Deloitte’s manufacturing research puts a number on something worth sitting with: agentic AI adoption in factories is expected to roughly quadruple in two years — from 6% to 24%. The industry’s own framing of why is blunt: copilots give answers, agents act.
That’s the “What, So What, Now What” gap in industry language. A copilot tells you what happened — demand shifted, a machine’s vibration changed. That’s the “What,” maybe stretching into “So What” if it explains why. An agent goes further: it infers the cause, checks what’s actually available, and proposes — or in bounded cases, takes — the next step. That’s “Now What,” done by software instead of a person digging through three systems to get there.
Here’s the detail that matters more than the excitement: even in the most advanced deployments, the autonomy is bounded by a human-in-the-loop gate, by design. The agent compresses the time to a decision. The human still owns the decision. That’s not a limitation vendors are apologizing for — it’s the actual product. Nobody serious is selling “AI decides for you.” They’re selling “AI gets you to the decision point faster, with the legwork already done.”
That’s the version of “Now What” worth wanting. Not AI that acts without you, but AI that stops making you do the “What” and “So What” work by hand, so the only thing left for you to do is the part that was always yours: the decision.
So what — of the tools your team uses today, how many actually get you to a decision, versus just a faster answer? Now what — what’s one recurring decision on your floor where the legwork, not the judgment, is the bottleneck?
Source: Deloitte, “From vision to value” (via Manufacturing Leadership Council survey)
Leave a comment