent.
Set the cap before the tool goes live
Decide the monthly ceiling before signing up, not after the invoice. Use built-in spend caps where they exist — Anthropic added budget alerts to Claude Enterprise for exactly this reason (Forbes). If a vendor doesn't offer one, the AI line item needs the same monthly discipline as any other recurring cost.
Cap and trace
A cap limits the damage; a trace tells you if the spend was worth it. Once a month, put AI spend in one column and hours saved or deliverables produced in the other.
The bottom line
AI agents genuinely pay off for small businesses — it's why mortgage brokers, commercial finance brokers, real estate agencies and aged care providers are adopting faster than almost anyone. But the gap between "pays for itself" and "quietly eats your margin" comes down to whether someone asked if AI was even the right tool, capped the spend, and checked the numbers. Big companies can afford to learn that the expensive way. Small businesses can't.
Weighing up an AI rollout and want the right tool — agent, automation, or plain code — chosen for the job from day one? That's exactly the work AutoCognition does. Get in touch.
Not sure whether you need an agent or an automation?
Get your free AI Assessment. We'll map your processes and tell you which tasks justify an AI agent — and which are cheaper, safer and better handled another way.