Publication date: Friday, 9 October 2026
Author: Malik Carter, Chief Intelligence Architect
Deverout and Associates | Strategic Intelligence for Transformation Leaders
The first AI challenge was access. The next is endurance: can the organisation make AI fit the work, understand its cost, and keep accountability visible as authority expands?
The Signal This Week
Enterprise AI is moving beyond experimentation. That sounds like progress, but it creates a more demanding problem. A tool can be impressive in a demonstration and still fail when it meets the organisation’s actual processes, data, economics, and responsibilities.
Three current signals converge. McKinsey’s October analysis argues that agentic AI is changing the constraints that shaped the modern organisation—human capacity, expertise, and coordination—and reports that 80% of respondents saw individual productivity gains while only 37% reported positive EBIT impact. Its strongest performers were far more likely to redesign workflows rather than simply add tools. Infor’s October research, based on a YouGov survey of 2,111 decision-makers across seven countries, reports that 68% of businesses believe off-the-shelf AI does not adequately address their industry-specific challenges, while 44% of AI-generated insights and workflows still require expert review or adjustment. The Linux Foundation’s October European research reports that 94% of organisations consider digital sovereignty important to their technology strategy, yet 31% still lack a formal open-source governance framework.
The common signal is simple: adoption is no longer the finish line. It is the point at which fit, economics, governance, and human judgement become unavoidable.
The Tool Has Entered the Work
Earlier AI programmes could be evaluated as applications: Was the model useful? Did people use it? Did the pilot produce a promising result?
That approach weakens when AI begins to participate in consequential work. An agent may access enterprise data, coordinate multiple systems, make recommendations, trigger actions, and hand work to another agent or human. The question is no longer only whether the system can perform a task. It is whether the organisation can operate the full chain.
McKinsey’s analysis describes this shift as a movement from an organisation chart to an orchestration chart: start with the outcome, then assemble the humans and agents needed to deliver it. Its reported 2026 survey figures are a warning against confusing personal productivity with enterprise value. The organisation must redesign work end to end, remove unnecessary handoffs, clarify decision rights, and make human accountability explicit.
This is the difference between adding intelligence to an existing process and redesigning the process around what intelligence now makes possible.
Fit Is a Strategic Requirement
Infor reports that 68% of surveyed businesses believe off-the-shelf AI does not adequately address challenges specific to their industry or business. The research was conducted by YouGov between 5 and 19 August 2026 across the United States, United Kingdom, Germany, France, Australia, Singapore, and Japan. It is vendor-sponsored research, so its figures should not be treated as independent universal benchmarks. The signal is nevertheless useful: technical capability is not the same as operational fit.
Fit includes the language of the work, the exceptions people understand implicitly, the quality of local data, the regulatory context, customer commitments, and the consequences of a wrong answer. It also includes culture. A workflow designed for one jurisdiction may carry different assumptions about authority, privacy, hierarchy, or acceptable risk in another.
The global AI conversation therefore needs a local translation layer. A model may travel; the operating model cannot simply be copied. Leaders must ask what must remain common across the enterprise and what must adapt to the community, sector, and jurisdiction being served.
The Cost Is Not the Licence
The second part of the post-adoption gap is economic visibility. The cost of AI grows through usage, data movement, model calls, system integration, human review, security controls, exceptions, rework, and the operational complexity of maintaining multiple agents.
Infor’s analysis notes that businesses are increasingly giving AI a role in consequential decisions while 44% of AI-generated insights and workflows still require expert review or adjustment. That review is not necessarily a failure. It may be the correct control. But it must be included in the operating economics.
The relevant calculation is not model price × number of users. It is total cost to achieve a trusted business outcome. That includes the cost of supervision, incidents, delay, data preparation, integration, training, change management, and the opportunity cost of work that remains unresolved.
Accountability Must Follow Authority
Infor reports that responsibility for AI governance is distributed across leadership: 23% name the CEO or executive leadership, 22% the CIO or CTO, 15% an AI committee or governance group, 10% a dedicated Chief AI Officer, 10% department heads, and 15% say responsibility is unclear or held by no single person. The study shows why a governance label is not enough. Every consequential AI action requires an identifiable owner.
The Linux Foundation’s 2026 State of Open Source in Europe report adds a broader sovereignty dimension. It reports that 94% of European organisations consider digital sovereignty important, while organisations with advanced open-source governance report an average return of more than four times their investment, compared with 3.6 times among organisations without formal governance. It also reports that 31% lack a formal open-source governance framework. These are report findings, not causal proof, but they point to a strategic pattern: control is strengthened not only by consuming trusted technology, but by participating in the systems and communities that govern it.
Accountability should therefore be designed at the point of work. The person who owns the outcome must know what the system is allowed to do, what evidence is required, what happens when it is wrong, and how authority can be withdrawn.
The Post-Adoption Control Loop
| Layer | Executive question | Minimum evidence |
|---|---|---|
| Fit | Does the capability understand this work and context? | Workflow map, local data, exceptions, jurisdictional requirements |
| Value | What outcome improves, and against what baseline? | Outcome metric, owner, target, and benefit-realisation method |
| Economics | What does the trusted outcome cost at scale? | Model, integration, review, incident, and rework costs |
| Authority | What may the system read, recommend, or change? | Identity, permissions, thresholds, and no-go actions |
| Accountability | Who owns the result when the system participates? | Named decision owner, escalation route, and intervention rights |
| Learning | How does the organisation improve without drifting into hidden risk? | Evaluation set, change log, incident review, and stop criteria |
The control loop should be reviewed whenever the workflow, model, data, vendor, geography, or level of delegated authority changes.
A 30-Day Post-Adoption Readiness Sprint
- Days 1–5: Name the outcome. Select one AI-enabled workflow and define the business result, baseline, owner, and affected community.
- Days 6–10: Test the fit. Document local language, data quality, exceptions, cultural and jurisdictional requirements, and points where the system depends on human correction.
- Days 11–15: Build the real cost model. Include usage, integration, supervision, review, rework, security, incidents, and change-management effort.
- Days 16–22: Set the authority boundary. Specify what the system may access, recommend, trigger, and never do. Assign a named person to each consequential decision.
- Days 23–26: Run the intervention test. Pause the workflow, revoke access, route an exception, and reconstruct what happened from the available record.
- Days 27–30: Make the endurance decision. Scale, redesign, constrain, or stop. Record what the organisation has learned and what must change before the next branch.
The Bottom Line
The question is no longer, “Can we adopt AI?” It is:
Can we make AI fit our work, understand its full cost, and keep human accountability visible as its authority grows?
The organisations that answer yes will not necessarily be those with the most models or the most pilots. They will be those that can turn adoption into a durable operating capability—locally intelligent, globally aware, economically credible, and humanly accountable.
This weekly publication continues the conversation in real time. Return to deuerout.com for the next issue, updated signals, practical frameworks, and the questions transformation leaders should be asking before capability becomes dependency.
Sources and attribution
- McKinsey, AI is changing work. Now it has to change the organization, 5 October 2026: source.
- Infor, Businesses are giving AI more authority. The structures built to govern it have not caught up, 6 October 2026: source.
- Linux Foundation, 2026 State of Open Source in Europe, 7 October 2026: source.
- Linux Foundation, The 2026 State of OSPOs and Open Source Management, 6 October 2026: source.
Editorial note: McKinsey, Infor, and Linux Foundation figures are attributed to their respective studies and reports. Vendor-sponsored or organisation-sponsored findings are not presented as independent universal benchmarks.
