The AI Adoption Gap: Why Context, Not Intelligence, is the Bottleneck

The current conversation about AI is stuck in a false choice: we are either racing toward a superintelligence or we must stop development entirely. This framing misses the point for most businesses. As Databricks CEO Ali Ghodsi points out, today's models are already smart enough to automate a large amount of work. The real problem is not a lack of model power, but a lack of institutional context. Companies are struggling to connect powerful models to the messy, internal reality of how they actually make decisions. For leaders, the advantage lies not in chasing the next big model, but in building the infrastructure needed to give existing models the internal knowledge they currently lack.

The Hidden Cost of Pacing the Frontier

The debate over slowing down frontier AI is often framed as a safety issue, but Ghodsi notes that pacing has nothing to do with security. When labs talk about development speed as a safety concern, they are trying to please two groups--those who fear AI and politicians who want oversight--without actually satisfying either.

The real danger, according to Ghodsi, is not some future superintelligence, but the immediate reality of cyber-resilience. As agents become more capable, the time between a vulnerability being discovered and its use in an attack has dropped from years to hours.

"The time it would take from a CVE vulnerability being sort of published until you see it actually weaponized in the industry would be like two, three years... now it's down to basically no time like things get immediately weaponized."

-- Ali Ghodsi

This shift forces the worlds of data and cybersecurity to merge. Because agents leave logs and digital trails as they work, the amount of data needing analysis has skyrocketed. Companies that rely on traditional security teams to manually check alerts cannot keep up with the speed of automated attacks. The advantage goes to those who treat security as an engineering problem to be solved with automated detection, rather than a policy problem to be managed through pacing.

The Ontology Moat: Why Context is the Real Product

Most companies are using AI as a faster version of Google Search. They fail to reach real automation because their models lack the institutional memory that separates a veteran employee from a new hire. Ghodsi calls this an ontology--a digital map that links concepts, departments, projects, and the unspoken knowledge of how work gets done.

"There's just a lot of ingrained knowledge that's sitting in everybody's heads, who knows how an organization works... how do you actually distill it down into a graph? Actually a digital graph that you can then feed to the AI."

-- Ali Ghodsi

The core insight is that organizations are usually structured as trees where information flows vertically and decisions get stuck at the top. By building an ontology, companies can decentralize decision-making. When an AI understands the context of the organization, it can analyze data, answer questions, and provide the reasoning that used to require a meeting. This creates a lasting advantage: while competitors wait for smarter models, the company that has digitized its internal context will operate at a different speed.

The 18-Month Payoff: Moving from Token-Maxing to Value-Maxing

The initial rush to use AI led to token-maxing, a period of wasteful and unconstrained experimentation. Moving toward maturity requires a shift to value-maxing, which means managing costs and using a mix of models.

Ghodsi notes that the most advanced organizations are moving away from the idea of one model for everything. They are using smaller, cheaper models for routine tasks and saving frontier models for complex, high-value problems. This requires the hard work of building evaluation frameworks, a task many teams avoid because it feels less exciting than testing the latest model. However, this effort is exactly where the competitive advantage is built. Teams that invest in testing their AI agents will survive the inevitable budget cuts that force less disciplined competitors to retreat.

Key Action Items

• Audit Your Detection Infrastructure: Move from human-led security to automated, agentic threat hunting. This is a priority given how quickly vulnerabilities are now weaponized.
• Start Building Your Ontology: Over the next few months, begin digitizing internal meetings and project relationships. You cannot feed context to an AI that has not been captured.
• Implement a Model Gateway: Use a gateway to manage token usage and budgets. This lets you route tasks to the most cost-effective model rather than letting engineers use expensive frontier models for simple tasks.
• Invest in Evaluation Frameworks: Treat AI development like software engineering. Build robust tests now; this will pay off in 12 to 18 months as your system grows and you need to ensure reliability.
• Adopt Agent-First Infrastructure: When choosing tools, prioritize those designed for agentic workflows, such as high-speed cloning and low-latency responses. This will be the standard for the next generation of enterprise software.