Just over a month after shipping GPT-5.1, OpenAI released GPT-5.2 — the latest version of its flagship generative AI model — seeming to confirm what many in the technology world already believed: AI is advancing faster than enterprises can adopt it. In the three years since ChatGPT popularized generative AI, the market has shifted from hype about models to hype about agentic AI, and enterprises often find themselves barely absorbing one technology before Microsoft, AWS, Google and OpenAI announce the next.
Intensity in the AI business
At times, the crowd of AI models and agentic tools simply feels overwhelming — a sentiment voiced repeatedly at the AI Summit in New York in December. Peter Guagenti, CEO of agentic AI vendor EverWorker, which helps companies create and deploy agentic AI workers without code, argued in an interview that the volume of hype and promotional noise from AI leaders makes the landscape genuinely intimidating for buyers rather than easier to navigate.
The promotional energy cuts both ways: it can make enterprises feel behind, or it can make the technology feel compelling. For Naomi Tadesse, an assistant architect at General Motors, walking a summit floor packed with dozens of AI vendors produced both reactions at once — excitement about AI integrated into manufacturing and automotive workflows that speeds up daily work, and unease about job displacement. The overall direction, though, seemed unambiguous: everyone is moving toward AI, and no one wants to be last.
Adoption starts with the use case
Guagenti’s advice to enterprises confronting this environment is to be deliberate rather than fast. “Start small, stay focused,” he said — identify where the pain sits in the business today, where hiring is impossible, and where costs are clearly not adding value, then begin with simple, obvious use cases that combine high reward with low risk. That framing mirrors what independent research consistently finds about the widening gap between AI capability and enterprise readiness: the constraint is rarely the model.
Data comes first
Beyond use-case discipline, speakers converged on a second theme: data. David Swan, a sales engineer at low-code platform vendor Retool, argued during a December 10 panel that trust in AI applications is impossible without visibility into the data underneath them.
Anuradha Mardapu, a manager of data engineering, analytics and data governance at American Airlines, made the complementary point: context matters enormously for AI, and pointing the right data at the right use case is much of the job. Data readiness — verifying that data actually fits the intended application and is properly prepared and governed — is frequently skipped in the rush to deploy. Notably, Mardapu inverted the usual complaint about slow enterprise adoption: from a governance seat, companies are moving too fast, skipping the underlying foundations, and the result is accumulating chaos rather than acceleration. It is the same conclusion reached by advocates of governance-by-design approaches to AI scaling.
Making it accessible
One way to cut through the clutter, Guagenti suggested, is to give everyone in the organization access to AI tools scoped to their department or use case, rather than concentrating them in a technical elite.
NBCUniversal offers an example of that philosophy in practice. Speaking at the conference on December 10, Chris Crayner, executive vice president and chief digital and technology officer for NBCUniversal’s destinations and experiences business, described the company’s approach to applying generative and agentic AI against friction points across the organization. The premise: whether someone is a front-line engagement leader or a data scientist, everyone has friction in their work — and the company’s task is to demystify AI as a tool while being explicit about the lens through which it will be used.
Reading the moment
A few cautions belong alongside this reporting. Nearly every voice urging calm, focus or broader tool access at an AI conference has a commercial stake in the answer — vendors benefit when enterprises buy deliberately but keep buying. The tension between the two dominant complaints (adoption is too slow; adoption is too fast and ungoverned) is real, and both can be true in the same company: fast at the pilot layer, slow and unready at the data and governance layer. For most organizations, the practical resolution remains unglamorous — pick one high-value, low-risk process, fix the data that feeds it, govern it properly, and let the release cycle noise wash past. Model versions will keep arriving faster than any enterprise can absorb them; a working use case compounds regardless of which model number ships next.
