# Roles Are Temporary Bundles A product manager can carry research, analysis, design, coordination, and coding in one role, but those capabilities can move independently. In a software team, they may shift to specialists, shared platforms, external partners, or AI-enabled tools. The useful question is not whether a title survives; it is where each capability, and the judgment to direct and assess it, now lives. Next: <ref slide="2">Nine Ways Capabilities Move</ref> names the different motions more precisely. # Nine Ways Capabilities Move Roles are temporary bundles; capabilities can move in nine distinct ways. This slide maps specialization, diffusion, centralization, integration, embedding, rebundling, externalization, elimination, and loss through product-development examples and their trade-offs. Pay special attention to the final distinction: **elimination** removes work that is genuinely no longer needed, while **loss** removes work before the judgment and learning it supported exist elsewhere. The next slide follows the feedback loops these moves can create. # Changes Create Feedback Loops A capability does not move through one orderly sequence. In a software product team, specialization, diffusion, centralization, and embedding can happen together: each creates a useful outcome and a tension that changes later choices. Use this systems map to ask what is gaining depth, access, consistency, or speed—and what handoffs, inconsistency, context distance, or loss of practice may be growing alongside it. This prepares the AI discussion in <ref slide="4">AI Changes Several Variables</ref>. ## AI Changes Several Variables AI coding and analytics tools can **diffuse** usable, codified expertise to local product teams. At the same time, models, data, infrastructure, standards, and defaults may **centralize** in a shared platform. Do not infer ownership or judgment from access to an expert-looking tool. AI adoption can also create specialist work in evaluation, orchestration, reliability, governance, and exception handling. This extends the feedback-loop view from <ref slide="3">Changes Create Feedback Loops</ref> and prepares the distinction between visible output and expertise in <ref slide="5">Artifacts Are Not Judgment</ref>. ## Artifacts Are Not Judgment AI can generate polished code, dashboards, and other familiar-looking outputs. Those artifacts are useful evidence, but they do not prove that the work fits the system around it. A codebase is a cumulative record of constraints, historical decisions, and prior failures. Evaluating or extending AI output therefore requires stakeholder context, evaluation criteria, coordination awareness, and attention to edge cases. This develops the caution behind <ref slide="6">Watch for Capability Loss</ref>: when routine work disappears, organizations must preserve the judgment pathways that make oversight possible. ## Watch for Capability Loss Routine work is often a learning pathway, not merely overhead. Repetition, feedback, hard cases, mentoring, and assessment gradually build the judgment needed to review work and manage exceptions. AI **augmentation** can accelerate tasks while preserving that path. AI **substitution** can raise near-term throughput while removing the practice that produces future reviewers. When automation removes routine work, deliberately create replacement learning loops: supervised review, exception drills, mentoring, and independent assessment. This extends <ref slide="5">Artifacts Are Not Judgment</ref>: a plausible artifact is not evidence that the organization still has the capability to assess it. # Use the Framework Carefully A product workflow can look faster while its capability system becomes weaker. Use a small, concrete workflow to inspect several motions at once: expertise may diffuse, embed in tools, centralize in infrastructure, externalize to vendors, and rebundle into roles. Check three practical risks alongside the movement map: who can assess AI output, what local context the system lacks, and how people will keep developing judgment. Technical change, adoption, and organizational adaptation run on different clocks. This framework is a way to make those interactions visible, not a classification exercise. Connect this diagnostic to <ref slide="6">Watch for Capability Loss</ref>: disappearing routine work can remove the practice pathways that make reliable assessment possible.