Huang and Dell Say Agentic AI Cuts Software Time at Dell World

At Dell Technologies World 2026 in Las Vegas, Nvidia CEO Jensen Huang joined Dell CEO Michael Dell onstage to argue that agentic AI — generative models given planning, memory and action capabilities — has reached practical use for enterprise tasks and can shorten software development cycles.

During Dell’s opening keynote, Huang traced a technical progression from content-generating models to systems that can reason and plan. He said generative models can “generate content to think with” and described the next step as agentic systems that carry out multi-step work. Huang estimated that agents can reduce some programming tasks from a month to a week and reported that Nvidia is using agents across software development, DevOps, site reliability engineering and continuous integration and testing.

Michael Dell addressed partners later in the conference and described the technology as early but improving. He said broad, large-scale adoption across millions of companies has not yet arrived and added that model capabilities are still advancing. “The models, they’re great! But they’re the worst they will ever be, and they’re going to keep improving,” he told attendees.

Steven Dickens, CEO and principal analyst at HyperFRAME Research, cautioned that usefulness varies by task and user. “Usefulness is in the eye of the beholder,” he observed, noting that earlier generative features were most helpful for summarization and content writing. He said agentic systems add value by performing actions as well as producing content and predicted wider acceptance as commercial tools make agent workflows easier to set up.

Speakers at the event pointed to concrete business uses. Presentations and demonstrations showed agents automating multi-step workflows such as software test cycles, managing CI/CD pipelines and supporting operations teams with proactive incident management. Executives said agents can coordinate tools, make intermediate decisions and carry out sequences of tasks that once needed human orchestration.

A practical barrier discussed at the show is resource consumption. Agentic systems typically use many model tokens and long compute runs, which drives up public cloud bills. Dell executives gave an internal example in which a single developer ran up about $3,400 in token costs over 24 hours while testing agents. They proposed on-premises hosting as one way to lower per-token costs, while noting that on-prem deployment requires companies to buy servers and accelerators and bear that capital expense.

Organizers and panelists also identified issues companies must address before moving agents into production: readiness, governance, cost and the need for human oversight. They said the technology is still in an early-adopter phase and that broader adoption will depend on easier tooling, cost management and operational safeguards.

Agentic AI refers to systems that combine large language models with planning and action capabilities so they can pursue multi-step goals with some autonomy. Proponents at Dell World described it as a next step beyond single-response generative models; critics and analysts highlighted limits and operational challenges that firms will need to resolve before scaling deployments.