Azimuth Labs
← All insights
AISmall BusinessOperationsLeadershipProcess

The Rigor Has to Go Somewhere

AI can draft your contracts, patient notes, listings, and client emails in seconds — but the careful judgment that work used to require doesn't disappear. It moves to deciding what's worth doing, checking the result, and owning the outcome. The businesses that win decide where that responsibility lands instead of letting it slip through the cracks.

By Azimuth Labs7 min read

A paralegal pastes a clause into a chatbot, and a contract appears in thirty seconds. A dental office turns a recorded visit into a tidy clinical note before the patient reaches the parking lot. A realtor writes a dozen listings over a single coffee. Work that used to eat an afternoon now takes a minute, and it feels like pure gain.

Sit with that feeling for a second, because it hides a question worth answering: where did the careful part go?

It did not vanish. When AI does the doing — the drafting, the calculating, the first pass — the rigor that used to live inside that work doesn't evaporate. It migrates. The phrase comes from a 2026 gathering of veteran software engineers wrestling with the same problem in their own field: if the machine writes the code, where does the engineering go? Their answer generalizes well past software. The rigor has to go somewhere. It moves to deciding what's worth doing, to writing down what "good" actually means, to verifying what came back, and to owning the result when it reaches a client. The only real choice is whether you route it there on purpose or let it leak out through the gaps.

The Staircase Problem

AI does not redesign how your business works. It runs your existing process faster.

Picture your workflow as a staircase. AI doesn't build you a better staircase — it lets you climb whatever staircase you already have at a sprint. If your client intake is sound, you reach the right floor sooner. If it's sloppy, you reach the wrong floor faster, with more confidence, and with more cleanup waiting at the top.

This isn't just a tidy image. DORA's 2025 research into how teams ship software found that the same AI tools pull in opposite directions depending on the ground they land on: paired with strong delivery practices, AI improved stability; paired with weak ones, it degraded it. As one practitioner put it, AI is an accelerator of whatever you already have. For a small business that means AI is not a fix for a broken process. It's a multiplier on the process you've got. A clinic with a real review step gets faster and safer. A clinic without one gets faster and more exposed.

The Cognitive Debt Nobody Bills You For

Here's the trap. The speed feels like progress, and the cost is invisible until it isn't.

In a careful study by METR, experienced developers were about 19% slower when using AI tools — while believing they were roughly 20% faster. The people doing the work could not tell the difference between feeling fast and being fast. Other research points the same way: a Carnegie Mellon team found AI raised short-term output while the complexity left behind quietly accumulated, and a Microsoft-led study found the clearest gains showed up for less-experienced people on well-defined tasks, not for everyone everywhere.

The lesson for an owner is uncomfortable: you cannot manage this by asking your staff whether AI is helping. Their honest answer will not be reliable. You have to look at the work itself.

There's a name for what builds up underneath: cognitive debt — what a team loses when it offloads so much that it stops understanding its own work. The output keeps flowing while the understanding drains away. In a small business that's especially dangerous, because the understanding usually lives in two or three heads. When AI handles intake, billing, and client replies, and no one on staff could reconstruct why a number is what it is — or catch the moment it's wrong — you've taken on a debt that comes due the first time something breaks: an audit, a complaint, a refund dispute, a deposition. A controlled trial at Anthropic found that AI-assisted engineers scored lower on understanding their own systems, with the biggest gap in debugging — the exact skill you need when things go wrong. The pattern travels: AI can quietly erode the muscle you most need on the day it fails you.

And most businesses are measuring the wrong thing entirely. They track adoption — how many people use the tool, how often. Usage tells you nothing about whether the work got better. It only tells you the staircase is being climbed faster.

The Cambrian Explosion of AI Agents

The pressure isn't confined to one team, and it's no longer just about chatbots that talk. Andrej Karpathy named the era in early 2025 with "vibe coding" — describing what you want in plain language and letting the AI build it, barely inspecting the result. That habit has spread far beyond code. The market for AI agents — software that doesn't just answer but acts on your behalf — is now measured in the billions and multiplying fast.

In practice, that means everyone in your shop is suddenly a builder. Your bookkeeper can spin up a dashboard. Your office manager can wire together an automation. Your marketing lead can ship a microsite by the weekend. Anyone can produce work that used to require a specialist — which is genuinely good, right up until you ask who owns it.

  • When your office manager automates appointment reminders, who confirms it handles cancellations and reschedules correctly?
  • When your bookkeeper's AI-built spreadsheet feeds a tax filing, who checks the formulas?
  • When a contractor's AI-drafted bid goes out, who's accountable if the numbers are wrong?

A 2026 study of "vibe coding" in practice found 36% of people skipped quality checks entirely, roughly 18% trusted the AI's output uncritically, and about 10% handed the checking back to the AI itself — a loop with no human in it. The researchers warned this is producing a new class of people who can create things they cannot fix. That description now fits a lot of small businesses, not just software teams.

Which is why the most useful framing we've heard comes from engineer Kent Beck: when anyone can build anything, knowing what's worth building becomes the skill. For an owner, that's the whole game. The constraint was never the typing. It was judgment — what's worth doing, what "right" looks like, and what you'll put your name behind.

Where to Send the Rigor

The rigor is going somewhere whether you manage it or not. Three moves point it somewhere useful.

Measure outcomes, not usage. Stop counting how often your team uses AI. Track whether the work is correct: error and rework rates, things sent out that had to be pulled back, client complaints — the things you'd actually be judged on. METR's finding is the warning here: even the person doing the work can't feel the difference, so you have to look at the output, not the activity.

Name an owner for every AI-touched output. The review step isn't an afterthought anymore — it's the job. Decide explicitly who signs off before an AI-drafted contract, clinical note, listing, filing, or client message leaves the building. "The AI did it" is not a defense to a client, a regulator, or a court. A named human owns it.

Invest where the rigor moved. The valuable skill is no longer producing the first draft; it's specifying what you want, checking what came back, and judging whether it's good enough to ship. Write down what "good" looks like for your handful of most important outputs — a checklist a new hire could actually follow. And keep at least one person who genuinely understands each critical workflow, not just how to prompt for it.

AI will keep getting faster and more capable, and the temptation to let it carry more will only grow. None of that changes the basic physics. The rigor has to go somewhere. In the businesses that come out ahead, it goes somewhere chosen — into clear specs, real checks, and a person who owns the result. In the ones that don't, it slips into the gap between "the AI handled it" and "nobody checked," and waits there for the worst possible moment.

References

Curious what this looks like for your business?

We help small and mid-sized teams turn AI from a headline into something that actually works. Let's talk about where to start.

Get in touch