Why did AI suddenly appear in every board conversation in late 2022? Businesses had used artificial intelligence for years before then — fraud scoring, recommendation engines, spam filtering — yet AI rarely made the agenda of a typical SME board meeting. Something changed in the final weeks of 2022, and it was not a leap in capability so much as a change in who could reach the capability.

On 30 November 2022, OpenAI released ChatGPT, a conversational interface built on its GPT-3.5 model, as a free research preview (OpenAI's announcement). The launch itself was a milestone with a clear date and source. Our reading of why it landed so hard — the interpretation that follows in this article — is a business trend analysis, and we keep the two distinct throughout.

What actually changed in November 2022

Earlier generations of business AI shared a hidden tax: using them required machine-learning expertise, integration projects, or both. GPT-3 had been available to developers since 2020, but it arrived as an API — a way for engineers to build products, not a tool a finance director could pick up. ChatGPT removed the intermediaries. You typed a question in ordinary language and got a useful answer, then a follow-up, then another. No ML team. No procurement cycle. No training course.

That natural-language interface is the accessibility shift, and it deserves to be stated plainly: for the first time, general-purpose language modelling was packaged so that any knowledge worker could test it against their own job within minutes. A commercial manager could draft a supplier email. An operations lead could summarise a policy document. None of it needed to survive contact with production systems for the penny to drop — people could feel the shape of the thing immediately.

To be clear, we are not claiming AI began in 2022. Businesses had operational machine learning well before this — credit decisions, demand forecasting, predictive maintenance — usually delivered by specialist teams and invisible to most staff. What changed in late 2022 was that the conversation became general. AI stopped being a capability you commissioned and became a capability anyone in the building could audition.

Why boards noticed

Adoption was extraordinary by any historical standard for a consumer technology launch. OpenAI's then-CEO Sam Altman reported that ChatGPT passed one million users in roughly five days after launch — a figure he published on X in December 2022 and that was subsequently reported by Reuters in February 2023. Whether or not you trust growth figures from an interested party, the direction is not in doubt: colleagues were arriving at work having already used the tool at home.

That bottom-up arrival is the part many organisations mishandled. The sensible executive question in early 2023 was not "should we adopt AI?" — your people already had — but "what are they pasting into it?" Customer lists? Contract drafts? Unreleased financials? The accessibility shift cut both ways, and the data-handling question arrived in the same envelope as the productivity one.

Illustrative example: imagine a 120-person professional services firm in early 2023. A handful of consultants begin using ChatGPT to tighten client reports and to rephrase awkward emails. Output quality rises on routine documents; nobody budgets for it or sanctions it. Then someone pastes a client name into a prompt to personalise a paragraph, and the managing partner realises there is no policy, no approved tool, and no log of what has already been shared. The value was real. So was the exposure.

Constraints and trade-offs, stated honestly

Even in its launch form, OpenAI documented real limitations: ChatGPT could produce plausible-sounding but incorrect answers, was sensitive to small changes in phrasing, and could not cite a source of truth for its claims (OpenAI, 2022). Those limitations mattered less for a Saturday-afternoon curiosity than for business use, where a confident, fluent, wrong paragraph is more dangerous than an obviously broken one.

The honest trade-off for 2022, then: the same fluency that made the tool accessible also made its errors easy to miss. Organisations that treated early output as a draft requiring a human check captured value; organisations that treated fluent text as verified fact discovered the bill later. Neither group was wrong about the technology's promise — they differed in how much verification infrastructure they built around it.

What this means four years on

Looking back from late 2026, the ChatGPT moment looks less like an invention and more like a door opening. The underlying capability was years in the making; the packaging made it a mass business conversation almost overnight. For established SMEs, the durable lesson is not about any single tool. It is that AI adoption is now a governance topic as much as a technology topic — it starts with what your people can already reach, and your first real decision is what structure you put around that.

Action checklist

  • Audit which AI tools staff already use, formally or otherwise.
  • Write a one-page acceptable-use policy before you write a strategy.
  • Identify the highest-volume written tasks in the business as candidate use cases.
  • Decide what may never go into a public tool: client names, personal data, commercial terms.
  • Assign one owner for AI questions so the board has a single point of accountability.
  • Treat every AI-produced draft as a draft until a named person verifies it.
  • Record one baseline metric now, so later improvements are measurable.

This article is the first in our series From ChatGPT to operational AI: 2022–2026. If you want a structured look at where AI already touches your operations — and what to put around it — Cooper Agile runs technical reviews as the FRAME stage of our FORT delivery cycles.