Enterprise AI: Some Interesting Observations

For the last year, I’ve sat in countless senior management meetings and executive strategy sessions looking at the exact same slide decks (the only difference being the vendor logos). We’ve all listened to pitches about the importance of building proprietary, custom AI infra from scratch. I have been pushing back for two reasons: 1) the velocity of AI development is so fast that today’s build might be tomorrow’s relics. 2) before industry’s open standards are set, what everyone’s offering is technically a “lock-in”.

Well, it’s mid-2025, the reality finally sets in. A massive chunk of enterprise AI spending isn't going toward building massive infrastructure or training custom foundational models. More than half of it is going straight to the application layer. Companies aren't building; they're buying. And frankly, my recommendation is validated.

The Myth of the Custom Model

Early on, there was a massive fear of missing out at the executive level (more so from my non-tech colleagues). There was a widespread belief that using a commercial API or an off-the-shelf SaaS tool meant giving up our intellectual property or "secret sauce." We watched organizations pour millions into hiring scarce AI talent to train custom models, only to realize that by the time their project was ready, a general-purpose model provider had released an update that made their bespoke problem not so special.

What I am urging our leadership teams to embrace is a heavy dose of pragmatism. Our competitive advantage isn't the underlying model itself—it’s our proprietary data, our unique workflows, and how effectively we integrate these tools into our core operations.

Moving at Twice the Speed of Traditional Tech

If you’ve spent your career guiding enterprise strategy, you know the inherent risk of a "pilot program." Traditional enterprise software pilots are notorious for dragging on for nine months, getting bogged down in committee, and ending with a whimper. Historically, only about a quarter of those pilots ever make it to a full corporate rollout.

AI is completely flipping that script. Pilot programs for AI applications are converting to full, company-wide production deals at nearly twice that rate—closing in on 50%.

The True Bottleneck: Engineering and Integration

Another important element that I have been arguing is that our focus must completely shift toward the "engineering" part of AI rather than the "science" part. We need to leave the complex scientific research, model training, and basic breakthroughs to the frontier AI companies who have the billions to burn on it. Our job as enterprise leaders is application, architecture, and deployment.

In this new era, if a Proof of Concept (POC) is still sitting stuck in a lab, it’s almost never because the AI itself failed. It happens for two distinct operational reasons. Either your tech team is completely siloed from day-to-day operations—creating massive, outside-in friction when it comes to change management—or there was no basic digitalization done in the first place. You cannot successfully airdrop sophisticated AI into a raw, analog landscape and expect it to stick. Success requires a solid digital foundation and tight cross-functional integration.

My Recommendation: Pragmatism (don’t buy into hype) & Agility

If our discussions are still stalled because we think we need a massive, multi-year infrastructure project to leverage AI, we are missing the window of opportunity. The market has shifted decisively from experimentation to execution.

The winners in this landscape will not be the companies trying to build the next great AI engine. The winners will be the organizations that find the best pre-built enterprise tools, secure them with the right governance, and integrate them deeply into their daily workflows.

In 2025, pragmatism beats perfection every single time. My recommendation to top management is clear: stop overcomplicating the technology, avoid the trap of building what already exists, and focus on deploying tools that drive measurable value for our business today.