The Third Wave of Generative AI – What Changes Now
What Does the Third Wave of AI Mean for Your Business & AI strategy 2026 ?
By Caroline Kennedy
Do you get that feeling - the one where you open your phone in the morning, see another headline about AI, and think: didn't we just do this?
Me too. And I want to tell you: that feeling isn't anxiety. It's calibration. The pace genuinely has accelerated, and your instincts are correct.
In the past few months alone, we've had the so-called SaaS-pocalypse, roughly US$300 billion in market capitalisation evaporated as investors began to grasp something fundamental: the per-seat pricing model that built Salesforce, ServiceNow, and Workday into empires may no longer make sense. If a single AI agent can now do the administrative work of ten to fifteen employees, why would you pay for ten to fifteen seats? That question, once theoretical, is suddenly very real.
We've also had the release of OpenClaw, which Jensen Huang, Nvidia's CEO, called "the most popular open-source project in the history of humanity" – one that "exceeded what Linux did in 30 years in mere weeks." At GTC 2026, Huang declared that every company in the world needs an OpenClaw strategy: "This is the new computer."
Here's what all of this is pointing to: we have entered the third wave of generative AI. And unlike the first two, this one doesn't just change what AI can do. It changes what you and your team do every single day. and it's important to think about generative AI business impact and AI workflow automation.
What are the three waves of generative AI?
It helps to understand where we've come from before making sense of where we are now. Each wave has been distinct – not just in capability, but in how it changed (or didn't change) the way people actually work.
Wave One: The ChatGPT Moment (November 2022)
Cast your mind back to late 2022. OpenAI released ChatGPT – a public-facing chatbot powered by a large language model – and the internet collectively lost its mind.
There were no network effects. No viral loops. No influencer campaigns. Just a text box that could hold a conversation, write an essay, explain a concept, and occasionally make things up with alarming confidence. It was so mind-bending that it became the fastest-growing application of all time.
The first wave was about discovery. It was about realising, often with a mix of wonder and mild existential dread, that this was real. That it worked. That it was going to matter.
For most organisations, the response was exploratory. People played. Teams ran workshops. Executives asked IT about "guardrails". A lot of policies were written. Not many workflows actually changed.
Wave Two: The Reasoning Shift (September 2024)
The second wave arrived more quietly, which is perhaps why its significance was underestimated.
In September 2024, OpenAI released their first reasoning model, the kind now labelled "Thinking" in most AI interfaces. On the surface, it didn't look that different. Under the hood, everything had changed.
What is the difference between a reasoning model and a standard AI model?
A standard large language model generates its response the way autocomplete works on steroids; it predicts the next most likely word, then the next, then the next, producing an answer almost instantaneously. A reasoning model does something different. It pauses. It considers the question. It thinks through the best approach, perhaps deciding to run a web search, perhaps working through a problem step by step, before it replies.
I've seen current thinking models take up to 20 minutes to generate a long, complex response. That's not a bug. That's a feature if quality matters more to you than speed.
The improvements on complex reasoning and problem-solving were real and measurable. That said, the hallucination story turned out to be more complicated than it first appeared – OpenAI's own internal testing later found that some newer reasoning models actually hallucinate more than their predecessors on certain benchmarks. The lesson: reasoning models moved the needle meaningfully on capability, but they weren't a silver bullet. And they still felt, for most people, like a better version of the same tool. The interface hadn't changed. The workflow hadn't changed. You were still typing into a browser tab.
Wave Three: AI That Actually Works With You (Late 2025 — Now)
What is the third wave of generative AI?
It's the point at which AI stopped living in a browser tab and started working directly inside your computer, on your actual files, in your actual workflow, in real time.
This wave arrived in late 2025 with the release of the Claude Code desktop application and, critically, the powerful model behind it - Claude Opus 4.5. Shortly after, OpenAI followed with Codex for desktop, built on GPT-5.3-Codex, bringing these same capabilities to roughly 900 million ChatGPT users.
The unlock? These models had crossed a new threshold in coding capability. And that threshold, counterintuitively , also unlocked something far more broadly useful: the ability to work with documents on your computer, almost flawlessly, in real time.
Let me make this concrete, because abstract claims about AI are everywhere and they stop meaning anything after a while.
I spent a few hours recently using Cowork, Anthropic's desktop collaboration tool built on this technology, to work on what became a 35-page Word document. It helped me write key sections. It brainstormed with me. It marked up the document with tracked changes and comments. When we were done, it did a fresh pass against the eligibility criteria, critiqued the document, and made specific suggestions for improvement.
Last year, that same document took me several days, with AI assistance. This year, it took hours.
I also used it to complete a vendor onboarding process for a new client. It read the requirements, then filled out two Word documents and a PDF questionnaire with our company information, drawing on a pre-built reusable Skill we'd created for exactly this kind of task. Start to finish: minutes.
The reason earlier AI tools struggled with this kind of work is worth understanding: uploading and downloading files to a browser-based chat interface was clunky, the context was limited, and the results - particularly with complex documents like Excel files - were often poor enough to require significant human correction. The new desktop models don't have those constraints. They work where your files actually live.
And the scale of what this unlocks is significant. Gartner predicts that by the end of 2026, 40% of enterprise applications will feature task-specific AI agents - up from less than 5% in 2025.
What Does the Third Wave of AI Mean for Your Business?
Here's the hard truth: if your AI strategy was written more than twelve months ago, it doesn't account for any of this.
Does my business need a new AI strategy in 2026?
Almost certainly yes. The first wave demanded that your team understand what AI was. The second wave demanded that they learn to use it well. The third wave demands something more fundamental - a rethink of how your teams cowork with AI.
Not which tools to subscribe to. Not which departments get access. The actual shape of the work itself: which tasks still require exclusively human judgement, which can be genuinely delegated, and where the new competitive advantage lies for organisations willing to redesign their workflows rather than simply append AI to the existing ones.
The SaaS-pocalypse wasn't just a market correction. It was the moment investors caught up to what operators are already experiencing on the ground. The economics of knowledge work are shifting - and the organisations that thrive will be the ones that ask not "how do we add AI?" but "how do we restructure around it?"
Jensen Huang put it plainly at GTC 2026: "Every company in the world today needs to have an OpenClaw strategy, an agentic system strategy. This is the new computer."
The third wave isn't coming. It's here. The only question is whether your organisation is catching it or watching it from the shore.
I'm already onto it, assessing my business operations and building out agents.
Frequently Asked Questions
What caused the SaaS-pocalypse in 2026?
The SaaS-pocalypse was triggered by the growing realisation that AI agents could perform the work of multiple software seats simultaneously. With one agent capable of handling the workload of ten to fifteen employees, the per-seat pricing model, the foundation of companies like Salesforce and Workday, began to look structurally broken, and no doubt they're contemplating how they remin relevant in the AI world. Nothing new here, think about Blockbuster, but it will be interesting to see how they innovate to remain relevant. This is not just a theory, becuase the data tells the story - around US$300 billion in market capitalisation evaporated as investors repriced the sector accordingly.
How is Wave 3 AI different from Wave 1 and Wave 2?
Wave 1 (ChatGPT, 2022) was about discovery, most people realised AI was capable and worth paying attention to, but few workflows changed. Wave 2 (reasoning models, 2024) improved the quality of AI outputs significantly, particularly on complex problems, but the interface was still a browser tab and the experience was largely unchanged. Wave 3 (desktop AI, late 2025) is different because AI now works inside your computer, on your files, in real time, with near-flawless results on documents, forms, and knowledge work tasks.
What is an AI agent?
An AI agent is an AI system that can take actions autonomously, not just answer a question, but complete a multi-step task. That might mean filling out a form, editing a document, searching the web for information, and compiling a report, all without a human directing each step. Jensen Huang describes agentic AI as "the new computer."
How long does it take to see ROI from AI workflow changes?
Based on direct experience, the efficiency gains in Wave 3 can be immediate and dramatic. Tasks that previously took days can take hours. The limiting factor isn't the technology - it's the time invested in setting up reusable Skills and restructuring workflows to take advantage of what the tools can now do.
If you're looking for a speaker who can make AI feel real and relevant for your team or event, let's talk. Check out my speaking kit below.
I write about this stuff every week. If this resonated, you're welcome to join our substack list - HERE