Capability Overhang

Highlights

Top Insights

AI is already powerful enough to trigger a transformation comparable to the computer or steam engine. Even if technical progress slowed tomorrow, today’s systems could still fundamentally change work, education, and business.

The gap between what AI can do and what organizations actually do with it may be becoming a more important competitive variable than access to the technology itself.

Source: The Next Phase of AI (Knowledge at Wharton)

Top News

1. Kimi introduced K3.1 Pro, its most powerful open-weight AI model yet.
2. Anthropic opened a U.S. preview of Claude Duet, a proactive cloud agent that monitors connected apps。
3. Google launched a Gemini business hub with brand, template, and connected-app tools tailored to small-business workflows.
4. Zoom announced a Unified Context Layer, APIs, MCP servers, workflows, and agent teams.
5. OpenAI added spreadsheet, presentation, and Word-document creation skills to hosted Agents API sandboxes.
6. Google Research demonstrated closed-loop AI agents that directed robotic experiments and improved confirmed aptamer hit rates across three design-test cycles.

Additional Insights

1. The Industrialization of Expertise: How AI Creates Value by Making Expertise Abundant (California Management Review)
AI’s greatest economic impact is not simply automating knowledge work, but transforming scarce, expensive human expertise into an abundant, scalable, and personalized resource. Historically, expertise in fields such as healthcare, education, law, and consulting has been constrained by professionals’ limited time and availability, forcing organizations to trade personalization for scale. Generative AI fundamentally changes this equation by delivering tailored knowledge and analysis to millions of users at relatively low marginal cost. The article identifies four mechanisms of value creation: expanded access to previously unaffordable expertise, mass personalization of services, augmentation of human professionals, and market expansion through new demand for accessible knowledge-based services. This shift could democratize expertise, empowering individuals, smaller businesses, and underserved populations without necessarily replacing human professionals. Crucially, as AI makes codified knowledge increasingly abundant, competitive advantage shifts from possessing expertise to effectively applying it through scarce complementary assets, including proprietary data, contextual judgment, trusted relationships, institutional credibility, and governance. The central strategic implication is that future winners will not necessarily be organizations with the most expertise, but those best able to combine widely available AI capabilities with uniquely human and organizational strengths to produce differentiated outcomes.

2. Blind Overseers: AI and the Erosion of Judgment (California Management Review)
AI dramatically increases productivity and lowers barriers to entrepreneurship, it also creates a widening “oversight gap” between what people can produce with AI and what they can independently understand, evaluate, or challenge. This erosion affects two critical capabilities: evaluative judgment, the ability to verify AI-generated outputs and identify errors, and generative judgment, the ability to imagine alternatives and pursue unconventional ideas. The resulting risks are twofold: “cognitive debt,” as employees increasingly rely on AI without retaining the expertise needed to assess its work, and diminished innovation, as AI-generated solutions converge toward similar, conventional approaches, reducing creative diversity and overlooking breakthrough opportunities. These risks are amplified by opaque AI systems, automation, competitive pressure, and dependence on a handful of technology providers. Importantly, experienced professionals who actively question and direct AI can strengthen their capabilities, while less-skilled users risk becoming increasingly dependent. To counter these threats, leaders should periodically test employees’ independent understanding of AI-assisted work, preserve human oversight, reward critical evaluation, deliberately use AI to explore unconventional alternatives, and diversify technology dependencies.

3. How to Level Up Systematic Innovation with AI (MIT Sloan School of Management)
Organizations can leverage artificial intelligence to make innovation more systematic, collaborative, scalable, and repeatable while maintaining human control over critical decisions. He outlines four key principles: (1) Optimize processes before introducing AI by mapping existing workflows, identifying inefficiencies, and breaking complex activities into smaller, clearly defined tasks; (2) Keep humans in charge, adopting an “AI in the loop” approach where AI augments human creativity and judgment rather than replacing them; (3) Master prompting and instruction design by providing clear objectives, contextual details, and examples that guide AI toward relevant and innovative solutions; and (4) Embed rigorous verification and validation to identify errors, evaluate AI-generated ideas, and prioritize the most promising opportunities. Eppinger recommends starting with simple AI-assisted experiments, refining instructions through iteration, and automating workflows only after they consistently produce reliable results.

4. Rewiring R&D for the agentic age (McKinsey)
The article highlights the transformative potential of agentic AI in research and development, estimating that it could generate $360–560 billion in annual economic value, accelerate complex product development by 20–80%, and deliver productivity gains equivalent to 10–15% of R&D spending. However, a significant gap remains between AI adoption and measurable business impact: while 77% of surveyed R&D leaders report improvements from AI, only 13% of identified use cases have scaled beyond proof of concept. The primary obstacles are poor data availability and quality (56%), difficulties integrating AI into engineering workflows (50%), and outdated organizational processes. McKinsey identifies five priorities for unlocking AI’s potential: (1) establish a quantified vision for AI-native product development, (2) scale successful pilots through standardized platforms and integration, (3) redesign operating models and workflows rather than simply adding AI tools, (4) invest in robust, reusable engineering data infrastructure, and (5) redefine engineers’ and scientists’ roles toward AI orchestration, critical judgment, and validation.

Innovation Radar