Breadth as differentiation. Depth as the survival constraint.

In April 2024, the same week I earned a peer-voted LinkedIn Top Voice PM badge, I published a post naming the failure mode that badge could easily represent: the Human-GPT. The credential and the critique arrived together, which felt right. Breadth-as-credential and breadth-as-commodity are separated by a single axis, and the axis is depth.

The Human-GPT failure mode is not that generalists are bad. It is that the credential gap between a broad generalist and a capable chat interface has collapsed. Breadth without a depth-axis is commodity. A hiring committee in 2026 gets generalist range for free. What they cannot get free is the operator with domain depth, production scars, and pattern recognition that breadth accelerates but cannot replace. The thesis is not breadth or depth. It is both, in that order of permanence.

The Human-GPT problem, named

Breadth without depth is now table stakes. And table stakes have no margin.

A Human-GPT is a person whose value proposition is breadth, availability, and general competence: the same things a chat interface provides for free. The Human-GPT post made five distinct moves. First: name the failure mode. Second: use ChatGPT itself as the evidence. Third: split depth into two valid axes. Vertical (industry or domain expertise) and horizontal (functional or specialization expertise). Fourth: permit pivots, as much as needed, but move quickly. Fifth: close with the observation that even the technology exposing the problem is moving toward domain-specific models. Generalization is the direction of commoditization.

The depth-axis is a free variable. Vertical depth (logistics, healthcare, fintech), horizontal depth (AI product management, voice AI infrastructure, LLM routing architecture). The choice is not optional. Standing still in pure breadth while treating the niche choice as deferred is how the Human-GPT failure mode becomes permanent.

The formation arc: 2018 to 2025

The diagnosis preceded ChatGPT by four years.

In March 2018, evaluating a team member at V2 Games, I saw it clearly in operational form: she was mostly the office generalist with an occasional bookkeeping assignment. A path to step up into project management was offered. She was not ready. The framing was operational, not theoretical. A generalist without a step toward depth is a role-gap, not a role.

By December 2019, the seed had a framework: the driver, mechanic, and engineer ladder. "A driver is adept in using a tool. A mechanic can repair it. An engineer knows it inside-out and can create similar tools from first principles. The world does not need more drivers. AI will make you redundant if not now then soon. Strive to learn concepts, not tools." The AI-will-make-you-redundant line was written in 2019. The prediction has aged into literal accuracy. The drivers are being automated, the engineers are not.

The label arrived in April 2024 and was applied immediately as a decision procedure. Inside-out: strong in product management, data, and applied AI. Outside-in: enterprise demand for AI PMs, LLM-based solutions, and production-grade agentic systems. Inside-out crossed with outside-in produced the depth-pick. The operating principle stopped being a published stance and became a navigation tool.

By July 2025, the arc extended further than expected: "We humans are built this way: to deeply feel, to be affected, and still have the resilience to bounce back. And no AI will ever take that away. This is both our greatest vulnerability and our greatest strength." The depth-axis extended to humanness itself as a professional differentiator, EQ, emotional precision, connection, imperfection as signal rather than liability. A September 2024 independent confirmation: "In the future, if you know how to use AI you will be efficient, but if you know how to connect with humans you will be loved." Knowledge becomes a utility. The depth that stays durable is being human.

Three operating constraints

The thesis produces three enforced decisions, not recommendations.

Choose a niche and invest. The axis is free. The choice is not. Every failed attempt adds a layer of depth. The point is not to lock in forever. The point is not to stand still in pure breadth while treating the niche choice as optional. Pick the intersection of demonstrated strength and market pull. Apply the inside-out, outside-in filter. Pick.

Pivot if wrong, but move quickly. Speed of pivot is the constraint. Slow pivots burn the sunk-cost trap. Fast pivots add domain exposure without destroying accumulated depth on the previous axis. Every failed depth-pick that exits cleanly still outperforms standing still.

Extend the depth-axis beyond technical expertise. Breadth keeps you efficient. Depth makes you valued at the level that cannot be commoditized. The April 2024 prescription opened with EQ before niche. Technical depth on an AI PM axis is the floor. The ceiling is the humanness-as-depth argument: the operator who can hold emotional precision, navigate institutional ambiguity, and form genuine connection in enterprise deals is not replicable at any price point by a model that optimizes for token prediction.

What changes in the AI era

The AI era raised the bar on depth. It did not change the direction of the thesis.

Two effects run simultaneously. First: range becomes more accessible, which compresses the premium on breadth alone. Any operator with a capable model now has access to broad pattern recognition across domains they have never worked in. The breadth premium erodes. Second: depth becomes more productive, which raises the ceiling on the IC who has it. An operator with genuine domain depth plus AI agents produces output closer to a small team than ever before.

At AIonOS, the AI PM role involved architecting a product from scratch against a voice AI production scale of 4M+ calls per year. That was an IC contribution with team-level throughput. Fifteen-plus enterprise POCs scaled from zero to $1.5M+ pipeline on that foundation. The breadth-as-differentiation thesis in AI-era form is not "be broad." It is "be deep enough that the model amplifies your judgment rather than replacing it." The operator who evaluated the inside-out versus outside-in depth-pick correctly and routed their career accordingly is the one who compounds. The one who deferred the choice deprecated their own leverage.

The compound configuration

Breadth without depth is commodity. Depth without breadth is a specialist trap that misses cross-domain signal. The compound is the only configuration that survives the AI-era talent market: domain depth that the operator chose deliberately, accelerated by breadth that speeds pattern recognition and sharpens the depth-pick over time.

The Human-GPT failure mode is avoidable. It requires one decision: pick the niche, invest, and refuse to let the breadth credential substitute for the depth work. The operator who gets the dependency direction right builds compounding leverage. The one who does not builds a profile that a chat interface can replicate for free.

Forward question: as models push further into domain-specific territory, the depth-axis for a human operator becomes less about domain knowledge per se and more about judgment and taste. Knowing which framework applies when, not just knowing the frameworks. What does deliberate investment in judgment look like, as a practice, when the knowledge layer is increasingly commoditized? If you're working through that question and have a working answer, I'd like to hear it.