AI doesn't have to be hard; start small and experiment - but you must begin says Pure Storage exec

Par Botes comes with more than 35 years of tech experience including large-scale storage systems, product development, and AI innovation. In between stints at Pure Storage he spent four years working on the AI behind self-driving cars, creating a successful product that has since been sold into car companies. Now he spends his days building new AI tech for supporting infrastructure.

It's a role that spans the entire company's products and roadmap, and also brings him into regular contact with customers. In fact, Botes was in Australia during October for the inaugral Pure Storage ANZ Accelerate conference to discuss AI, and kindly spent time with iTWire talking about the trends he is finding across the globe.

"Most customers haven't yet begun the AI journey in earnest," he said, "They want to talk about how to set it up, the pitfalls they'll face, and the processes and ecosystems required."

{loadposition david08}

It fundamentally begins with having better data and better data management, Botes explained. "This is how I think about AI," he said. "We're familiar with source code, compilers, and programs. You check your source code into a version control system, you get a compiler from a manufacturer, and you write or run programs which you test and verify and use."

In the AI world, "the source code of AI is data. The model is your compiler. And inference is your program."

And, instead of version control systems, you need to manage the lineage of your models. "It's critical, whether performing deep training or using smaller models, or fine-tuning a model, or using a RAG (retrieval-augmented generation)."

Of course, it's not easy as you and I both know. "In the world of self-driving I found the hardest problems were building intelligent systems that performed predictably," Botes said. "Outside of mega-big companies I don't think this knowledge is super widespread either."

AI is like no other tech we've seen before. You might think your experience in building for databases has an equivalent in building for AI. Things like transactional, fast I/O, good latency make sense for your database but instead "AI has everything. A small piece of data must be fast, but other pieces of data must be more like a data warehouse - all the patterns in AI are like different faces. You can't take your experience from pre-AI. It needs to be a whole new beast. It's much more diverse than in the past. Performance has a whole lot of new dimensions."

Fortunately, here is where Botes and is team comes in. "At Pure Storage we're building reference architectures. We look at ecosystems that we and our partners then bless and endorse for specific use cases. We've done this for specific industries like medicine with drug discovery."

This knowledge, this skill, is what Botes helps Pure Storage customers set up for themselves. "In 2024 and 2025 people are experimenting, but in 2026 we'll start to see real value in corporations," he predicts. "I encourage people and business to run experiments right now. Put the guardrails on, accept failures, and control the cost. But in 2026 you'll have experiences in organisations when you have new applications that will bring new capabilities."

It's an exciting future Botes sees. "When the incremental computer becomes very affordable we can do things we never dreamed of. That's the power of the cloud. It was the same thing with shopping online. The shop was incrementally added to, like with Amazon, and it's forever changed how shopping is done."

While chatbots are the very first layer for many AI experiments, they're a minor step of the whole journey. "You'll be able to transform a whole bunch of things," Botes said. "I'm a big believer in agentic AI. It will change AI and be the tech that will come and really change things."

For clarity, agentic AI is when your AI tech can not only make recommendations, but action those recommendations too. "Imagine you have two databases," Botes said. "One that does payroll, one that does sales forecasting. It's hard to make an AI to span both but you can make one that understands HR and another that makes sales forecasting. Then you make those two AIs interact and exchange data - that's agentic AI."

Let's get back to the practical elements. What's worked for Pure Storage itself, that we can all apply to our own companies? "Internally, we have a small team of people skilled in the art," Botes said. "They partner with other parts of our company - marketing, logistics, engineering, and so on."

"Our experts ask the team to come up with an idea where they think AI could matter. We figure out what they're trying to accomplish. We explore what works well."

The second step is to have some governance and direction. "Don't start out in the most sensitive part like revenue. Start in small parts," Botes said. "We have lots of contracts with boiler plate terms like who owns IP and delivers schedules, etc. We had AI analyse our contracts and ensure they conformed."

By starting small and working upwards "we've made people more productive," he said.

While we might think of AI as a solution to the greatest problems and challenges we face, the reality is AI can bring value quickly by focusing on assisting your workers in removing distractions, removing repetitive tasks, and generally freeing up time to focus on higher-valued, more important, activities. This is the message Botes gave; experiment and explore, but importantly collaborate and collectively build up knowledge within your organisation.

"One of the benefits of being a techie for a long time is I've seen two big transformations," he said. "The reason I picked the UNIX ecosystem in the 1990s was because of the client/server transformation and the end of the mainframe era. It was earch-changing. Then cloud changed how we build systems."

"And now we have this, we have AI. It's a big challenge for tech. But new engineers are coming into the world with these skills. The best AI people will be a new generation of people thinking differently about the world."

"It makes me feel young being able to hang out with smart people," he joked.