Learn concepts, not tools. The durable IC investment in an era of tool churn.

STP, Kotler, via Wendell Smith, 1956. Is still live in AI-PM practice today. The segmentation-targeting-positioning framework has not changed in seventy years. The tools running segmentation have changed dozens of times. That asymmetry is the point. Frameworks endure. Tools rotate. Learn the concept that spawns the tools, not the current implementation of it.

AI makes the rotation speed faster, not slower. The half-life of tool-specific knowledge is compressing. The concept underneath stays constant.

The three-type taxonomy

Three types of practitioners: the driver uses the tool; the mechanic repairs it; the engineer knows it inside out and can create similar tools from first principles. AI commoditizes drivers. "Being technical is not about knowing a technology but using the technology". Concept-application over tool-knowledge.

The engineer survives the rotation because they can evaluate, adopt, and deprecate tools against the underlying concept without losing their footing. I have watched this play out across four AI tooling generations at companies ranging from gaming (V2 Games, 2014) to logistics SaaS (FarEye) to enterprise AI (AIonOS, 2026). The practitioner anchored to a specific orchestration framework rewrites onboarding materials every eighteen months. The one anchored to how agents reason, how context windows shape output, and how retrieval changes answer quality evaluates any new RAG implementation in an afternoon.

Concept depth at the AI PM layer

The PM skill set includes concepts that AI cannot commoditize: JTBD (what is the underlying job the user is hiring this product for?), NFR specification (what are the non-functional requirements that block enterprise compliance?), V/V/U framing (viable, valuable, usable), RWDA (real, winnable, worth-it, aligned). These frameworks predate every current AI tool. They will outlast every current AI tool.

The learning investment that pays: when AI commoditizes a tool layer, identify the concept the tool was running. ChatGPT commoditized surface-level research. It did not commoditize information architecture, synthesis judgment, or knowing which framework applies to which problem class. Those concepts are the durable layer. Double down there.

Concept fluency as the hiring signal

A PM candidate who reaches for named frameworks in technical discussions is demonstrating that concepts are internalized, not merely recalled. STP cited back to the professor who wrote it on a whiteboard in 1956. Greene's Mastery cited on deliberate practice. Frameworks deployed as the answer architecture, not as a vocabulary flex. That signal is visible to senior hiring managers and invisible on a resume that lists tool names.

The diagnostic: audit the learning diet for concept-to-tool ratio. A course that is 80% tool-specific. Shortcuts, UI flows, vendor features. Has a short shelf life. One that is 80% concept-specific. Mental models, decision frameworks, system architectures. Compounds across the career. Most candidates get the ratio wrong, and it shows in the first twenty minutes of a technical screen.

The compounding math

The frameworks in use at V2 Games in 2014 are still the PRD scaffolding at AIonOS in 2026. Zero relearning cost across twelve years. The same JTBD framing and NFR specification discipline that produced $1M ARR at FarEye in eighteen months and compressed a 60-day data onboarding process to seven days are the same tools in use today. That is the compounding case made concrete.

What the concept layer does not do: eliminate the need to stay current on tool evolution. The engineer who understands the concept still needs fluency in the current implementations. The question is which direction the investment runs. Tool fluency earns the first interview. Concept depth earns the promotion, and the role after that.