Why the Top 5% of Business AI Users Are Getting 10x the Value

We got a pair of major enterprise AI reports this week—one from OpenAI, one from Menlo Ventures—and taken together, they’re telling a very clear story.

Menlo calls generative AI "the fastest scaling software category in history," now at $37 billion in spend and capturing 6% of the global SaaS market just three years after ChatGPT dropped. Here's what I think actually matters in these reports.

AI coding ate. Enterprise spent $4 billion on AI coding this year—55% of all departmental AI spend. Marketing AI? Just 9%. Code agents specifically were up 36.7x. In many ways, 2025 was the year of AI agents like people thought it would be. It was just coding agents, not general agents.

Anthropic flippened OpenAI on enterprise. Menlo puts Anthropic at 40% of enterprise LLM spend, up from 12% in 2023. OpenAI went the other direction—from 50% down to 27%. This is exactly why OpenAI is as aggressively focused on Codex as they are.

The leaders are pulling away from the laggers. OpenAI found that "frontier workers" (95th percentile of adoption) generate 6x as many messages as the median worker, send 17x as many coding-related messages, and report dramatically higher time savings. Same pattern at the firm level. The gap isn't closing—it's compounding.

Agents are still nascent—but there's a clear maturation path. Only 16% of enterprise deployments qualify as "true agent systems," and copilots represent 10x as much spend as agents do. But OpenAI's data shows something important: the leaders aren't just using AI more, they're using it differently—19x growth in custom GPTs and projects, deeper integration into core workflows, more automation. The infrastructure for agents is being built. The question is how long until enterprises have the data pipelines, governance structures, and operational models in place to let agents actually run.

If you're looking for evidence that this is a bubble, you're not going to find it in enterprise adoption.

Tomorrow’s Headlines

• Report claims DeepSeek using banned Blackwell chips to train

• McDonald’s pull AI-made Christmas ad

• 1/3 of teens using chatbots daily

• An AI model gets trained in space

• A new eval for real world tasks

Is this the new OpenAI image model?

Today’s Episode

Liked this? Forward it on.

Thanks for reading. Catch you tomorrow. NLW