CFO 2030 and 2035: who orchestrates enterprise value?

Koko: Welcome to the CFO Future Guide. I'm Koko, joined by Sam and Max. Our starting point: CFOs must protect, create and orchestrate enterprise value through Trusted Finance. By 2030, safe delegation could depend on financial meaning, valid authority and verified outcomes. By 2035, that ability may constrain growth. The question is how to earn the right to delegate.

Sam: Sam here. PwC and Deloitte connect reliable data and governance with useful AI in finance, but guidance and selected survey evidence do not prove delegation outcomes. What makes trust a growth constraint rather than a compliance checkbox?

Koko: Conditional hypothesis: if agents act faster than your authorization chain can verify, exceptions accumulate faster than anyone can clear them. Hackett flags segregation of duties and ongoing control monitoring as scaling requirements, not one-time setup tasks.

Max: Max here. Operationally: who owns the boundary? If an agent prepares a cash transfer and nobody has named the approver for that amount and counterparty, the control doesn't exist. The policy document just says it does.

Sam: But more controls slow decisions too. The real alternative is that governance theater costs more than the errors it prevents. What signal would change your view?

Koko: If review effort grows faster than agent-delivered value, that's the stranded-intelligence warning from our 2035 scenarios. Delegation scope should shrink until the authorization chain is clean.

Max: The action is concrete: trace one material cash decision end to end, name every owner, test an unauthorized action. Decision gate: no clean trace, no expanded autonomy. Internal audit stays independent throughout.

Koko: Next we look at what happens when knowledge itself becomes a finance product.

Koko: Conditional hypothesis: reusable knowledge may compound in ways model access alone doesn't. By 2030, finance could help design enterprise decisions. By 2035, a record of context and outcomes might improve those decisions over time. That's the claim, not established fact.

Sam: Foundation Capital calls decision context a trillion-dollar layer, but that's an investor thesis, not a verified market estimate. And recorded history preserves mistakes too. A concession made during a supply crisis shouldn't quietly become standing policy.

Max: That is a plausible operating risk: a temporary exception could become a default. What would make knowledge compound in value rather than just accumulate?

Koko: Clear ownership and current policy. Dehghani's data mesh article emphasizes domain ownership and data as a product. Our extension is to distinguish authoritative policy from historical context. W3C's provenance vocabulary helps record who acted, on what and when; it does not prove correctness.

Sam: McKinsey's FP&A agent piece envisions anticipating performance gaps, but the cases are illustrative. We still need to measure maintenance and adoption costs. What's your decision gate before scaling?

Max: Reuse reducing expert rework on a second workflow. One domain proving out is evidence; two domains drawing on the same data product is the signal.

Koko: That's the action: CDO with finance as co-sponsor funds one decision domain—agreed definitions, lineage, policy versions, explicit ownership, outcome measures. Scale only after reuse improves the economics.

Sam: If definition disputes persist or agents perform worse because stale rules were treated as policy, the hypothesis weakens. Simpler models and direct judgment may outperform the maintenance cost.

Max: So by 2035, the question is whether that maintained evidence record becomes strategically important as autonomy expands—but only if it's been governed, not just stored.

Koko: The next hypothesis: headless SaaS and agentic protocols could shift bargaining power toward whoever controls decision context and execution rights. SAP describes an architecture direction; Salesforce documents API-invoked agents without a screen. Those are vendor perspectives. The CFO question is who owns the meaning behind the action.

Sam: MCP security guidance addresses authorization risks. Koko's extension is that a working connector doesn't establish financial authority or meaning. Two agents can exchange messages and still disagree on a credit exception. Communication protocols don't replace reconciliation.

Max: If I can't export my decision records when I want to change a supplier, I've already lost the negotiation. Oliver Wyman argues portability and audit rights must be contracted before dependence grows, not after.

Koko: McKinsey's agent-economics interview, with a measurement vendor so caveat applies, argues real cost is per accepted outcome including rework. If context is locked to one platform, exit cost compounds quietly.

Sam: Alternative: this stays a stranded-intelligence problem. Enterprises buy orchestration, reconciliation breaks increase, and the CFO funds exception management they never budgeted. What changes your view?

Koko: If a supplier starts charging separately for decision-record export, bargaining power has already moved. That's the signal. The hypothesis weakens if open protocols genuinely keep context portable at no incremental cost.

Max: The action is concrete: CFO and CIO run one real workflow across two applications, export the evidence, verify a prohibited action is blocked, price the supplier exit. Do that before committing capital.

Sam: If evidence can't be exported or the block doesn't hold, that's your gate. By 2030 you want to know which decisions depend on that layer. By 2035, in the compounding scenario, that context is strategic capital.

Koko: Here's the workforce hypothesis. By 2030, finance roles may shift toward exception judgment and decision design. By 2035, apprenticeship could be the scarcer asset. The risk isn't automation itself. It's harvesting efficiency today while quietly emptying the learning pipeline tomorrow.

Sam: Gartner's 2026 survey found finance AI investments lean heavily toward productivity, not decision quality. That's a stated priority, not a realized outcome. And EY flags a gap between value ambition and actual leadership action. So the pipeline risk may already be underpriced.

Max: Operationally, the question is: what replaces the work junior staff used to learn from? Automating reconciliations before redesigning the learning path could weaken judgment over time. That is a risk to test, not an inevitable result.

Koko: Exactly. IBM's execution research links AI benefits to operating maturity, though it's survey-based, not causal. The pattern suggests workforce capability is a prerequisite, not a byproduct.

Sam: The counterclaim: maybe the new generation learns faster through supervised exception review than through manual reconciliation. That would change the hypothesis. We'd need to see decision quality improve, not just hours saved.

Max: The signal to watch is whether senior reviewers can actually explain and override model outputs. If they can't, the expertise gap is already visible, regardless of what the efficiency dashboard shows.

Koko: The action is concrete. Map changed tasks and decision rights now. Build supervised practice on real exception workflows. Measure adoption and decision quality, not training attendance. The CFO owns this with the CHRO.

Sam: The decision gate matters: before booking any efficiency benefit, show demonstrated proficiency evidence at the next workforce review. That connects 2030 execution capacity directly to the 2035 judgment mandate.

Koko: Here is the conditional hypothesis. Quantum, physical AI and compute are advancing fast enough that a fixed technology roadmap may be worth less than staged commitments with explicit expand, pause and exit gates. The CFO needs a portfolio structure that moves with evidence, not vendor confidence.

Sam: Koko's review of Gartner's physical-AI research covers only its abstract, which warns about hype, risk and uncertain outcomes. A demo working in a lab is not a fleet working at operating cost in your environment.

Max: Bain's 2025 humanoid analysis discusses dexterity, power and operating conditions. Koko's financial test is productive hours, actual uptime and intervention costs. Those older assumptions need checking against the proposed deployment.

Koko: DARPA's benchmarking initiative is trying to separate quantum hype from reproducible economic value, with a 2033 evaluation horizon. That is a program milestone, not a commercial date. The IEA adds that lower energy per task can coexist with rising total consumption.

Sam: The alternative scenario: slower progress makes disciplined optionality beat speculative commitment. What changes the hypothesis is a use case that clears a classical comparator on full cost, not just throughput.

Max: One action independent of all that uncertainty: NIST has standardized post-quantum algorithms now. The CFO and CISO can fund a cryptographic inventory and migration plan without waiting on any quantum computing business case.

Koko: That is the structure: separate foundation funding like post-quantum security, bounded pilots with real cost accounting, and frontier options that stay small until evidence justifies more. The gate is unit economics in the intended environment, not the demo room.

Sam: And that connects 2030 to 2035. Options held cheaply now, with clear triggers, let the CFO move fast if evidence arrives, without stranding capital if it does not.

Koko: Final hypothesis: agent governance could move from occasional compliance review to sustained enforcement at every material action. By 2030, the CFO may need a running control layer. By 2035, jurisdictional differences may force genuinely different operating models. That's the conditional claim.

Sam: I'll push on it. NIST's AI Agent Standards Initiative is a work program, not a completed regime. Workday's agent registry was announced as in development. A dashboard showing which agents exist doesn't prove any of them are actually constrained.

Max: With governance tools from three vendors, the question is whether their controls reconcile. We should test that, not assume failure or success. Hackett argues governance needs ongoing funding, not just launch spending. That is an operating-budget decision.

Koko: ServiceNow documents reviewing risk posture. Koko's interpretation is that visibility must be distinguished from permission enforcement inside transaction systems. Binding authorization rules with audit requirements, or a material loss tied to an ungoverned action, would change the investment case.

Sam: If standards stay voluntary, governance remains a differentiator for some and an afterthought for others. Nothing in the current landscape guarantees convergence by either horizon.

Max: So the action is one owner, one live register of agents, permissions and obligations. Test that a prohibited action is actually blocked and rollback works. That's the acceptance test before expanding any operating envelope.

Koko: That is the through-line for Koko Knows: earn the right to delegate. The CFO 2035 reading and value agenda set out four conditional futures and five investment tests. Choose one material decision, name its owner, and fund the evidence and controls it needs before buying more autonomy.