In this article PwC’s Tim Walding and Roshini Rajan explain how dependable, repeatable, industrial-grade AI requires a shift from fragmented pilots to governed, end-to-end agentic workflows.
There’s a pattern playing out in boardrooms across nearly every industry. Nearly 80% of executives say they are still at least 12 months away from meaningful returns from their AI investments, according to PwC’s 2026 Global CEO Survey, and deciding where to place those bets is a strategic choice. PwC’s lead, lag, or exit framework gives CEOs a way to prioritize. The problem may not always be ambition or technology. It’s that enterprise AI is still largely being built as a series of disconnected point solutions without a shared foundation. Until that changes, the gap between AI investment and AI value may continue to widen.
The handcrafting problem
Let’s consider how enterprise AI is being built in companies today. A team spots a use case, spins up a model, builds the integrations, tests it, and ships. Then a second team does the same in a different system, with its own guardrails. The cycle repeats across the company. Every initiative starts from scratch. None share any common infrastructure underneath.
The result is what you might call artisanal AI: bespoke, duplicative, and fundamentally unscalable. Every new use case starts from scratch. Every agent operates in its own silo. There’s very little shared context, no common orchestration layer, no unified approach to governance.
The companies seeing returns on their AI investments aren’t always smarter or luckier. They’ve stopped treating each AI initiative as a standalone build and started treating AI as something that needs scaling infrastructure—a shared, governed foundation from which individual use cases can be deployed.
What the shift actually delivers
PwC has seen clients achieve roughly 30% reduction in core operational workflow cost after making this transition.
A few themes mark the foundation of enterprise AI:
- Generalist agents cap scale and concentrate risk. Purpose-built agents define roles, work together in an end-to-end workflow, and help enhance accuracy and reusability.
- Centralized, evolving memory replaces agents carrying their own context, confirming that knowledge persists across the entire workflow.
- Centralized orchestration and embedded governance enable systems to run themselves, with humans handling exceptions.
All of this is enabled by a strong foundation that provides scalable compute and data access for all agentic systems.
None of this requires ripping out the foundation that already exists. The new AI layer sits on top of the systems you already run, tapping into them for data, business rules, and security rather than replacing them.
From handcrafted to industrial
The real question for many enterprises isn’t whether AI will pay off. It’s whether their architecture is built to drive the return. That architecture rests on five layers and each earns its keep. A common foundation lets you build once and reuse the same orchestration, controls, identity, and telemetry everywhere. Embedded governance bakes in trust from the start, so quality, cost, and risk stay consistent at scale. Centralized orchestration lets systems run themselves and routes humans to the exceptions that need them. Redesigned workflows move work end to end rather than layering AI onto old tasks. Purpose-built agents replace risky generalists, separating execution from validation so accuracy and reuse can climb together.
Fragmented deployments rebuild integrations, controls, and monitoring with every use case, and costs compound. Consolidate onto one platform and the math inverts: The upfront investment amortizes, each new agent is a marginal add, and simpler tasks route to cheaper models. That is the difference between AI as spend and AI as return.
This transformation starts with three moves:
- Pick your highest value big bets.
- Build architecture and governance into your DNA, not bolted on later.
- Invest in your workforce and help build capabilities of the future.
The 30% isn’t a ceiling. It’s what becomes possible when architecture becomes strategy. For a deeper look, see “From pilots to production: AI architecture that scales” at PwC.com.
Note: This content was created by PwC.