Models Commoditize. Data Gravity Compounds.

Engineers love staring at leaderboards. I see it constantly in our Slack channels across APAC—teams wasting weeks debating the nuanced differences in the latest April 2026 LLM Providers rankings.

It is a distraction. A comfortable one, but still a distraction.

When you are building software that actually moves the needle, the foundational model is rapidly becoming the least interesting part of the stack. Models are commoditizing. The real battle—the one that actually dictates enterprise value—is where the data lives. When I was scaling engineering orgs at ShopBack and GoPomelo, the hardest truth to teach young developers was that infrastructure always beats algorithms.

Look at the enterprise giants. They do not sell models; they sell gravity. Oracle tying Fusion Agentic Apps and their AI Database directly to their cloud story is exactly how this plays out in the real world. Why? Because an agent is utterly useless if it has to pull context across three different cloud environments with a 2,000-millisecond latency. The winner in enterprise AI isn't the one with the smartest model—it is the one with the lowest friction to proprietary data.

Boring Problems Produce Compounding Returns

Founders in Singapore pitch me their "AI for X" every week. Most are thin wrappers over an API, searching desperately for a problem. They optimize for the demo. I optimize for the deployment.

If you want to understand what actual, hard-won value looks like, look at the plumbing. Consider the fact that Persistent just launched a merchant fraud tool on Databricks.

This is not sexy. It will not win a design award. But it is a pure math problem. Fraud detection is about hunting anomalies in massive, unstructured data lakes. You plug the AI directly into the pipeline where the data already sits—in this case, Databricks—define the metrics, and let it run. If you can reduce merchant fraud by even a fraction of a percent at scale, that compounds. It drops straight to the bottom line. Value in AI is created by doing one highly specific, measurable task an order of magnitude better than a human.

Time is the ultimate constraint. As an operator, you have exactly three domains to allocate your finite time to: your career, your family, your finance. Wasting six months of runway building a conversational agent that occasionally hallucinates internal HR policies is a catastrophic allocation of career capital. Pointing compute at fraud, churn, or supply chain bottlenecks? That pays dividends.

The Macro Reality Check

We operate in a tech bubble where AI is treated purely as a software feature—a lever to increase developer velocity or lower customer acquisition costs. The capital markets view it entirely differently. They view it as a systemic risk.

Jamie Dimon’s yearly letter dropped recently. Amidst the standard macroeconomic analysis, he explicitly highlighted the dangers in geopolitics, artificial intelligence, and private markets.

Stop and think about that grouping. The CEO of JPMorgan Chase is bracketing artificial intelligence with global conflict and opaque private market liquidity. Are we paying attention?

At the highest levels of finance, they understand that automated agents operating at scale introduce unprecedented volatility. It is not about a rogue superintelligence. It is about thousands of agentic applications reacting to the same geopolitical news trigger simultaneously, moving capital faster than human oversight can intervene. We are building systems with feedback loops we do not fully understand yet. When you deploy an AI agent that can execute transactions, you are no longer just writing code—you are participating in a global financial surface area.

The Operator's Mandate

The hype machine wants you to believe that every application needs to be rebuilt from scratch with a natural language interface. The reality on the ground is much colder.

If you are leading a technical team today, your job is not to chase the latest model release. Your job is to ruthlessly evaluate where intelligence can remove friction from your core business logic, and ignore the rest.

Deployments beat demos. Margins beat mindshare. Build accordingly.