Technology & Data: The Foundation Teams Stand On
Capabilities four and five. A distributed, easy-to-use technology environment and data embedded everywhere — the shared foundation that lets every pod move fast without rebuilding the basics.
If every pod has to rebuild the basics before it can start, you don't have ten teams moving fast. You have ten teams paving the same road.
Capabilities four and five pair naturally, because they answer the same question: what does a pod stand on? Capability four is the technology environment — the platforms, tools and infrastructure teams build with. The word McKinsey leans on is distributed: instead of a central group every team must queue behind, the building blocks are self-serve services any pod can pick up without asking permission. Capability five is data — not locked in silos, but organised, trustworthy, and reachable wherever a decision or model needs it. Treat them as foundation, not features: build it once, make it self-serve, and every pod starts halfway up the hill instead of standing up its own plumbing.
Data is now the bottleneck for agents
If the 2023 edition treated data as the fuel for analytics and models, the agentic era raised the bar again. The single most common reason agent programmes stall isn't the model — it's the data: roughly eight in ten organisations cite data limitations as what blocks them from scaling agents. Agents need to read messy, unstructured data — documents, tickets, transcripts — and act on it reliably, which most data foundations were never built for. So capability five now climbs a longer ladder:
Why “easy to use” is the whole point
A foundation that's powerful but painful gets routed around. If using the shared platform means a two-week ticket and a meeting, pods will quietly build their own, and you're back to ten roads. The same goes for data that technically exists but takes a month and three approvals to reach — it may as well not. Usable beats complete. The test isn't whether the capability is there; it's whether a pod reaches for it by default, because reaching for it is easier than going around it.
Where it goes wrong
Two opposite ways. No foundation at all — every pod a sovereign island rebuilding deployment and data access, until you have ten incompatible stacks and no leverage. Or over-centralising in reaction — one platform team that owns everything and becomes the bottleneck every pod waits behind. The narrow path is shared building blocks that are genuinely self-serve: common where it helps, optional where it doesn't, never a gate.
Try this
Ask a team trying to ship something new: how long from idea to running in front of a user, and how much of that time is plumbing they're rebuilding rather than the actual work? The plumbing fraction is the size of your missing foundation. The closer it gets to zero, the more your pods are doing the work you hired them for instead of paving roads.
Grounded in Lamarre, Smaje & Zemmel, Rewired (McKinsey), and McKinsey on building the data foundations for agentic AI at scale.