Welcome to the 3rd wave of gen AI

Do you get the feeling that the pace of generative AI news and disruption is getting even faster? Me too.

In the past few months we have had the SaaS-pocalypse (roughly US$300 billion in market capitalisation evaporated), the release of OpenClaw (which is the fastest-growing open source software project ever), and Jensen Huang declaring AI agents ‘the new computer’. And the reason why?

We have entered the 3rd wave of generative AI.

If we take a look back, the first wave was sparked by the launch of ChatGPT in November 2022, the first public-facing chatbot powered by an LLM. Despite having no network effects (as people in Silicon Valley would say), it was so mind-bending it became the fastest growing app of all time.

The second wave was quieter. In September 2024 OpenAI released their first Reasoning model (often labelled ‘Thinking’ in AI chatbots), and suddenly ChatGPT became markedly smarter, if you were prepared to wait for an answer.

Unlike a traditional LLM, which outputs the answer immediately, generating the next word as a prediction of what is most likely to come next, a Reasoning model takes a beat, thinks about what you’re asking for, reflects on the best way to approach the task (maybe a Web search for example), and then will reply. I have seen current Thinking models take up to 20 minutes to generate a long response. If you’re not in a hurry, and quality is more important than speed, the release of Reasoning models by all of the AI labs moved the needle - in particular in reducing hallucinations.

Which brings us to this third wave, which we entered in late 2025 with the launch of the Claude Code desktop app, and perhaps more importantly the new very powerful model behind it. The combination of the app and the new model (Claude Opus 4.5) suddenly unlocked using AI chat for working with documents on your computer, and getting perfect results nearly every time.

Anthropic’s releases were closely followed by OpenAI Codex for desktop (using the GPT-5.3-Codex model), providing these same capabilities to the ~900m users of ChatGPT.

Although working with documents has been possible with AI chatbots for a while, it was largely a poor experience due to having to upload/download files to the browser, and the results were often not great (in particular with Excel files).

These latest models have crossed a new threshold for coding, which also unlocks their ability to work with documents nearly flawlessly. I spent a few hours today using Cowork to collaborate with me on what ended up being a 49 page Word document. It helped write key sections, it brainstormed with me, it marked up with track changes and comments.

When we were done, it did a fresh pass against the eligibility criteria and critiqued the document and made suggestions. It was truly amazing. Last year it took me several days (with AI help!) to work on this document, this year it took me just hours.

I also used it to complete a vendor onboarding process for a new client - it filled out two Word Docs and a PDF questionnaire with our company information (using a prebuilt reusable Skill we have created).

The boost to operational efficiency is undeniable. The impact on coding is also monumental, but a subject for another post.

Thanks to this third wave, businesses now need to rethink how their teams work cowork with AI. Your AI Strategy from a year ago does not take into account these new capabilities - which in my opinion are even more compelling and impactful than previous waves.

If you are a CEO and would like a no-obligation private conversation on how this third wave impacts your company, please reach out.