Sid Pednekar: Judgment Is the Scarce Resource in AI Adoption

I am Sid Pednekar, an AI, technology and data executive in asset management. I want to explain what organizations that lead in AI adoption do differently, using two hypothetical firms and their research workflows as a running example.

AI access is widespread but outcomes are uneven. I cite surveys reporting that some companies see productivity improvements while fewer report measurable cost reductions or revenue gains. These figures depend on the surveys and their definitions; the broader point is that deployment does not automatically create business value.

I organize the discussion around three pillars: design the right problem, build shared capabilities, and compound the advantage. They apply to individuals and teams across functions, not only to one industry.

As AI makes many forms of analysis and answers more abundant, more output does not necessarily mean more value. The scarce resource is judgment: choosing what matters, recognizing quality in context, and weighing tradeoffs when there is no single obvious answer.

Judgment is exercised throughout an organization, from executives and engineers to portfolio managers and data scientists. Because it is limited, the useful question is where our best judgment is worth spending.

The first pillar is choosing the right problem. It is tempting to start with a new model or copilot and ask where it can be inserted. I suggest starting instead with the business bottleneck and what would improve the decision.

In my hypothetical firm B, analysts summarize earnings transcripts faster with an AI tool. That may save time, but summarization can also discard context. Firm A asks why important signals reach analysts too late and reexamines the entire workflow around that bottleneck.

The relevant bottleneck could be research capacity, decision speed, data quality, latency or reliability. Leaders, in my account, are selective about which decisions deserve attention rather than treating every possible AI feature as equally valuable.

Experimentation is useful, but it is different from innovation theater or productivity theater. A firm with disconnected summary, copilot and retrieval pilots can appear active without improving an important outcome. I advocate concentrating engineering, analytics and human judgment on a small number of valuable workflows.

Redesigning work also means changing roles. Rather than merely accelerating a report, an analyst could define an investment thesis, monitor evidence for and against it, investigate exceptions and spend time on the cases most likely to affect a decision. The distinction is between speeding up a task and reconsidering the system around it.

Once the problem is selected, the second pillar is shared capability. Technology and business roles are converging: engineers move closer to business outcomes, while business specialists build tools and automation. Bringing the right people together requires crossing organizational silos.

I call this standardizing the core and personalizing the edges. Shared infrastructure and modules are familiar ideas, but AI changes what can be shared across analysts, clients, engineers and data scientists while still allowing each group to adapt its own workflow.

For data scientists, a shared core might include datasets, semantic definitions and a record of what worked, what failed and where the gaps are. The next person should be able to build on that work instead of rebuilding it from scratch. Ask what a successor should not have to reconstruct.

Enterprises were largely designed around people using individual applications. Agents can move across applications and workflows, so leaders need to understand where infrastructure and data quality must change. Reusable data is a common challenge.

Reusable knowledge can support different kinds of team agents: subject-matter specialists, workflow agents that coordinate tasks, or agents that track organizational state. Their value depends on the shared knowledge and processes available to them.

The final pillar is compounding. When an initiative fails, what can the organization learn? When it succeeds, how does that become part of the system? An individual may learn from a correction, but the organization benefits only if useful evidence and learning are fed back into tools, context or process.

AI can free capacity, but leaders still have to decide where that capacity goes. Redesign the system around important measures, monitor the results and build a feedback loop. My closing question is the same one I started with: when everyone has AI, who has the advantage?