文字起こし 🔗
AOM News2: Welcome to the latest edition of the Agentic Operating Model newsletter. This week, I found evidence of a shift in attention from what AI agents can do to how organizations can make their work reliable, accountable, and valuable. Is this a result of shifting opinions and global events? US midterm election results may well result in a greater focus in the future on the work implications of AI. Rather than this year's focus on its generative output. We found three articles worth your attention that offer interesting perspectives. 1. IBM examines enterprise learning and value creation. 2. DXC focuses on workforce transformation. And 3. Forester addresses enforceable controls in software development. Together, they suggest that implementation capability, not autonomy alone, Is becoming the central management challenge. The main article is from IBM: Turning Operational Experience into Enterprise Value. Published in November, IBM Institute for Business Values Research with Microsoft proposes an agentic AI value model spanning six connected capabilities: ecosystem strategy and design, governance, controls. Work design and workforce readiness, execution, and an applied learning mechanism. Its most distinctive contribution is that the report emphasizes feedback loops that convert operational experience into improvements. Organizations observe agent performance, evaluate outcomes, and use the resulting lessons to refine systems, shared knowledge, and working practices. The strategic proposition is that access to similar technology need not produce similar results. The ability to learn from using it may become a source of differentiation. The point identifying, sustaining, and growing agentic value requires a capability many leaders may be overlooking, stood out as crucial to an agentic operating model, and I was surprised by the further finding from those surveyed. About how most employees regard the technology as only 25% pointing to transformation, new ways of working, as opposed to 41% of survey respondents identified efficiency as AI's primary benefit. Still some work to do to push value creation as a key driver in the agentic operating model strategy. They are just survey-based associations, not proof that adopting the framework causes superior performance. The public summary does not provide enough methodological detail to fully assess selection effects. It is a shame there isn't a worked example and case study with a real company. The second article is a commercial piece from the DXC website, Designing the Human Agent Workforce. DXC's November article, Agentic AI Workforce Transformation, focuses on managing work performed jointly by employees and agents. Its recommendations place organizational responsibility alongside technical capability. Assign a named human owner to each agent supported workflow, define decision rights, establish escalation paths, and govern agent identities throughout their life cycle. Agents can begin in shadow operation, move to execution subject to human approval, and eventually undertake bounded automation. This approach allows organizations to test performance and controls before widening permissions. It also avoids treating the removal of human involvement as the default measure of success. Many of the same old messages are repeated, but it is worth reading for an insight into their methodology. Key points DXC's survey of 2496 decision makers found 86% will use AI to shift job responsibilities, 89% will create new roles to manage AI systems, and 87% are retraining staff. Workforce readiness for human agent collaboration starts with AI fluency, meaning people know how to supervise. Correct and orchestrate agents. Exponential is the method or blueprint developed by DXC which may well warrant further research, and there are a few case study examples on the method's website. The final article is a short piece from Forrester Making Controls Enforceable. This introduces its Agentic Development Security Tools Q4 2026 evaluation. Covering 12 providers against 29 criteria. The focus extends beyond inspecting generated code to governing what coding agents actually do, invoking tools, choosing dependencies, accessing resources, and undertaking remediation. It follows on from a post earlier in the year on Agentic Development Security why AppSec needs a new operating model. Three distinctions are relevant to operating model design. An organization must distinguish between telling an AI agent what it should do and having mechanisms that actually control what it can do. For example, security instructions are not equivalent to enforced restrictions. A proposed fix is not equivalent to validated remediation. And An inventory of available tools is not equivalent to the ability to stop a risky action before execution. And now the newsletter spotlight. A look back at how the OECD report is being interpreted. Public responses to the OECD's agentic AI in organizations, early insights from practitioner interviews, have concentrated on practical implications. Rather than methodological criticism. Three examples illustrate how its findings are reaching legal, workforce, and policy audiences. One Evershed Sutherland Accountability Across Boundaries The law firm's October Global AI regulatory update highlights the movement from experimentation into organizational workflows while governance remains immature. Its discussion draws attention to evaluation and assurance, cybersecurity, and accountability across organizational boundaries and interactions between agents. The operational implications include human approval for high stakes actions such as payments, restrictions on permissions, and registries of approved agents. For leaders, the useful question is not simply whether someone owns AI. But whether responsibility and traceability remain clear when several agents and systems contribute to an outcome. 2. European Commission Platform, Capability Without Dependency. The Digital Skills and Jobs Platform summary, updated in October, emphasizes structured tasks, verifiable outcomes, and gradual increases in autonomy. Its distinctive contribution is the workforce lens. Agents can support learning, but excessive reliance may erode expertise, critical thinking, and learning opportunities. Training therefore needs to preserve judgment while improving tool use. Finally, the emerging themes from research this week: Bounded autonomy. Expand agent permissions gradually, matching freedom to task structure, risk, and verifiability. This is from the DXC report. Accountability. Keep human ownership and traceability clear across agent interactions and organizational boundaries. This is also from the DXC report. Workforce capability. Develop AI skills without weakening professional judgment or opportunities to learn. This is from the European Commission report. Governance maturity. Replace policy statements alone. With enforceable controls, evaluation, and operational feedback. This is from the IBM article. Thank you for listening to the Agentic Operating Model Newsletter Podcast. We will include all links to the articles and reports mentioned in this podcast in the description, as well as a link to the newsletter on LinkedIn.