The clouds can’t stop spending on AI, even if their customers are starting to question their own investments. Such disillusionment may be temporary, as Amazon CEO Andy Jassy suggests. In his words, today the vast majority of enterprise workloads remain tethered to on-premises data centers, but AI promises to change that. He says, “The ability to use AI more effectively is going to be one of the many drivers” that convince enterprises to move applications to the cloud.
This makes it clear why the cloud vendors are so focused on enabling AI workloads. They, along with Meta, have collectively spent more than $100 billion this year on capital expenditures, while signaling plans to spend even more. As Alphabet/Google CEO Sundar Pichai put it, “When [the industry is] going through transitions like this … the risk of underinvesting [in AI] is dramatically higher than overinvesting.” Okay, but although it’s great the clouds are building more and more infrastructure and services, what’s even more important is guidance on how customers can use AI effectively. There’s still way too much hype about possibilities, and not nearly enough substance.
So, what are some practical ways an enterprise can develop its AI expertise?
Making AI real
“There is still an issue of translating this technology into real, tangible economic benefit,” argues Forrester senior analyst Dario Maisto. I’ve definitely seen this in my work running developer relations at MongoDB. I don’t spend time with the executives telling Wall Street how AI will transform their businesses, as has been commonplace on corporate earnings calls. Instead, I work with the developers tasked with turning dreams into reality.
As I wrote in June 2024, most companies seemed to be succeeding with smaller-scale retrieval-augmented generation (RAG) investments. This makes sense given the relative immaturity of the industry. To do AI well, you not only need to get your data in shape, you also need experienced employees. And even if LinkedIn is telling you that your job candidate was a low-level data analyst last year but now has flowered into an experienced data scientist, the reality is different. Most people are far better at positioning themselves as AI experts than actually demonstrating the requisite background in artificial intelligence and machine learning.
As such, it’s perfectly appropriate for a company to start building up AI muscle with RAG applications or other table-stakes workloads. That’s where you’ll also begin to develop your employees. You have to start somewhere, and, with a Deloitte study finding enterprises new to AI get just 0.2% returns on their AI investments, it’s best to start now, even though the real payoff may come much later.
One other thing I’ve found to be surprisingly effective? Hackathons.
My team has run a number of community hackathons, like this one we did with AWS, designed to help developers begin building with vector embeddings and more. We’ve seen several developers successfully build projects, but we’ve seen even more success when we’ve focused the hackathon concept within an enterprise. Rather than having strangers competing with dummy data for AI fame and glory, corporate hackathons allow development teams to use real company data to experiment with real AI applications. At one corporate hackathon my team staged recently, we saw the teams go from tire-kicking on the first day to a sophisticated agentic system (which won the hackathon) on the second day. We’re now running these corporate hackathons for a range of Fortune 500 companies and expect similar results.
A competitive but nurturing atmosphere is not the only way to develop necessary AI skills within your development teams but I’ve found it to be a great way for developers to move from curiosity to actual success.