Private view of invited-reviewer visits, time on site, and AI chat interactions. JP activity is excluded from all metrics.
EY has a 90-day opportunity to decide whether it should own the governed systems layer for agentic AI. Keystone is the contained proof: one Gate, one banking proving ground, one partner commitment, one measured agent, and one scale-or-stop decision.
Big tech owns the models. Clouds own the compute. Startups own the speed. EY owns trust, domain complexity, regulated access, and transformation credibility, but that advantage only holds if EY builds the system layer between frontier AI and enterprise adoption. That layer is Keystone.
EY already has the ingredients: a firm-defining bet on EY.ai, AI embedded across proprietary platforms like EY Fabric, which EY reports serves 60,000 clients and 1.5M+ unique users1 and powers 80% of the firm. It has shipped the EY.ai Agentic Platform with NVIDIA,2 EY.ai enterprise private with Dell and NVIDIA,3 and risk solutions on the agentic platform.4 The ingredients exist. What is missing is the governed systems lab that turns frontier capability into regulated enterprise infrastructure.
EY leads where it counts most: 150 tax agents now support 80,000 EY tax professionals,2 and EY.ai for Risk is live.4 But rivals are racing to build the connective infrastructure around their agents, and that is the gap.
Keystone folds every one of these into a single governed lab. The simulation and evaluation harness, regulated digital twins that stress-test banking, healthcare, and energy workflows before they go live, is a foundational system. The workforce-transformation engine is how we build. The proving grounds carry the systems beyond tax and risk into every regulated domain. EY does not need to chase four separate moves. It needs the one lab that makes all four compound.
Apps solve tasks. Foundational systems create new operating capacity. The whole enterprise exists to move the firm, one capability at a time, from the left to the right.
Keystone is EY's proposed frontier systems lab for governed agentic AI. At the executive layer it is exactly four things — each links to its deep dive below.
Forward-deployed teams turn frontier capability into reusable, governed systems — with the hyperscalers, inside EY's constraints.
The enterprise compiled into a live governed environment: policy, controls, data, workflows, agents, and evidence as versioned code.
The trust boundary: identity, evaluation, independence, red-team, attestation. Nothing operates without a passport.
Banking first: real regulated workflows, measured against baselines, on governed data — proof before production.
We don't pitch the transformation. We compile it, run it, and let you watch.
Short on time? Skip the deep dives — the first proof · the operator · the 90-day ask
Keystone runs as three layers: a Frontier Lab that discovers what frontier AI can now do, a Foundational Systems Assembly that turns that capability into reusable governed systems, and Industry Proving Grounds that apply them first in banking, risk, cyber, tax, and finance. Inside, an idea enters at the Frontier and only leaves through Graduation, and only if it clears the Gate.
Most agentic efforts stall on execution, not technology. The fix is not a better model. It is forward deployed teams that embed in the work, rebuild the process around the agent, and reskill the people who run it.
Every engagement upskills the teams it touches: EY professionals and client teams learn to design, govern, and operate agents. The pyramid of manual juniors becomes a diamond of agent supervisors. Training is not a side effect of Keystone. It is how the capability spreads, sticks, and compounds across the firm's 300k+ professionals.
Before Keystone changes a single client, it changes EY. The firm becomes the first user of its own governed agentic systems — Keystone turns trust into infrastructure: policies become rules, controls become runtime checks, quality standards become evaluations. That is what earns the right to sell the transformation, and what lets EY look any regulated client in the eye and say: we did this to ourselves first.
The point of Keystone is not fewer people. It is more capable ones. Scarce expertise — the firm's best risk, tax, and independence judgment — is encoded once, governed at the Gate, and put in reach of every team: 300k+ professionals amplified inside EY, and client teams that keep the capability after every engagement. Capability uplift is the product.
Keystone AI OS turns a regulated enterprise into an executable model. It converts policies, controls, data boundaries, workflows, agents, customer behavior, adversarial risk, and evidence into versioned runtime components — then compiles them into a live environment that can be tested, certified, and improved before production.
The enterprise stops being described in documents. It becomes an executable, certifiable model.
Domain expertise, assurance heritage, independence discipline, control knowledge, delivery patterns.
Versioning, testing, secure delivery, observability, reusable primitives, governed promotion paths.
Agents that work inside defined boundaries — identity, permissions, evidence, evaluation, human authority.
The trust boundary that certifies a system is safe, useful, governed, independent — and ready to graduate.
The new human operating model: experts supervising systems, exceptions, controls, judgment, outcomes.
The simulation is not the product. The simulation proves the product. The product is governed operating capacity.
The first engagement creates the environment. The second improves it. The third scales it. The hundredth becomes an industry operating standard.
We compile it. We run it. We attack it. We certify it. Then we help you operate it.
Keystone AI OS converts EY knowledge into governed enterprise capability: modern dev practices make it executable, AI makes it scalable, the Gate makes it trustworthy, and capability building makes it stick.
The Keystone Truth Layer is the knowledge substrate under the Lab: it ingests EY’s public record, the Keystone thesis, partner capability evidence, market signals and authorized internal material into a versioned evidence graph — claim-level truth, bound to exact sources, scored by authority and freshness.
RAG retrieves documents. The Truth Layer retrieves claims.
The briefing agent on this page runs on it. Every answer now carries its evidence basis — what is EY-official, what Keystone proposes, what is inferred — with claim-level citations. Open it and ask: “What does EY already have, and what is Keystone adding?”
A striking lab is nice. A lab EY can defend to audit committees, banks, and hyperscalers is one EY can put its name on. This is the real moat: not raw compute or agent count, but the ability to industrialize AI under regulatory, ethical, and independence constraints, which is the one thing competitors still only experiment with. Every risk maps to a control at the Gate.
Trust is not a one-time gate. Every governed agent runs a full lifecycle, cradle to retirement, tied to EY's independence and assurance standards. Clearing the Gate issues an Agent Passport: the certificate that the agent is fit for regulated operation.
Non-sensitive parts of the governance framework are published openly, so regulators, academics, and clients can inspect them. Trust through transparency, not just assertion. The whole system runs under a standing board: EY.ai, Risk, Independence, Cyber, Alliances, and Financial Services.
These three workflows are demonstrators of the deeper systems layer. Each has a named buyer, a real pain, and a measurable target. What they prove is the governed substrate underneath, and that is the asset that compounds.
The first outputs may look like banking, risk, tax, or cyber agents. Those are not the product. They are demonstrations of a deeper capability layer: the governed agentic substrate every future EY service will run on.
Policy, risk, independence, data access and controls are embedded before it operates.
Every material output traces to the data, agent, workflow, policy, tool and evidence that produced it.
It performs real enterprise work — not demo-path automation.
Agents, workflows, controls, prompts, tests and evidence patterns become reusable primitives.
Every deployment improves the library. Every library improves the next deployment.
EY's edge is not compute or agent count. It is trust, independence discipline, regulated domain complexity, and live EY.ai momentum. Each side brings what the others lack, and the Lab is where they meet under one governance spine.
Partners contribute compute, embedded engineers, and early access to frontier models and hardware. EY contributes regulated-enterprise reference deployments, domain-governed blueprints, and co-go-to-market. Big tech gets a trusted beachhead into banking and other regulated industries it cannot reach alone. EY gets frontier capability and embedded engineering it could not stand up at this speed on its own.
EY owns the Lab as a strategic asset. Then it draws in regulated industries, banks first: they have the regulatory pressure and no safe way to experiment with agents in production. The Lab gives them a governed proving ground. Then insurers, life sciences, healthcare, energy, and government.
The market puts the Lab to work. Client use cases and governed data make every system better. The best systems clear the Gate, graduate into EY delivery and the shared systems library, and every engagement makes the next build faster. The capability compounds.
The execution risk in a venture like this is not ambition. It is the operator search. The job is to stand at the intersection of executive strategy, engineering reality, regulated-enterprise risk, security, independence, model governance, and hyperscaler execution — and hold all of those rooms at the same time. That profile is rare. It is the job John Petty has already been living.
IC-grade analytic tradecraft and adversarial thinking — the discipline of assuming a capable adversary.
Built agentic product capability end to end — from zero to a working system.
Inside a regulated institution where independence, audit, model risk, data sovereignty, and cyber resilience are the environment, not theory.
The operating model, the Gate, and the systems architecture that make AI-native transformation governable, repeatable, and scalable.
He translates fluently across C-suite buyers, engineers, risk leaders, security teams, alliance partners, and hyperscalers — moving from vision to operating model, operating model to systems architecture, and architecture to governed deployment.
Executive leadership of the Keystone Lab — a firm-level AI systems venture reporting into EY.ai leadership, with the authority to coordinate across practices, industries, alliances, risk, independence, cyber, and delivery.
A multidisciplinary organization, operating horizontally across the firm.
Not another AI offering. An operating system for trusted agentic transformation — one EY can run, prove, sell, and scale.
The ask is contained: approve a 90-day foundation — operating model, governance gate, partner terms, first prototype — then make a scale-or-stop decision on evidence, not enthusiasm.
Keystone runs as three layers, each proving the next. Nothing scales until the layer beneath it is certified at the Gate. EY commits to a stage, sees the proof, then opens the next.
The visible outputs are banking, risk, tax, and cyber demonstrators. The real asset is the governed agentic operating layer underneath them. That is what compounds, and what big tech cannot build alone.
The deepest objection to any agentic venture is binary risk: it works, or the money is gone. Keystone is engineered so that is never the choice. Two design moves remove the all-or-nothing.
A basket of agents, each one a one-off build. If a given agent misses, that spend is gone. Binary, fragile, and hard to defend to a board.
The governed machine that builds repeatably: the Gate, the playbooks, the eval harness, the upskilled people. Apps, solutions, agents and modernized core systems are the output. The capability is the durable asset that compounds.
No party bets the farm. Every gate produces reusable value, partner leverage, client proof, and governance capability. Limited downside. Measured upside. Reusable assets. Clear exit ramps.
The fallback is never "we lost money." For every party, the fallback is a real asset.
Reusable Gate methodology, control models, delivery playbooks, and marketplace packaging, plus sharper EY.ai delivery, governance, and internal productivity.
Regulated reference deployments, buyer signal, and co-sell paths, with preferred status by use case rather than platform-wide lock-in.
A governance assessment, workflow value map, and AI control blueprint: a board-ready path to agentic adoption, on client-owned data.
"If it scales, EY owns the governed systems layer the whole market will run on. If it stops early, EY still owns the control model, reference architecture, client pipeline, and delivery playbooks. This is not an innovation spend. It is a staged strategic option."
A governed agentic capability performing real enterprise work.
Agents, workflows, prompts, tools, controls, evaluations and evidence templates — ready to use again.
People who know how to supervise, govern, improve and scale the system.
The first build proves the pattern. The second improves it. The third accelerates. The hundredth becomes an industry operating standard.
EY-public facts are cited to EY's newsroom. Peer moves are cited to the firms' own publications. Everything Keystone proposes is labeled a proposal, not a fact. The references used above:
Anything not cited here — the Lab, the OS, the Gate, the proving grounds, the agent targets — is Keystone-proposed: authored strategy for EY leadership to evaluate, not an existing EY program.