01 The first 90 days should produce decisions
The practical objective is a small portfolio of capabilities with enough evidence to decide what to scale, redesign, stop or preserve as an option. The following is Koko’s proposed agenda. Targets and thresholds should be set against each enterprise’s baseline and risk appetite; there is no universal ROI hurdle or headcount reduction implied.
| When | Work and deliverable | Accountable owner and partners | Decision gate |
|---|---|---|---|
| Days 1–30 | Map a small set of material enterprise outcomes and the capabilities that deliver them. Select one spanning at least two functions. | CEO/COO sponsor; business owner; CFO | Is the outcome material, measurable and owned? What problem remains if AI is removed from the proposal? |
| Days 1–30 | Establish the value baseline, current delays, quality losses, control obligations and total operating cost. | CFO with capability owner | Can benefits be distinguished from activity, capacity and avoided cost without double counting? |
| Days 1–30 | Map critical data, definitions, policies and decisions. Name a data-product and knowledge owner. | CDO with domain experts and CIO | Are meaning, provenance, permitted use and current policy reliable enough for the chosen action? |
| Days 1–45 | Define and test authority: identities, access, spend, approval, segregation, escalation and recovery. | Capability owner with CISO, CIO, legal/risk | Can an unauthorized action be prevented and a material outcome reconstructed? |
| Days 15–60 | Redesign the work, supervisory load and learning path; provide realistic practice. | CHRO with COO, CAIO and functional experts | Can people handle exceptions and recover the service? Is expertise being developed as routine work changes? |
| Days 30–75 | Run the bounded capability with a meaningful baseline or comparison. Measure quality, value and full cost. | Business owner with CIO/CTO and CFO | Does the improvement survive review, correction, integration, support and transition costs? |
| Days 45–90 | Test reuse in a second context and a supplier exit or recovery exercise. | CIO/CDO with procurement and CISO | Is reuse cheaper and dependable? Can evidence and critical context be exported and interpreted? |
| Day 90 | Decide scale, redesign, stop or hold. Update the scenario and capital assumptions. | Executive portfolio forum; single outcome owner | Is the next capital commitment supported by observed evidence and operational capacity? |
Oliver Wyman highlights foundation-funding gaps; Capgemini treats autonomy as an explicit decision; Deloitte connects orchestration to workforce design. These support the agenda’s direction. The sequence and gates are Koko’s proposed practice. Oliver Wyman ↗ Capgemini ↗ Deloitte ↗
02 Fund data products and knowledge as maintained services
Start with a decision that needs better information. Agree the definitions, ownership, allowed uses, quality expectations and consumers. Fund the maintenance and support needed to keep it useful. Test whether a second capability can use it without rebuilding its meaning.
Knowledge engineering adds the relationships and operating context: why a policy applies, which exception is permitted, which evidence supports a recommendation and which version was used. A context graph is one possible representation. It is not a requirement to purchase a new platform before doing this work.
The investment case should include fewer definition disputes, less repeated interpretation, better decisions and reuse. It should also include maintenance cost, access controls, bias correction and deprecation. Recorded history must not silently become approved policy. BCG ↗ Forrester ↗ Andreessen Horowitz ↗
03 Upskill through work, not attendance
Define proficiency for each changed role. An agent supervisor may need to diagnose a failed workflow, challenge an unsupported conclusion and recognize a material exception. A domain knowledge owner may need to maintain definitions and resolve conflicting evidence. An executive may need to evaluate a capability’s economics and authority boundaries.
Provide supervised practice, feedback and a route to increasing responsibility. Preserve the work experiences that build future experts or replace them with credible learning opportunities. Measure demonstrated ability, adoption, workload and service quality.
Historical task studies and newer surveys show why generic productivity assumptions are insufficient. METR’s 2025 result should not be reused as today’s effect; its 2026 update and survey describe different evidence and limitations. METR ↗ METR ↗ METR ↗
04 Sustain governance as the capability changes
Maintain an agent and service inventory with accountable owners, versions, purposes, permitted actions and dependencies. Assess material changes in models, tools, data, policy or supplier terms. Revalidate controls where the change affects the risk. Capture decisions and actions in a form that can support incident investigation and independent assurance.
Governance should include permission denial, escalation, rollback or recovery, monitoring quality and human response capacity. An inventory alone does not prove control. A guardian agent is another system that must be tested. NIST and the WEF/Capgemini playbook provide useful guidance, while the IIA helps define the separate assurance role. NIST ↗ World Economic Forum ↗ The Institute of Internal Auditors ↗
05 Price the whole capability
A practical value review considers revenue and retention, margin, cash and asset productivity, risk and resilience, reusable capability and strategic options. Those categories are not additive accounting balances.
Include people, models, compute, data, integration, evaluation, supervision, security, support, transition and supplier exit. Separate cash realized, productive capacity released, losses plausibly avoided and options created. Avoid converting every hour saved into payroll reduction.
Procurement should compare buying an application, contracting for an outcome and building a differentiated capability. Sequoia and Bessemer’s services theses broaden the choices; they do not settle the economics. Sequoia Capital ↗ Bessemer Venture Partners ↗
06 Keep the two Future Guides connected
Enterprise Future Guide owns the broader organization, leadership and capability hypothesis. CFO Future Guide owns the deeper finance, Trusted Finance and capital agenda. Each retains its own source records and edition history.
When evidence changes a shared assumption, update the affected passage in both guides and record the reason. Keep shared scenarios consistent, while allowing sector and functional implications to differ. New sources should earn their place by changing a decision, strengthening a rationale or exposing a countercase.