The Year Everything Changed

The first nine months of 2022 were steady. Good client work. Mental Health Foundation continuing to grow. New engagement with a logistics company that pushed our integration capabilities. The team was stable, the work was solid, the industry was doing what the industry does: incremental progress, incremental complexity, incremental improvement.

Then August happened.

Stable Diffusion landed. Open source image generation that actually worked. I wrote about it last month. At the time, I thought it was the most significant technology release of the year. I was right about that for approximately three months.

Then November happened.

I've been building software for fifteen years. I've watched every "next big thing" come and go. Mobile. Cloud. Blockchain. AR/VR. Each one had genuine substance underneath the hype, and each one took years longer than the breathless predictions suggested. I'd developed a healthy immunity to technology excitement.

ChatGPT broke through that immunity in about ninety seconds.

I opened it on a Thursday evening. By midnight I was still there, testing it, pushing it, trying to find the edges. I called the team the next morning. "You need to see this."

The thing that got me wasn't any single capability. It was the breadth. It could write code. It could explain concepts. It could hold a conversation with context. It could adapt its tone. It could reason about problems. Not perfectly - it made mistakes, hallucinated facts, and sometimes confidently said things that were completely wrong. But the baseline capability was so far beyond anything I'd seen that the mistakes felt like details, not disqualifiers.

In the span of five months, two things happened that I believe are genuinely significant:

Image generation became accessible. Not possible. It was possible before. Accessible. Anyone with a graphics card can now generate images from text descriptions. The quality is imperfect but improving rapidly. The implications for creative industries, marketing, content production, and design are real.

Language capability became accessible. Same pattern. Large language models existed before ChatGPT. GPT-3 has been available via API since 2020. But ChatGPT put a conversational interface on it that anyone could use, for free, without an API key. One hundred million people signed up in two months. That's not hype. That's demand meeting capability.

The common thread is accessibility. These technologies moved from research labs to consumer products in months. The speed is what's different from previous technology waves.

I need to be careful here because the temptation is to declare everything different. It's not.

Enterprise adoption takes time. ChatGPT is a consumer product. Enterprise integration requires security, governance, reliability, and customisation. Those requirements haven't gone away. Every enterprise technology wave - cloud, mobile, SaaS - took 3-5 years from consumer breakthrough to mainstream enterprise adoption. I don't see why AI would be faster.

Data quality still matters. AI models are powerful but they operate on data. If your data is messy, siloed, and inconsistent, AI won't fix that. It might make the mess more visible.

The fundamentals of good software are the same. Clear requirements, clean architecture, thorough testing, thoughtful design. AI might accelerate some of these. It doesn't replace any of them.

Most enterprise problems are people problems, not technology problems. No AI tool has changed the fact that organisations struggle with alignment, communication, change management, and prioritisation.

We don't have a strategy yet. Claiming we do would be dishonest. What we have is a direction:

Learn deeply. Every person on the team is spending time with these tools. Not casually. Seriously. Understanding what they can do, where they break, and what they need. We can't advise clients on technology we don't understand ourselves.

Experiment internally. Before we take anything to a client, we're using it on our own work. Can AI improve our code review process? Can image generation accelerate our design exploration? Can language models help with documentation? We're finding out.

Stay honest. Clients are going to start asking about this. Some already have. Our answer right now is: "This is significant and we're studying it. We don't think it's enterprise-ready yet, but we think it will be sooner than most people expect. Here's what we're learning." That's an honest answer. It's not as satisfying as "we have a complete AI strategy ready to sell you," but it's true.

Watch the rate of improvement. The thing that genuinely concerns me - in the productive sense of concern - is the speed. Stable Diffusion in August was good. By November, the community had made it significantly better. ChatGPT today is impressive. If it improves at the same rate over the next twelve months, the enterprise readiness question moves forward dramatically.

I don't know what 2023 looks like. For the first time in a long time, I genuinely don't know. The carefully planned consolidation year turned out to have a plot twist in the final act.

What I do know is that we're paying attention. We're learning. And when we're confident enough in the technology and our understanding of it, we'll start applying it to client work in ways that are responsible, measurable, and genuinely useful.

Something shifted in 2022. We felt it. Everyone felt it. The question isn't whether this matters. It's how much, how fast, and what we do about it.

That's the work for next year.