John Roese, Dell's global CTO and chief AI officer, argues that enterprise AI succeeds only when tied to measurable business outcomes, and that the next phase, agents, will digitise an enterprise's skills just as generative AI unlocked its data. He ends up holding that agent-interworking standards will arrive by consensus within the year, that AI will create specific new jobs led by construction, and that viable quantum computing will be a bigger disruption than ChatGPT.
ROI first. Roese says early generative AI enthusiasm produced ideas without context, and that technology matters only when it improves a measurable process: profit, revenue, margin and cost for a company, educational outcomes for a university. AI that gives salespeople more customer time counts; AI that makes them slightly happier is meaningless unless measured.
From 800 ideas to four pillars. After Michael Dell's company-wide note, about 800 ideas arrived, many, Roese jokes, from people who had seen Star Trek. Dell found it could not choose among them, so it asked where AI should apply, picked supply chain, sales, services and engineering, and found the biggest lever in each, such as sellers' content preparation and core coding.
Escaping PoC prison. Dell's AI work began eight or nine years ago with bottom-up permission to experiment, at 500–1,000 projects a year, 80% with no business impact. Roese draws a bright line at production: there must be a material ROI, the project must fit "modern Dell" rather than cover bad processes, and only then come technical viability, security and compliance.
The flywheel. Well-chosen first projects generate ROI, and later ones reuse their technology at lower cost and higher speed. Starting with a goodwill project, Roese says, gives the flywheel no fuel; G&A waited. Sales came last because its data needed fixing, then Dell Sales Chat reached 20,000 sellers. He tells every organisation to start by naming what makes it special.
What agents are. Enterprises differ by proprietary data and unique skills. RAG chatbots unlock the first and account for almost all of Dell's production projects; agents digitise the second, taking an objective and working autonomously. Today's agents are narrow, like geofenced robotaxis or airport trains, and whole-human AGI/ASI is far off. The human moves from in the loop to on the loop, and the architecture adds a knowledge graph and tool use around the LLM.
Teams of agents. Dell's agent systems have always had at least two agents and a human; one flexed to 1,600 agents. The industry lacks ways to authenticate, authorise and share knowledge across companies. Roese credits MCP and Google's Agent2Agent, announced the day before recording, and is "100% confident" of de facto approaches before year-end, ring-fenced at level four autonomy and built from existing tools such as homomorphic encryption.
New jobs. Roese names software composers, thermal plumbers for liquid-cooled GPU clusters, and AI explainers who deliver results such as genomic risk empathetically, and says such roles are appearing at Dell. His caveat: the biggest job creator is construction, an AI build-out larger than the US infrastructure bill of several years ago.
Quantum. Roese says AI is accelerating quantum computing's development, and that once viable at scale, "probably not for a few more years" but not decades, quantum computing will make AI three to five orders of magnitude faster and better. Enterprises, he says, should prepare as if they knew the ChatGPT moment was coming.