On 30 November 2022, OpenAI released ChatGPT as a public research preview. Artificial intelligence had existed for decades and large language models were not new. What changed was access.
A conversational interface allowed people without a machine-learning team to test advanced generative AI directly against everyday work: writing, summarising, explaining, translating, brainstorming and code.
That did not prove that AI would replace professional work. It did something more immediate: it made ignoring AI a business decision.
Key takeaways
- The turning point was access, not the invention of AI. A general-purpose conversational interface dramatically reduced the cost of experimenting with advanced models.
- What was visible in November 2022 was capability mixed with unreliability. ChatGPT could produce useful work, but it could also produce confident errors.
- The durable business question became operating leverage. The issue was no longer whether AI existed, but which parts of knowledge work could be accelerated without abandoning human accountability.
What happened then
OpenAI described ChatGPT as a research preview and explicitly asked users for feedback on strengths and weaknesses.
The product could maintain a dialogue, follow instructions and generate apparently coherent answers across a wide range of subjects. It also had important limitations. OpenAI itself warned that the system could produce plausible but incorrect or nonsensical answers.
Those limitations matter when reconstructing the moment honestly. In November 2022, nobody could responsibly know that later systems would become reliable agents, replace entire professional teams or eliminate particular occupations. Those were scenarios, not facts.
What was already observable was simpler: millions of knowledge workers no longer needed specialist infrastructure to discover what a large language model could do.
Interpretation: the interface changed the economics of experimentation
A technology can exist for years before it becomes operationally important.
ChatGPT compressed the distance between sophisticated model capability and an ordinary user’s task. Instead of procuring a specialised system or building an internal machine-learning project, a person could describe a problem in natural language and iterate immediately.
That mattered because experimentation is itself a cost.
When the cost of asking, rewriting, testing and trying again falls sharply, many more tasks become candidates for augmentation.
How the mechanism works
The transmission mechanism is straightforward:
accessible interface → widespread experimentation → useful workflows emerge → the cost of cognitive iteration falls → organisations reorganise work around the new capability.
The first changes did not require autonomous agents.
Drafting could begin faster. Long documents could be reduced to structured starting points. Code could be explained or generated. Research questions could be decomposed. Translation and rewriting became cheaper.
The professional still had to determine whether the output was correct.
That distinction remains fundamental. Faster production and reliable judgment are not the same thing.
The strongest countercase
The strongest objection is hindsight bias.
Because today’s systems are more capable, it is tempting to describe November 2022 as the obvious beginning of an inevitable revolution. It was not.
The first ChatGPT was unreliable, its knowledge and reasoning had obvious limits, and many early demonstrations confused fluent language with factual accuracy.
The correct historical claim is narrower: ChatGPT made advanced generative AI sufficiently accessible that businesses could no longer treat it solely as a specialist technology question.
What changed since then?
GPT-4 arrived in March 2023 and materially widened the range of tasks for which generative AI appeared professionally useful.
Adoption then spread through businesses of very different sizes. OECD research published in 2025 found that 31% of more than 5,000 surveyed SMEs across seven countries were using generative AI. Among those firms, reported benefits were common, but 83% said GenAI had not changed their overall staffing needs.
That evidence is more useful than the early prediction that AI would simply remove jobs. The observed mechanism has so far been more complicated: some tasks disappear, others expand, workflows change and the value of review, judgment and domain knowledge moves.
Scenarios, not forecasts
One scenario is commoditisation: model capability becomes widely available and the advantage moves to proprietary data, workflow design and execution.
A second is deep operating leverage: increasingly reliable agents allow very small teams to coordinate work that previously required larger organisations.
A third is a governance ceiling: confidentiality, liability, weak data or unacceptable error rates limit automation in high-stakes work.
Which path dominates will depend on observable outcomes such as measured productivity, error rates, enterprise adoption and rules governing professional responsibility.
Practical consequences
For a small international business, the lesson is not to replace judgment with generated text.
It is to identify where intelligence is expensive because people spend time searching, transforming, drafting, checking and coordinating information — and then test where AI reduces that cost without weakening control.
There is also a cross-border consequence. A business can become more digital and more operationally lightweight without becoming legally placeless. People still reside somewhere. Companies are still incorporated and managed somewhere. Contracts, banking relationships, regulatory obligations and tax nexus still attach to facts in the real world.
AI can reduce the minimum infrastructure needed to operate internationally.
It does not abolish international structure.
Sources
- OpenAI, Introducing ChatGPT, 30 November 2022: https://openai.com/index/chatgpt/
- OpenAI, GPT-4 Research, 14 March 2023: https://openai.com/index/gpt-4-research/
- OECD, Generative AI and the SME Workforce, 2025: https://www.oecd.org/en/publications/generative-ai-and-the-sme-workforce_2d08b99d-en.html
- ILO, The impact of GenAI on jobs, productivity and work organisation, 2026: https://www.ilo.org/publications/impact-genai-jobs-productivity-and-work-organization-review-empirical
Disclaimer
This Insight provides general information and analysis. It is not legal, tax, employment, investment or regulatory advice. The impact of AI depends on the specific activity, jurisdiction, data, contractual framework and level of human oversight involved.
