Deep Dive
Three-layer strategy for AI adoption
Kunal Malik opens by rejecting the incrementalist approach. Bolting AI into specific tasks isn't enough; you need to rethink the entire workflow. OpenAI structures this as three concentric layers: amplifying individuals with AI tools so they get 5-10x throughput, scaling that to team-level end-to-end workflows, then company-wide. The operating model shift is stark — everyone becomes a builder because the people closest to the actual problem are best positioned to solve it with AI. This flips traditional IT hierarchy, where problems flowed up to centralized help desk queues.
Replacing the help desk ticket bottleneck
The old model: file a ticket, wait hours or days for an IT human. The new model: an AI agent responds instantly, either unblocking you with guidance, escalating to a human with full context, or resolving it outright. This freed the human help desk team to focus on high-security tasks and problems only humans can solve. Malik notes the ripple effect: support staff started building their own agents and micro-sites to handle routine work, reinventing their own jobs in the process. The validation loop matters — teams share learnings, other teams accelerate past them, the whole organization moves faster.
Software licensing audited in weeks, not months
OpenAI had hundreds, possibly thousands, of software licenses scattered across the company with no unified view. ChatGPT grouped them into roughly 30 categories and delivered an initial assessment: keep this third, retain that third, defer the last third. The analysis identified that 70% of software needs could be met with existing tools without buying new software. Malik is explicit: this analysis would've taken him a couple months to produce manually. The speed matters because fast-growing companies can't afford quarterly software audits — they need real-time visibility. The CIO recommendation is blunt: make big bets on the right foundation, then focus on experiment-test-deploy cycles to unlock AI value faster.