What Utterback Taught Us About Innovation And What It Means for Generative AI

I’ve had the privilege of learning directly from James Utterback, a dear friend, a mentor, and one of the most generous and brilliant scholars I know. Jim has spent his career helping the world understand how innovation really works and how it diffuses. Not just the breakthroughs, but the patterns behind how new ideas grow, compete, and eventually shape the world. Jim was my thesis advisor at the Massachusetts Institute of Technology and his work continues to influence how I think about what we're building over at OpsZ. Nonetheless, in our most recent conversation, like many before it, Jim and I found ourselves talking about patterns of innovation and the focus was on something new, exciting and uncertain: generative AI. We talked about the tools, the hype, the potential, and the very real risks of locking in too early.

Jim, along with James Abernathy developed a model decades ago that still shapes how I think about technology shifts today. It’s called the innovation cycle, and it breaks into three phases:

Utterback's - 3 Phases of Innovation

Utterback - Abernathy 3 - Phases of Innovation

Fluid Phase – The experimental stage. Lots of different designs compete. The winners aren’t clear yet.

Transitional Phase – One or two approaches start to stand out. Design Dominance is established. Standards form. Competitors follow suit.

Specific Phase – Everyone builds around the dominant design. Innovation focuses on speed, efficiency, and cost.

This idea helps explain the past, like how cars or computers evolved, but can also be used to help make sense of what’s happening right now with generative AI. Notable models like ChatGPT, Claude, LLaMA, DeepSeek and many others are impressive, but perhaps the bigger question is: where are we in the cycle?

Some people will argue that we’re still in the fluid phase. New models appear every few months (or days). Labs are experimenting with size, architecture, and training methods. No single model or company has locked in the "winning" formula. Yet.

Others believe we’re entering the transitional phase. Most GenAI tools can do similar things write emails, summarize documents, write code. Their APIs and capabilities are starting to feel the same, at least to me. Fine-tuning and enterprise use are becoming standardized. If that’s true, the next step is coming fast where we stop talking about which model to use, and focus instead on how to build with them.

As an example at OpsZ we’re building with the expectation that GenAI models will become commodities and interchangeable tools you can plug into any workflows or operational tasks as needed. How you connect them (and there will be LOTS of them) is critical. The interfaces, the orchestration layer, and the data integration is where long-term value is created. So we feel it's prudent not to paint ourselves into technical corners by baking this or that LLM specifically into our platform, they are just another pluggable feature, like Kafka for NATS or RMQ.

There’s also the hype cycle to consider, another pattern that shows how new tech (even game changing tech) usually rises fast on excitement, drops into disappointment, and eventually finds real value. GenAI models seem to be nearing or cresting the "peak of over inflated expectations" and will undoubtedly be heading into that middle phase toward the “trough of disillusionment” where the sparkle fades and reality sets in.

But that’s not a bad thing. It’s a sign we’re moving from magic to maturity. From sizzle to structure. From headlines to habits. The enlightened ones that move to the boring old "platform of productivity" and actually get the job done.

Jim once said that innovation isn’t just about coming up with new ideas — it’s about recognizing patterns. And right now, the patterns around GenAI are looking familiar, at least to me because I watch for them. If we’ve seen them before, we can use that history to avoid repeating old mistakes.

I don’t speak for Jim but his work has taught me to stay grounded, ask better questions, and design for change. Generative AI is exciting. But the real opportunity lies in building systems that adapt when the dust settles, not in trying to pick the winner when the race is just getting started.