Industry executives and experts share their predictions for 2026. Read them in this 18th annual VMblog.com series exclusive.
By David DeSanto, CEO of Anaconda
If you walk into my office, virtually or in-person, the first thing you’ll see is a wall of built-ins lined with LEGO sets: Rivendell, The Razor Crest, The Titanic, and others I built since the start of the pandemic. I was a LEGO builder as a child, drawn to the idea of building new possibilities from a collection of pieces, imagining worlds and building them one brick, one plate, and one slope at a time. Looking back now it is obvious why I gravitated to engineering early in life and have spent the majority of my career with R&D.
A lot of people see LEGO as a hobby, and it very much is, and it became a hobby for me at the start of the pandemic when almost everything closed around the world. But I also see it as a philosophy for how technical architecture and complex systems should be built. Building with LEGO requires imagination and discipline to try again when something doesn’t fit quite right together. LEGO has taught me how to build with intention and building enterprise AI systems demands the same from every organization striving to compete.
2026 will be the year leaders stop treating AI like a silver bullet for everything and start treating it like the complex system it is, one that requires technical architecture, thoughtful implementation, and a purposeful goal to achieve. With that shift underway, 2026 will redefine the AI ecosystem as we currently understand it. This includes measuring AI by impact tailored to each organization’s desired outcome, rewarding companies that construct real AI-native solutions, and renewing investment in open source as the shared foundation of secure AI systems at enterprise scale.
Constructing what success looks like for your build
For the last several years, AI ROI has been flattened into the single goal of efficiency. Boards are conditioned to ask, “How quickly can we see returns?” while internal teams have been pushed to justify AI spend solely on the basis of time savings. This thinking isn’t only flawed, it’s misleading.
Fintech company Klarna’s story is a cautionary tale. They implemented a hiring freeze and carried out major workforce reductions, only to realize that their AI investments didn’t deliver the dramatic efficiency surge and high customer experience they expected. When companies measure AI by the wrong KPIs, they make the wrong decisions.
In the real world, efficiency as time savings is not the universal metric for AI success and shouldn’t be seen as “the one ring to rule them all”. AI ROI must become bespoke, as no single metric can measure its impact by itself, and include unique team and organization specific measurable outcomes to ensure everyone is accomplishing what they’re trying to achieve.
When I’m building a complex LEGO set, I don’t step back and judge the final build by how quickly it came together. I judge it by how authentic it is to what inspired it, how the build experience was, how well it is assembled, and if the mechanics work as expected. A LEGO set is best when these core principles are true. AI should be evaluated the same way.
At GitLab, we measured AI value through DevOps Research and Assessment (DORA) Metrics, with a focus on Change Lead Time, as it reflected the time maintainers spent on code changes and how many changes needed to occur based on feedback from the maintainers. This, combined with code quality goals, worked for us as it aligned with the company’s Iteration value which is core to how GitLab operates.
The leaders who move toward proving ROI with their actual goals, developer workflows, and the specific problems they’re solving, not a generic measure of “efficiency”, will be the ones who succeed.
Building beyond the packaging
The rapid rise of generative AI has unleashed a wave of innovation, with AI research labs racing to make large foundation models easier to adopt, deploy, and use in production while reducing “hallucinations and confabulations”. With this wave, companies quickly brought AI into their products, by wrapping the APIs of these new AI models, providing a thin layer of implementation between their original application and this new technology. Today, these early products are referred to as “AI wrappers” as they sit on top of a model and provide a thin workflow around it.
And let me be crystal clear: these AI wrappers have played an essential role in accelerating AI adoption. They lowered the barrier of adoption, enabled easy experimentation, and helped organizations understand the potential of AI long before they were ready to build custom systems. But 2026 will mark a critical juncture in how we build AI.
A way to visualize AI wrappers is to look at how some LEGO enthusiasts knoll their pieces before building, sorting and organizing everything meticulously. But knolling doesn’t get you closer to the final build or create anything new. AI wrappers work the same way. They take inputs and sort them into prompt structures needed for the model it uses. The companies providing true AI-native capabilities will be different. They’ll solve real-world problems, integrate into key workflows, and redefine how work gets done.
Shared bricks, stronger builds
Open source is not just a way to build and license software, it is the architectural foundation of the future of AI-native development. No single company can move as quickly, securely, or imaginatively as a global community contributing shared components, libraries, packages, and improvements that everyone can build upon.
The companies that understand this will build with higher velocity while not compromising on interoperability, security and governance, and scale. The opposite will also be true. Those that attempt to go it alone, ignoring the value of the broader AI community, will be missing out on leveraging the richest set of “bricks” available.
Every LEGO set or MOC (my own creation) build is anchored by the same universal principle: a shared system of bricks that anyone can use to create something extraordinary. Open source is the “brick library” of AI. It provides a common set of interoperable pieces that lets organizations build ambitious, enterprise-scale systems without starting from scratch or compromising on security, scale, and velocity. As AI adoption accelerates, this shared foundation will become even more essential.
The future of AI isn’t predetermined. It’s something we must assemble, one purposeful brick at a time. And if we build together through open source contributions, everything is awesome.
##
ABOUT THE AUTHOR
David DeSanto is Chief Executive Officer at Anaconda, where he leads the company’s mission to empower the world’s data science and AI communities through open-source innovation and secure enterprise solutions. A proven product and technology executive, David brings more than two decades of experience spanning cybersecurity, developer platforms, and enterprise software.
Most recently, David served as Chief Product Officer at GitLab, where he led the global product organization in delivering a comprehensive, AI-native DevSecOps platform with more than 50 million registered users worldwide. During his six years with the company, he helped transform GitLab from a high-growth startup into a publicly traded, industry-defining leader of the DevOps Platform category.
Before GitLab, David held leadership roles in cybersecurity and product management at Spirent Communications, NSS Labs, and ICSA Labs, where he led distributed teams focused on vulnerability and malware research, network security, and security product validation. He has also served on the OpenSSF Governing Board, helping shape global software supply chain security initiatives.
David holds a Master of Science in Cybersecurity from New York University and a Bachelor of Science in Computer Science from Millersville University.
