Host Intro (Gene Kim)

Okay. The next speaker is someone who should be familiar to most of us in this community. Dr. Mik Kersten wrote the awesome book Project to Product four and a half years ago, which is being used by so many organizations to change how they think about their software efforts. He is now currently CTO of Planview.

Over the years, he's taught me so much about architecture, which dictates so much of how organizations are wired, which is a topic of the book that Steve, Dr. Steve Spear, and I are working on.

Mik and I were talking earlier this year, and when he told me about a conversation that he had with a mutual friend of ours, I immediately thought that this should be a talk that he should give to this group. So I asked him if he could talk about his own reflections as a skeptic about most things, and ponder how things like AI will change the software profession as we know it.

Here's what's going to happen next. Mik will give his presentation, and I've asked Dr. Steven McGill, VP of Innovation at Stype, one of my favorite research collaborators, to provide a rebuttal to a certain statement that was made. Then he, along with three other people, will give a five-minute mini lecture where we've asked people to either show us something cool they've done with AI or teach us something about AI.

So let's start off with Mik.

Dr. Mik Kersten

Hello, everyone. For my entire career, I've been looking for where the 10x productivity improvement in how software is built is. That journey took me through working on new programming languages, developer collaboration tools, and more recently to value stream management and flow.

To my shock and amazement, I think much more than 10x productivity gains are actually going to come from where we're headed with generative AI and large language models. We're seeing this age start now.

This talk will not be about the social implications of this. It'll be how to understand what's happening to help you and your organization thrive through, I think, the biggest disruption that any of us are going to see in our entire careers. In terms of setting up your organization for success for this, I'm going to outline a few principles around it and how I think you should start adopting some of these technologies and experimenting with them today.

Just for fun, I did actually use DALL-E from OpenAI and Stability AI to generate all the art in this talk in the style of Dutch masters, since we're here in Amsterdam. This is a portrait done in the style of Rembrandt. This is Rod Johnson, a close friend of mine who's now coding back in the 1600s on his laptop.

Rod hasn't been coding quite that long, but he is known as an open source luminary by virtue of having created the Spring Framework and a lot of other open source technologies. He still codes on a weekly basis, sometimes a daily basis even. He has been my go-to person for understanding some of these major shifts that we have in programming.

Back in 2009, Rod and I were sitting at a cafe in Amsterdam. He was preparing his keynote for the SpringOne conference. We were reflecting back then, remember 2008, 2009 was the financial crisis, on how some banks and financial institutions were adopting open source very rapidly at that time, even though, of course, they had major budget cuts. Some of them adopted this new way of consuming and evolving software through open source. Those are the ones I think have actually thrived and succeeded more over the past decade.

What I think will actually happen in terms of large language model adoption, GenAI, will be much more significant than that. When Rod and I were digging into this, we had a conversation at the start of the year. He actually said he thinks it's the end of what he's been doing, which is having coding as both a hobby and as a profession. It'll happen in at most five to 10 years.

I think one of the main constraints, certainly in every company and every product I've worked on, has been delivering the code, building the functionality, delivering that digital experience. All of a sudden I think that constraint will be gone. Large language models, I believe, will actually remove that constraint on software delivery, which is a very, very fundamental change.

I think it's actually going to be an order of magnitude bigger than what we've seen with open source, with Agile, with DevOps. I do think, by the way, that it's still very important to listen to all the talks at this conference, because if you've not put in place those DevOps practices, you are not structured properly to leverage the change that's coming with generative AI.

The thing that I'm most concerned with is that some organizations will be properly structured and set up for this, and will take this move seriously and move very quickly, while others won't, because we're going to get into this very big frenzy. The goal here is to make sure that you bring this back to your organization, your leadership, and make sure that you lean into this change, because we don't want your organization to get left behind.

I actually see this as one of these major technological revolutions. For those of you who have read Project to Product, it talks about these five big technological revolutions, from the Industrial Revolution to the age of steam and railways. This is documented by the work of Carlota Perez, who wrote the book Technological Revolutions and Financial Capital.

Interestingly, in early 2019, I had a call with Carlota, and I asked her, I was reading a bit more about AI, I still had no idea that we would be where we are today with large language models, but I asked her: what do you think the sixth technological revolution will be? I said it seems like it's going to be AI, but that probably won't happen for another decade or two, which, of course, was completely wrong in terms of the power that we're seeing now.

She said one thing we know about these revolutions is, before the installation period, you don't know what it's going to be yet. There are too many things being tried and experimented with. But as soon as something fundamentally changes around the constraint on production, and that's what's always happened: electricity changed the constraint on a factory, where Henry Ford was all of a sudden able to not be constrained by a steam engine in the middle of a factory floor, but get power to the whole assembly line and scale production and bring about the age of oil and mass production.

As soon as that constraint is gone, the same thing will happen. Now that main constraint of building software, of coding, is gone. I think we're now in the age of AI and in this period of massive frenzy.

The key thing is now that software has become much more scalable and much cheaper to produce, how do we approach this? How do we make sure that the organizations that get built up in the frenzy include your organization as well? What happened back then in the early 1900s is there were over 300 car startups in Detroit, and only some of them survived that frenzy and went past the turning point.

Of course, for me, like for everyone else, this is early days in terms of understanding this. But I've really looked at it in terms of three different productivity multipliers that we're going to get from generative AI and large language models.

There are the code agents that are going to multiply how quickly we can deliver code, how quickly we can refactor it, how quickly we can modernize things: really down at the code level.

There are the value stream agents. Think of agents as connecting GPT and large language models to various tools that can produce outcomes. Value stream agents will fundamentally change how we manage at the team-of-teams level.

Then strategic agents: you've heard Gene and Steve Spear talk about rewiring organizations. We're going to be able to, of course, have agents that take what good looks like, such as in their upcoming book, and actually implement that, because apparently without AI we've been very slow to do it for most organizations.

The key thing is that what we'll be doing is leveraging the machines, leveraging the AI. At each one of these levels, it's not that we'll have a tenth of the number of programmers. As Sam Altman himself said, he thinks we're going to have 10 times the software with the same number of programmers, because the appetite for productivity and for delivering great experiences through software and through digital is just massive.

The interesting thing is we'll have this at every level. The question is: how do we not just get that leverage that we can from people being amplified through AI, through machines, at the code level, but at the value stream level, at the leadership level, all the way to the most senior levels in the organization?

I'll quickly dive into my current work in progress on each of these layers.

In terms of the first level, the code agents, I think code generation is here now. If you're not leveraging things like GitHub Copilot in your organization today, I urge you to do that and to make sure your leadership removes the impediments on that, because this is how programming will happen.

We have a new breed of co-ops that just came into Planview. In the selection process, they asked us: are we going to be able to use GitHub Copilot? If not, then we don't want to come here. That's how profound this thing is.

Code brushes can already make your code much better. Code brushes will make it possible to do things that are very painful, such as colleagues who are still behind on Angular-to-React migrations. Those kinds of things will become automatic. All of this will be automatable through these large language model prompts.

As we heard Patrick Debois mention today, prompts will become the code. A lot of what we're executing, for example, my data science team over the summer implemented sentiment analysis using the AWS sentiment service. Now that's just a little bit of prompt engineering, not a summer's worth of coding activity and curating the data.

These prompts will become a core part of the way that we code, and the large language models will just become a part of our programming environment. The way to dig into this and understand this better is to dig into what Auto-GPT does, when it's not just a single prompt, but you've actually created a loop and a feedback loop with a large language model that's going towards some goal.

That goal might be refactoring your whole application. Maybe one of the harder things you could imagine is doing your SAP system modernization. I think that would probably be the limit, but it is theoretically achievable. This will fundamentally change how we write code and the effectiveness with which we can do this.

Above the code level is what's really interesting to me, and this is where most of my time is going right now, where we have the architecture, the organizational design, and the value streams. Think of the team-of-teams level.

Flow optimization is the thing that I care deeply about through value stream management and the Flow Framework. What we've seen is that we're surfacing these organizations and showing organizations: you've got too much work in progress; you've got too many dependencies between value streams; you've got all this waste in the system, all these queues; but not enough is happening about that.

What we actually will do is now be able to automate flow optimizations, to automatically reject work when the value stream is overloaded, to accelerate flow, to remove burden from the system, to make sure that we've got investments happening in architecture and tech debt that perhaps the business is not prioritizing by making it clear to the business what will happen if you don't do that, but provide that feedback in a very natural language way.

Roadmaps can be automatically managed this way. A lot of companies already know what good looks like in terms of striking that balance between tech debt and feature investment, risk reduction, and so on. Now all of this can be automated by bringing those best practices from tech giants and from startups to every organization through automatically managing roadmaps through these kinds of copilots that we will also have for the code level.

This next one just blows my mind every time I try to think about it. With Auto-GPT, you can actually iterate and bring the copilot into the loop and have it managed to some goal. That goal can be making a piece of code beautiful or an architecture better. We'll be able to do that at the value stream level itself.

Set an objective and key result of saying we want to increase retention by driving adoption of this particular product. Let's automatically generate the features that will do that to that goal iteratively until we see that, and create that fully closed feedback loop in terms of delivering value to a customer, while having people directing that rather than implementing it at a more detailed level.

That is actually when value streams themselves will become generative. The productivity increases from this are truly mind-blowing. Why stop there, because we can bring this right up to the organization level, where the agents are actually working on the organizational design and structure itself.

Today we do reorgs basically through PowerPoint. All of this data is available. All of the flow metrics are visible to an agent, and they can actually help refactor the organization to, let's say, target a goal of increasing happiness. Whenever we do any restructuring, any team changes, we always look at the employee net promoter scores for every value stream and target to that goal. From having done that for years now, I know how much more effective a large language model would be, with, of course, the guidance of the humans behind it.

Technology optimization: the whole notion of tech debt is very different when you can actually modernize the applications, your whole technology stack, allowing your organizational design and your software architecture automatically.

Another fascinating one is investment allocation. Too often the way that budgeting is done is without any real view of capacity or flow. All of this becomes connected. Budget allocation, capital allocation in the organization, is also supported by copilots that have much more depth and much more best practices and much more visibility to all of the data than any single human can have.

The challenge is that most organizations are just not ready to scale this. This is real data that we see of value streams using our tools. What we're seeing across our customer base is that 8% of the time is spent in development. If you now 10x your development capacity, you're not speeding up at all. You've got these large upstream and downstream bottlenecks in terms of approvals, processes, waterfall compliance and such.

The challenge is most organizations are not set up to do this. One of my main calls to action is we need to move faster on this front or more organizations will get left behind.

We've got more detailed data now. With the fifth anniversary of Project to Product, we ran a study combining both data from 3,600 value streams across 34 organizations. It's all live system data from Jira, ServiceNow, GitLab, GitHub, all those sorts of tools, as well as survey data, and we put those together.

What we noticed is that there's a 10x mismatch between the demand put on technology teams and the capacity of those teams, because whenever everybody creates their strategic plans for the year, they don't take into account the capacity.

Forty percent of team efforts are wasted due to overload and bottlenecks, largely caused by that mismatch: too much work in progress and too many dependencies. This last one is shocking: even though we're over a decade into DevOps best practices, 30 to 40% of that flow time is actually spent after code is written and deployed. It's usually just getting deployed into staging or development and not actually getting into production.

Eighty percent of value streams don't proactively invest in tech debt. This is a place where we can leverage the computers, leverage the machines, to make that non-negotiable, to reduce that tech debt and to always invest in it, and to remove the burden from the system.

Ninety percent of value streams measure outputs instead of outcomes, which is a challenge because we need to bring those outcomes to the teams to know what they're contributing to, who the customer is, and start with the customer in mind. We can't leverage the benefits of value stream-level agents unless we actually have that outcome there, have that outcome connected.

The challenge right now is that if your organization has these three layers disconnected, the strategic layer, the value stream layer, and the code team layer, you're not set up properly to leverage everything that's coming, everything that's there already. At the bottom, the code copilots are there already, and what's coming is these value stream-level agents and these strategic agents.

The main thing is make sure you connect these layers. Make sure you've got OKRs flowing down, you've got visibility coming up across your entire tool set to leverage that 10x productivity gain that's there at the code level. At the value stream level, I think there's another 10x to be gained. At the strategic level, where the organization is being rewired to drive better happiness, better outcomes, better delivery, another 10x. Those things will multiply in some way. That's very mind-blowing to me.

The key thing is set yourself up for this, because all this is coming soon.

In terms of guidance, in the end, this is about people. This is a Vincent van Gogh representation of Maya Leibman's talk this morning, where you've got a pizza party. Make sure that you do bring the people together to understand this, plan it out, create your AI committee, and really focus it on each of those levels: the code level, the value stream level, and the strategic level. Make sure that people are learning and adapting to this new way.

Just like no developer is going to be successful five to 10 years from now without leveraging AI, technology leaders won't be there. I think this really will elevate what we can do, the value we can deliver to our customers and to our organizations.

To set yourself up for this, you have to connect all of the work from strategy to delivery. That data has to be available, it has to be connected, and it has to be able to flow to a platform that can leverage it.

You also need to complete the shift from project to product. If you don't have product value streams, if you're still working in that old way of throwing work over the fence to IT, it won't work. There won't be anything to optimize. That's the challenge.

As soon as you get back home, make sure that your organization is adopting what's there already today. The code agents like GitHub Copilot are there to be used today, and it'll help you learn across the organization, across your teams, how these things work and how much you can leverage them.

As you're doing this, make sure to measure flow, to measure the productivity, so we have a sense of the baseline that's there and how much productivity is there to be gained. In the end, we need to make the economic case to organizations, to our leadership, and our boards on how much this can help us accelerate.

Keep in mind that others are doing this already. In that survey, we noticed that a third of organizations that were there have actually made the shift and are in various levels of scaling from project to product. It's happening. The value streams are there to be optimized.

It's the two-thirds that I'm most concerned with because, again, the guidance and the signals have been there. It's there in terms of adopting best practice on agility and on DevOps. Everything that's to come with generative AI and LLMs is going to build on top of that.

To learn more, you can check out flowframework.org in terms of how you can measure and create and set up those value streams. We'll be creating posts and announcements on the Planview blog as well in terms of those value streams and strategic levels and how you'll be able to leverage that within your own organization.

I think it's very exciting times. I know some of it can seem a little bit scary because, again, I've never been through this much change. I have never watched this many YouTube videos in a two-month period before. I was actually trying to avoid watching them as much. But I think it's more exciting than scary, especially if we can make sure that your organizations and the broader economy are leveraging all of these developments to build truly great things.

With that, thank you very much.