The New A.I. Power Class Will Be Built on Context, Not Compute

It’s been more than three and a half years since ChatGPT was launched, and now nearly every company has an A.I. policy, an enterprise A.I. subscription and increasingly strong opinions about models, tokens, token budgets and how A.I. should be deployed internally. If you go by the headlines, you’d think A.I. is going to replace all white-collar work and leave millions without jobs. But the reality is more complicated. MIT’s 2025 GenAI Divide report found that only five percent of the integrated generative A.I. pilots were extracting significant value. The other 95 percent showed no measurable profit-and-loss impact. 

What’s going on here? It’s easy for executives to look at their own organization’s lack of progress and conclude that A.I. is all hype or that today’s models simply aren’t smart enough. This is the wrong takeaway. The reality is that the models lack one of the most valuable assets an organization has: context. Most organizations haven’t structured that context in a way that A.I. systems can actually use. 

Anyone who has watched financial news or talked with a financial advisor has heard about the first A.I. power class: the labs, the chipmakers, the “hyperscalers” and the investors funding the scale-up. My own financial advisor once told me the best hedge against the downside scenario of A.I. replacing jobs was to own the infrastructure. This is probably right.

But there’s a second A.I. power class emerging, and it’s far more subtle. It’s the gatekeepers. The people who decide what context enters the system, how it’s organized, and what the organization defines as “good.” Put differently, the new power broker inside companies is the arbiter of “taste.” They decide what the A.I. is allowed to learn from, what information it can trust and what standards it should emulate.

How does a new employee ramp up to creating value after their first 90 to 180 days within an organization? It’s the repeated feedback from their team that refines their initial output into something they actually send to a client, partner, customer or executive. It’s not that they read all of the onboarding materials and listened in on dozens of Zoom meetings. It happens through the edits and repeated rounds of feedback. Inaccuracies get fixed. Language and tone sharpen. Client context is added. The employee’s output moves closer to what the organization considers good. The employees who ramp quickly are described as the ones who “just get it.” What they’re really absorbing is judgment. 

The gap between the initial draft and the final version captures taste, judgment, client context, institutional knowledge and an organization’s tolerance for risk. I call it the edit trail, and it becomes the organization’s proprietary context asset.

Take a communications team responding to a negative story. You could drop the story into ChatGPT and instantly get a concise summary and a plausible response. It will probably include familiar language about how the company “takes this matter seriously.” But the real work is not in producing a plausible statement. It’s knowing what the company can actually say. Have we used that phrasing before? Which executive can credibly deliver this message? Which journalists have covered the company critically or favorably? How did the last statement change after communications, legal and the CEO signed off on it? What will legal approve?

That context almost never lives in a single document. It lives in the edit trail, the difference between the first draft and that final version that was ultimately published. A company that captures this trail is not just saving drafts. It’s codifying what good judgment looks like inside the organization. The person who decides which of those edits becomes reusable institutional knowledge is, in effect, shaping how the organization’s A.I. systems learn. 

Internally, organizations have an advantage. The curator of context chooses which information, processes and data A.I. systems have access to. Externally, however, there’s a far greater challenge. 

As more people turn to A.I. engines in their decision-making process, reputation within the A.I. layer is more important than ever. Companies need to understand which sources influence A.I. answers, what narratives those sources reinforce and where misinformation or outdated information enters the system. 

A prospective customer will likely ask an A.I. assistant how your company compares to competitors. A software engineer may ask Claude what it’s like working for you. An investor may ask about the biggest risks facing your industry. It’s therefore equally important for companies to think as carefully about curating the context available to public A.I. engines as they do the context inside their own walls. They also need feedback loops to ensure that information stays accurate and current over time. 

A.I. may not be changing everything, but it’s unquestionably changing something. Models will become smarter, faster and cheaper. What companies can’t afford to do right now is nothing. 

An easy place to start is by asking yourself—and then two other colleagues—the same question: Who’s the best editor in this organization? If the same name comes up more than once, then there’s a high probability you’ve identified a good candidate to become your curator of context. Should that person sit in IT? Product? Brand? Communications? Legal? The answer is going to be different for each organization. 

The role will likely have many different titles, but its function matters more than its name. The important thing is to begin. Someone needs to own the questions of what organizational context A.I. can access, what “good” looks like and how the company’s edit trail can be transformed into a reusable strategic asset. That is where the next A.I. power class is being built.