IBM's Arvind Krishna: 'Pick Three to Five Things You Can Scale Like Crazy'

“Please don’t do 100 experiments — pick three, four, five things which you can scale like crazy.” That’s IBM CEO Arvind Krishna’s central advice to enterprise leaders on Bain’s Winning with AI podcast, drawn from a decade of his own company’s trial and error. Krishna is candid about IBM’s mistakes after the 2011 Watson launch: it built monolithic vertical applications instead of modular components, and it entered healthcare without doctor relationships or regulatory expertise. The lesson shaped IBM’s current scaling program, which targets roughly 200 enterprise processes at $100 million-plus investment thresholds — 60 leaders volunteered to pilot it, about 60 processes are now complete, and 70 more are queued. As one concrete example, Krishna described an employee-verification process that used to require 17 human touchpoints and roughly an hour of work, now resolved through a 15-second AI agent interaction. IBM has proven out three use cases reliably enough to scale: customer experience, customer service, and coding.

Krishna also predicted quantum computing reaches business-impact breakthroughs by 2029 — his phrase was “the ChatGPT moment of 2022” — pointing to Cleveland Clinic’s modeling of 12,000-atom protein fragments as early evidence. That timeline runs well ahead of Gartner’s own read: VP Analyst Chirag Dekate has said true quantum computing “is not ready for any production AI workload and will most likely not be for the rest of this decade,” and that most vendor “quantum AI” marketing actually describes hybrid or quantum-inspired techniques running on ordinary classical hardware. Gartner puts no peer-reviewed evidence behind quantum advantage on any production AI workload through at least 2028.

The disagreement matters less than what it reveals: even IBM’s own CEO is willing to bet on a specific, falsifiable date for a technology Gartner says isn’t close — a useful reminder that “the era of experimentation is over,” Krishna’s framing for narrow AI, doesn’t yet apply to quantum. Enterprises should scale the AI use cases with proof behind them now, and treat quantum timelines as exactly that: a bet, not a roadmap.