How frontier firms are pulling ahead

OpenAI has a new report out, self-branded as B2B Signals, and the framing is blunt: some companies are getting dramatically more value from AI than others, and the gap is not closing. It is widening. The firms OpenAI calls frontier enterprises are not just using ChatGPT for drafting emails anymore. They are wiring Codex into real engineering pipelines, letting it write, review, and ship code as part of agentic workflows that run with minimal human babysitting.

That shift matters because it changes what counts as a competitive edge. A few years ago, having access to a capable model was the differentiator. Now the differentiator is organizational: how deeply a company has restructured its processes around agents that can plan, execute, and iterate on their own. OpenAI's argument, unsurprisingly self-serving but not baseless, is that the companies doing this well are building an advantage that compounds. Every month spent refining agentic workflows produces institutional muscle memory that a competitor starting from scratch cannot buy overnight.

Codex is doing a lot of the heavy lifting in this narrative. It is the wedge product letting engineering teams hand off entire chunks of work, not just autocomplete suggestions. And once a team trusts an agent enough to let it touch production code, the door opens to using similar agent patterns elsewhere in the business, from customer support triage to internal data analysis. OpenAI frames this as a flywheel: deeper adoption breeds more trust, more trust breeds broader deployment, broader deployment breeds harder-to-replicate advantage.

The uncomfortable subtext is that this is also a story about inequality between companies. If frontier firms really are separating from the pack the way OpenAI describes, then plenty of businesses risk falling permanently behind simply because they moved cautiously or lacked the engineering culture to absorb agentic tools quickly. That is a harder problem than just buying a subscription.