The AI Gold Rush: It Still Takes Teams

A conversation between Norbert Korny, Jeff Allison, and Sean Murphy on the AI Gold Rush and why teams are still required for success.

The AI Gold Rush: It Still Takes Teams

Teaser quotes

Sean: We’re in the age of gentleman science. They’ve deployed these models, but they don’t really have a use case. It’s kind of like they’ve deployed new instruments and now we’re learning how to play them.
Jeff: It’s not going to happen overnight. And how is that change going to be proliferated into your organization? How are you going to leverage this new technology? It’s teams that develop products, not just individuals.
Sean: Westinghouse would pick a hard problem to solve that was the first example of a new kind of use case. The cost reduction drives more use and then more use leads to novel structures.
Jeff: Okay, time out here. This is how this thing works. This is how we want, this is how we need to implement it going forward.
Sean: NVIDIA is not talking about laying people off. Right? NVIDIA’s probably embraced this and they’re doing some amazing stuff and they’re not saying yes. And what we hope next year is to lay off 20% of our staff.

Edited Transcript

Norbert: This a conversation with Sean Murphy and Jeff Allison, both of whom have spent decades in Silicon Valley high tech. We break down what AI is, what it’s not. And how organizations can adopt it without losing the craft of engineering. I’m Norbert Korny, let’s jump right in.

Sean: I’m really excited to be here, to be in a conversation with some other guys that actually know what they’re talking about, so that I can, by association, at least appear as if I belong in the room. I’ve worked on introducing various forms of automation, computerization, and digitalization to teams and organizations since 1980.

I think that AI provides new affordances and new capabilities. But there are a number of useful rules of thumb for innovation that will still hold. We are in the process of discovering what these new possibilities can do for us. But my focus has always been, going back to Doug Engelbart, the augmentation of human intellect, not the replacement. The trick is to augment people who are good at their jobs to help them develop and extend their expertise.

This idea of technology change, managing the introduction of new capabilities into organizations, is something that I’ve worked on for small firms and large firms for a while. In 2003, I started SKMurphy, and I now work with early-stage technology companies, helping them navigate the introduction of new products and methods to business customers.

Jeff: My name is Jeff Allison. I’m so glad that you’re pulling this together, Norbert. There are so many conversations happening around AI, and so many that need to happen.

Every day, I open my newsfeed and read that AI is going to do this, AI is going to do that, and wonderful things are going to happen. And I think they will, but the conversations need to get into more detail about the realities of AI and how it can help.

I have spent more than 30 years in the Valley. The Valley is just an innovative hotbed. It’s full of very creative people who have created fantastic technology. There are so many unsung heroes. You walk down the street and may meet someone who was involved in developing IBM’s ATM infrastructure years ago. They were able to do that because they were supported in the challenges they faced. Their organizations invested in tools and methodologies to enable what they developed individually and in teams.

I think AI is going to help push innovation to much higher levels, but it’s not because they’re using AI; it’s because of how they’re using it and how we create infrastructure to enable that. It’s not just about saving cost; that’s not the best way to look at it. It’s about how to help people do their best in a shorter period of time in a way that can be absorbed by the overall team.

Individuals come up with great ideas, but teams make products. Companies invest in teams. I have not mastered the new AI tools, but I do know how to introduce new technologies into organizations and what that entails. And it’s not as easy as, well, we’ll just bring this thing in, and people will start using it, and things will go great. There are all sorts of issues that need to be figured out. And I think being able to share some of my experiences with how difficult that has been over the years will help introduce this new technology into the development process.

“The future is already here–it’s just not evenly distributed.”

Norbert: I know you like the William Gibson quote, “The future is already here–it’s just not evenly distributed.” What do you think that quote gets right about AI in 2026?

Sean: It’s spot on. The new path forward only becomes clear in hindsight.

There’s an article going around that we’re in the age of gentleman science, where individual kind of amateurs are creating breakthroughs using chatbots and AI tools. And I think the major AI companies, they’ve deployed these models, but they don’t really have a use case. It’s kind of like they’ve deployed new instruments and now we’re learning how to play them. And so a lot of people are exploring the landscape, And I don’t think we’ve quite figured out the implications of what it means. And so I think what it gets right is that we’re back in an age of kind of individual exploration, individual tinkering. And we’ll look back and realize that something that happened six months ago or nine months ago, or maybe a year and a half ago, actually was the branch of the tree that we’re following. But right now we’re in this forest and it’s hard to tell.

It’s spot on. There’s an article going around that we’re in the age of gentleman science, where individual amateurs are creating breakthroughs using chatbots and AI tools. And I think the major AI companies have deployed these models, but they don’t really have a use case. It’s like they’ve developed new instruments and now we’re learning how to play them.

A lot of people are exploring this new landscape, but I don’t think we’ve quite figured out the implications or what it means. We’re in an age of individual exploration, individual tinkering. At some point, we’ll look back and realize that something that happened six months ago or nine months ago, or maybe a year and a half ago, actually was the branch of the tree that we’re now following. But right now we’re in this forest, and it’s hard to tell.

What’s Driving the AI Gold Rush?

Jeff: People who are doing new things or just starting out in their endeavor to create a new product or deliver a useful innovation, pick up new methods naturally. But 90% of us continue to stick with “the old way” because it’s proven, predictable, and we understand it. We need to talk about how those folks can get on board with this new product development paradigm. And I don’t think there’s enough being talked about that.

Also, the rate of change in AI tooling is still so high that new methods are evolving faster than they can be documented and debugged. So this transition from explorers and early adopters won’t happen overnight. A lot of what’s being described is experiments, pilot projects, or proof-of-concept. There is much less about how a successful pilot is proliferating as a proven methodology in an organization. Much less about how to use AI technology to improve your methods and provide better products for customers. I think that’s where we’re at.

Norbert: Is the Gold Rush really about efficiency, innovation, or a competitive arms race?

Sean: I think our sense of a Gold Rush is driven by competition and a fear of obsolescence. I think there’s been a lot of messaging that “Everything you know is wrong” or “Everything is going to change.”

Lately, I’ve been fascinated by the introduction of electricity into both industry and daily life. In some ways, it was a struggle between Edison and Westinghouse. Edison was a tinkerer who would try many things in parallel. One reason I think Westinghouse ultimately set the direction toward an alternating current architecture was that he chose hard problems and solved them for industrial use. This was formed by his early sales to the railroad of an air braking system that had to work reliably.

When he wanted to test his AC dynamo paired with an AC motor, he went to a remote mine in Colorado and ran it all winter, debugging problems as they came up. His approach was to solve a real problem with a satisfactory level of performance. That’s missing from much of today’s embrace of AI. Many announce tiny incremental breakthroughs, typically based on contrived benchmarks, not real-world outcomes. Everyone wants to be five minutes ahead, driven by a massive fear of missing out.

How do we Define and Adopt a New Paradigm?

Jeff: New tools, whatever they may be, can help us come up with very creative ways of developing products. We have been doing this for decades in the Valley; it’s great. The more you can automate and enable individuals to be more productive, the more wonderful it is.

The challenge is that teams develop products, not just individuals. Whenever you’re bringing in something new, you have to consider how it will work within the team.
It’s great to become more efficient in one area, but that may account for only 10% of the overall product development cycle. It’s an improvement, but if it slows down 90% of it, then it’s not really a win. It takes a team to develop today’s complex products that deliver value for customers.

Part of what I worry about, and what I want more discussion on: how do we leverage this technology so that the team gets more efficient, the team gets more competitive and the team is able to innovate beyond current constraints and not just accept them.

When we would develop algorithms or integrated circuits to perform critical tasks, we would then review results and say, “Okay, this is good, but we need to make it better.” So, if this technology allows us to make things better, I think it’s a great thing. But how do we do that at a team level, not just at an individual level?

I think the true Gold Rush is about the teams that really embrace this technology and are able to use it at the team level to collaborate and to develop product, they’re going to win. The guys that just take it and say, okay, this makes me more efficient, and they’re in a little box doing it, They’re not going to win here. They’re just going to get more efficient at what they do and it’s not going to affect the overall product.

Norbert: Yeah, and it seems like the adoption is driven by the individuals, unlike previously for enterprises, even better yet, the army before then. And it acts like a force multiplier. To be best, you need to be good at the beginning.

Jeff: Yes, you do. And we observed this when we were putting automation into the hardware development process for developed systems on chips. We got really good at doing specifications, but then we realized that we were really good at creating millions of gates, but now we have to get really good at testing them. So then we had to bring more people on and develop methodologies around that.

So I think this is what’s going to happen with this AI. AI is going to allow us to move very quickly in certain areas. But this will uncover blind spots that we will need to address. We’ve moved really quickly here and were really innovative there. But what does that mean to the overall development process? What else do we need to do to get this thing to actually realize everything that is promised?

Sean: Couple of quick thoughts. In the beginning, most technologies get adopted because of a cost reduction of some sort, or they come in as toys or kind of an artisanal tool.

I think a lot of AI has been sold to the CFO, by promising to cut costs by allowing you to lay people off. The most recent example is Block just laid off 40% of their people. I think that’s a smokescreen for years of bad decisions on their part. When you actually look at who’s getting laid off, it’s a lot of people with expertise. So I don’t buy that story.

The next step after cost reduction is more use, and then or use drives the creation of novel structures. Right now we see time savings at the individual level and thus more use.

Norbert: I agree.The story Jack Dorsey is running with is that they built smaller teams to make them more effective. Is this is the best thing we can do right now, because an individual wouldn’t make that big a difference in an enterprise? Small teams, maybe yes, but if you got 10,000 people, is it enough?

Individual Insight Can Enable Breakthroughs

Sean: I think an individual can have a significant insight that leads to an architectural or a novel combination breakthrough. The breakthrough always comes first in one person’s head, or maybe two people working in an intense collaboration. But then, to actually ship a product or deliver an industrial-strength or highly reliable processor, normally requires a cross-functional team of four to twelve to look at it from all the right angles.

If we go back to the Gold Rush, you can trace an evolution of individuals panning for gold—which is an artisanal model—to industrial-scale mining and extraction methods. Today, I think we’re still in the panning-for-gold phase. I don’t think we’ve even figured out how to facilitate a team of six to twelve.

Jeff: Well, this technology was initially developed for consumer or individual use, not industrial. I don’t see much of enterprise or industrial focus on what teams of engineers or scientists need.

Sean: It’s hard to track everything that’s going, but the Perplexity guys seem to be passing a test of giving answers with substantiation with links. As opposed to providing an answer that you have to treat as if it was delivered by an oracle. So Perplexity is at least a half-step towards deploying something that could be part of a reliable business process.

Jeff: That’s a big step from a consumer-based product to company-base. We’ve all done that. If anyone’s trying to bring technology into their organization and it works well for an individual, there is still a long process to get it adopted by one more more teams.

It requires a higher level of reliability, support, and well-defined infrastructure to enable all of that: standards, databases, defined development processes, revision control, methodologies, etc. These support infrastructure elements all take time to figure out and develop, regardless of whether they are for an outside vendor or an internally developed tool. There are many steps from a proof of concept to getting something widely deployed, making many people more productive.

Different parts of the organization may have different architectures and different cultures. An established business unit can have a big centralized architecture, but newer smaller business units may be more decentralized with small teams, some scattered all over the world. And then Different organizations have different architectures, right? They work differently and may benefit from different support and infrastructure models.

I think there’s a lot to be discussed here. And I’m happy, Norbert, that you’re doing this and trying to shine some light on some of these issues so that we can have these kinds of discussions and kind of work through it. Because I think some of these things are just not getting talked about enough.

Norbert: Is the software something where we’ve seen the biggest impact? Because from my point of view, it seems to me the software seems to be reinventing itself. We introduced the concept of shared libraries to reuse logic. AI recently introduced skills. Now you need a skill repository, and there are “malicious skills” you need to scan for security threats. And this is a whole software development lifecycle you need to revisit. And we’ve been through this already.

Adoption Takes Months to Years

Jeff: We could draw lessons from earlier disruptive technologies for software development. How long did they take to go from individual adoption mainstream use. It wasn’t days or weeks. It was months and years.

I don’t know how long AI will take to become a reliable mainstream development environment for even small teams. But I think to Sean’s point where you’ve got people that are making decisions on this technology, they really don’t understand the technology, but they are just looking at cost savings. I can pay for an agent that can develop a product that costs X vs. 5-10X for a human being.

That’s a metric, but to me, it’s very short-sighted. It’s no way to run the business.
I think we are at a point where the business guys are running the show, but I’m hoping that the engineering and development organizations step in here and to say, “Hey, okay, time out here. This is how this thing works. This is how we need to implement it going forward, and this is what’s going to make sense for us rather than just counting the dollars.”

Sean: The business guys remind me of the early rapid early evolution of HTML. Netscape on a whim decides to add the blink tag or to implement cookies in a certain way. I think we’re going to look back on MCP and on some of these other things constructs and conclude that they should have never left the lab.

MCP is an interesting first effort, but it’s not a reliable, scalable protocol, and it throws away 90% of what we know about security and protocol design. I’m okay with the individual experimentation and allowing individuals to scout and report on what they were able to do. But because of this consumer focus and this fixation on replacing professionals, knowledge workers, or white collar workers, they really haven’t engaged the people who define processes, the methodologists who look at this in a structured way. And it doesn’t seem like those people are on staff at the major AI companies; they seem to hire people with different expertise.

Another interesting thing for me is that Jensen Huang at NVIDIA is not talking about laying people off. NVIDIA has deeply embraced AI and is doing some amazing stuff, but they are not saying, “Next year we hope to lay off 20% of our staff.” They realize they are in a Red Queen Race, running faster and faster just to hold on to their position in the market. They understand that the bar for acceptable performance will be continually raised.

Jeff: I think NVIDIA is airing this technology out internally and they’re probably developing methods, processes, maybe developing tool chains around this to enable them to develop their product. I think those tool chains and those methods will become more available to people over time.

This reminds me of design automation in the early days of hardware. The first Design Automation Conference was held in 1964, but it wasn’t until 1982 that they formally recognized independent vendors with an associated trade show. EDA was incubated by larger organizations for almost 20 years. IBM and other big companies developed technology internally, so they understood it.

And then, those technologies became more commercially available, creating a whole market. So I think that has to happen here too. And I think NVIDIA is probably taking the lead on it. And there may be more companies out there that are doing stuff behind the curtain, right? That, you know, they’re not talking about right now, but they’re maybe saying, okay, we’re going to develop this next generation of product or this product, you know, we’re going to use an AI model on this product and we’re going to air it out. And once we figured out how it all works, then we’re going to make it more mainstream in our organization. Maybe that’s going on. That’d be great. It’d be great to talk to people that are doing that. The AI automation gurus of the future, right? Where are they and what are they doing?

Sean: It seems like there’s two tracks running here. So, I agree it’s very possible that there are organizations that are getting value and deploying this to make internal processes or design processes more efficient, and they’re getting big advantages that they’re not disclosing. Because most of the stories that are being told seem to be told from the perspective of, “Don’t worry, we’re not falling behind. Here’s some examples of how we’re using AI.” But when you dig into most of those stories, it puffery.

We are still in the early stages. The Design Automation Conference was a community of practice that allowed competitors to compare notes and learn. People would not give away everything they were working on, but there was still value in solving common problems.

I remember watching a YCombinator LightCone podcast where they essentially concluded, “We think we can solve electronic design kick the currently EDA industry in the ass in a year from now.” Not so much. The semiconductor industry and the EDA guys have been running a Red Queen Race ever since they passed Moore’s Law. They are incorporating this technology. We haven’t so far seen anyone come out who’s made current practices obsolete. I think the current vendors are obsoleting their current tools, but that’s different from some outsider using this to destabilize the industry. All the majors now have a VP of AI. Everyone on the semiconductor side, the systems houses, and the design tools are all looking at this. I think it will be like electricity: It’s going to raise the level for everyone.

Jeff: VP of AI? Good.

Sean: You asked where we’re seeing the biggest impact. I think right now it seems like it’s amongst experienced developers working on greenfield designs. That to me is kind of a splinter or a scouting activity. There’s some changeover point where you’ve got a working proof of concept and now you actually have to harden it and deploy it and support it. And that’s not going to be normally just one person. I don’t know that we’ve got the models for that yet.

Norbert: Yeah, but there’s the rush for AI to replace humans, especially the juniors. Do you see the impact on the job market in Silicon Valley?

Sean: That’s absolutely going on.

M	T	W	T	F	S	S
		1	2	3	4	5
6	7	8	9	10	11	12
13	14	15	16	17	18	19
20	21	22	23	24	25	26
27	28	29	30

The AI Gold Rush: It Still Takes Teams