Capability without position is invisible to buyers who would pay for it
The AI market in its current form has a structural problem that benefits practitioners who understand it: enormous demand for specific technical capability, combined with a near-total absence of clear signal about who has what. Buyers cannot efficiently find practitioners. Practitioners cannot efficiently reach buyers. The market clears through network and reputation rather than through search — which means practitioners who are not legible to search are effectively invisible regardless of their actual capability.
Most positioning failures are not capability failures. The practitioner has genuine depth. The problem is that the depth is not expressed in language that matches how buyers describe the problem they need solved. A practitioner who describes themselves as an "AI engineer with experience in LLMs and agents" is not wrong — but they are also indistinguishable from the fifty thousand other practitioners who describe themselves identically. The vocabulary failure produces the same market outcome as a capability failure: no inbound, no differentiation, no ability to command above-market rates.
Positioning is the decision about which slice of the AI stack to claim, combined with the communication of that claim in language that buyers can act on. It is not marketing. It is not branding. It is a strategic decision about where to concentrate depth so that the depth becomes a surface — something that a buyer can find, evaluate, and select based on the evidence the practitioner has produced. The decision precedes the evidence. The evidence validates the decision. Both are required.
This page addresses the positioning decision first and the communication problem second. The order matters. Most practitioners attempt to solve the communication problem without first solving the positioning problem — they try to make a broad, generic capability legible rather than narrowing first and communicating second. Legible genericism is not a position. It is a description of a commodity.
Three positioning failures that feel like capability problems
The three most common positioning failures are structural — they are produced by reasonable decisions that are wrong for reasons that are not obvious until the market response makes them visible.
The generic AI practitioner
The practitioner claims the full stack — models, orchestration, applications, evaluation, infrastructure — and describes themselves in terms that apply to the entire field. This feels safe because it does not exclude any buyer. The market effect is the opposite: it signals to every buyer that the practitioner is not the specialist their problem requires. Buyers who need a specialist in evaluation-backed agent pipelines do not select a generic AI practitioner — they continue searching until they find someone who specifically describes what they need. The generic practitioner is competing with every other generic practitioner for the buyers who could not find a specialist, which are the buyers with the least budget and the least defined problems.
Vocabulary mismatch
The practitioner has genuine depth but describes it in technical language that buyers do not use. The practitioner says 'cross-encoder reranking and hybrid retrieval architectures.' The buyer says 'our AI search isn't finding the right answers.' Both are describing the same problem and the same solution — but the practitioner's description does not match the buyer's search query, their vocabulary in meetings, or their framing of the budget request. The practitioner loses the opportunity not to a more capable competitor but to a less capable practitioner who uses the buyer's language. This is the most common failure mode for technically strong practitioners.
Premature adjacency
The practitioner establishes a position in one area and then, before the position has produced returns, begins developing depth in an adjacent area — driven by interest, by a client request, or by the perception that the first area is too narrow. The result is two shallow positions rather than one deep one. Neither area has the proof-of-work depth required to generate the inbound that makes the position work. The adjacency move is not wrong in principle — it is wrong in timing. The second area of depth should be chosen after the first position has generated enough gravity that the practitioner can afford to invest in developing depth in a new area. Adjacency before that point disperses the investment that would have compounded.
Depth × domain — the intersection that produces a surface worth owning
A defensible position has two properties: it is narrow enough that the practitioner can develop genuine depth in it, and it is durable enough that the depth does not expire with the current model generation. In a field where the underlying technology is evolving as rapidly as AI, durability is the harder constraint to satisfy.
Positions built on specific model capabilities are fragile — the capability that made the position valuable may be automated or commoditized within one model generation. Positions built on architectural patterns and operational problems are more durable, because those problems persist even as the underlying models improve. Evaluation, observability, deployment reliability, and domain application are structurally durable — they become more important, not less, as the systems they govern grow more capable. The Technical Readiness framework uses this same durability assessment — the practitioner who can own the operational layer of an AI system is positioned above the layer that models will automate first.
The depth requirement is calibrated by the buyer, not by the practitioner. The relevant question is not "how deep am I" but "am I deeper than the buyer can go without me?" A practitioner who is the second-best person in the world at RAG retrieval architecture for healthcare data is more valuable to healthcare buyers than the best generalist AI practitioner in the world, because the generalist cannot substitute for the domain depth that the specialist brings.
The Value Engineering framework provides the decision logic: filter for positions where the gap between what the practitioner can do and what the buyer can do without them is largest. That gap is the value delivered. The positioning decision is a bet on where that gap will be largest and most durable over the practitioner's planning horizon.
Where to position — layer-by-layer analysis
The AI stack has six layers, each with different positioning properties. The diagram shows how the three practitioner position types — generic, narrow-deep, and adjacent-domain — occupy the space differently. The defensible positions are always narrow and deep, never wide and shallow.
| Layer | Code | Positioning Notes | Defensibility | Entry |
|---|---|---|---|---|
| Domain | L6 | The vertical where AI is applied — healthcare, finance, legal, logistics. Positioning here requires both technical depth and domain fluency. The hardest to enter, the most defensible once held. Buyers in domain verticals pay the highest rates because the practitioner cannot be replaced by a generalist. | Highest | High |
| Applications | L5 | Product-level AI systems — chatbots, copilots, agents, automation pipelines. Broad market, high demand, but also the most competitive layer. Positioning here requires a specific application type or integration surface — not 'I build AI applications' but 'I build evaluation-backed agent pipelines for regulated industries.' | Medium | Low |
| Evaluation | L4 | The quality and reliability layer — evals, regression detection, deployment gates. Chronically undersupplied. Practitioners who understand both the technical implementation and the conceptual problems (what 'failure' means for a non-deterministic system) occupy a position that few can credibly fill. | High | Medium |
| Orchestration | L3 | Multi-step agent workflows, tool use, memory systems, retrieval pipelines. High demand, moderate supply. The practitioner who has built and debugged production orchestration systems — not just connected an API — has real signal here. The failure mode is claiming this layer without the operational record to support it. | Medium-High | Medium |
| Models | L2 | Fine-tuning, RLHF, alignment, pretraining. Requires significant compute access and research-level understanding. Defensible for practitioners who have the background — inaccessible for those who do not. Most independent practitioners should not claim this layer without the receipts. | High | Very High |
| Infrastructure | L1 | Serving infrastructure, GPU clusters, inference optimization, deployment systems. Requires deep systems engineering background. Highly defensible, highly specialized. Largely orthogonal to the practitioner who is building on top of existing model providers. | Very High | Very High |
The table reveals a consistent pattern: the layers with the highest defensibility also have the highest entry bar. This is not a coincidence — it is the mechanism that makes them defensible. A position that anyone can claim is not a position. The entry cost is the moat. For most independent AI practitioners, the highest-leverage positioning target is the intersection of Evaluation (L4) or Orchestration (L3) with a specific Domain (L6) — high defensibility, meaningful but achievable entry, and durable demand that increases as AI systems become more capable and more consequential.
Translating technical depth into language buyers can act on
Legibility is a separate problem from positioning. The positioning decision is about which surface to own. Legibility is about whether a buyer who encounters the practitioner's public artifacts can understand the value of that surface in terms that connect to their own problem. The failure mode is assuming these are the same problem and solving both with the same vocabulary.
Technical vocabulary is necessary for the artifacts — the build logs, the framework documents, the research investigations — that constitute the proof of work. A buyer evaluating whether the practitioner has the depth they need will read those artifacts and the technical precision is the signal. But the entry point — the positioning statement, the one-sentence description of what the practitioner does, the language on the first page a buyer sees — must be in the buyer's vocabulary, not the practitioner's.
The buyer's vocabulary is the problem statement, not the solution description. A buyer does not search for "hybrid BM25 and vector retrieval with cross-encoder reranking." They search for "AI search that returns wrong answers" or "RAG system not finding relevant documents." The practitioner who positions as the person who solves that problem — in those terms — is findable. The practitioner who positions as the person who implements the technical solution is not findable by the buyer who has not yet identified the technical solution as the answer.
The translation is not simplification. The technical depth remains — it is the evidence in the artifact, the substance of the engagement, the actual value delivered. What changes is the entry point: the problem language that makes a buyer recognize themselves as a potential client before they understand the technical method. Both layers are required. Neither replaces the other.
The communication layer — how to construct the narrative that connects problem language to technical depth — is documented separately in the Technical Storytelling page, which is currently in development. Legibility is the goal; storytelling is the mechanism.
Choosing the second area of depth from the first
Every practitioner eventually develops more than one area of depth. The question is not whether to develop a second area — it is when, and which one. Both questions have answers that compound the primary position or dilute it.
The timing answer is: after the primary position has generated inbound that the practitioner cannot fully service. At that point, the primary position is producing returns, and the investment in a secondary area comes from capacity that the primary position has created rather than from capacity that should be deepening the primary position. Before that point, adjacency investment is premature — it disperses depth before depth has compounded.
The selection answer is: the secondary area should be chosen because it makes the primary area more valuable, not because it is independently interesting. There are three categories of useful adjacency. The first is the layer above or below in the stack — a practitioner positioned at Orchestration (L3) develops depth in Evaluation (L4) because every orchestration client eventually needs evaluation, and the practitioner who can provide both is more valuable than two separate practitioners. The second is the domain vertical — a practitioner positioned at a technical layer develops depth in a specific domain, producing the domain-layer defensibility that the stack diagram describes as highest-value. The third is the operational layer — a practitioner who understands how to build systems develops depth in how to operate and observe them, because operational credibility is what converts technical knowledge into trusted ownership.
The Researcher's Path framework describes the underlying dynamic: depth compounds when the second area reinforces the first. A practitioner with two independent areas of depth is not twice as valuable as a practitioner with one — they may be less valuable, because the cognitive overhead of maintaining two independent positions dilutes both. A practitioner with two adjacent, mutually reinforcing areas of depth is not twice as valuable as one — they may be five times as valuable to the specific buyer who needs both, which is the buyer worth attracting.
Four-dimension diagnostic
Four diagnostic dimensions. Each addresses a different failure mode in the positioning system. A strong position passes all four. A weak position fails one or more — and the dimension it fails identifies the specific intervention required.
A specific position is one that a buyer can find when they are actively looking for it. 'AI practitioner' is not searchable — it describes half the people publishing online. 'RAG pipeline architect for regulated data environments' is searchable. The test is whether a buyer who needs exactly what you do can find you by searching for what they need.
- 01Can you state your position in under twelve words without using 'AI,' 'machine learning,' or 'data science' as the primary descriptor?
- 02If a buyer searched for the specific problem you solve, would your artifacts surface in the results?
- 03Can you name three practitioners in the world who occupy the same position — and distinguish yourself from each of them specifically?
- 04Does your position reference a specific layer of the stack, a specific application type, or a specific domain vertical — not all three vaguely?
No to any: the position is too broad. The next move is to narrow — not to a different area, but deeper into the current one. Width feels safe. Depth is what buyers pay for.
A position without evidence is a claim. Claims are evaluated against the prior probability that the practitioner is telling the truth — which for an unknown practitioner is low. Evidence changes the evaluation. The evidence required is public proof of work in exactly the position being claimed, not adjacent to it.
- 01Do you have at least one public artifact — a build log, a published system, a framework document — that demonstrates capability at the layer you are claiming?
- 02Is the artifact specific to the position, or is it general enough that it could have been produced by someone in any position?
- 03Has the artifact been encountered by buyers in the context of the problem it addresses — not just promoted by you in your own channels?
- 04Does the artifact demonstrate operational experience, not just conceptual understanding?
No to any: the position is not yet credible. The next move is artifact production before positioning communication. Communicating a position without evidence degrades the position — buyers who encounter an unsubstantiated claim and then find no supporting artifacts draw conclusions that are hard to reverse.
Legibility is the gap between the practitioner's technical description of their work and the buyer's vocabulary for the problem they are trying to solve. The gap is almost always the practitioner's responsibility to close — buyers do not translate practitioner language into their own vocabulary, they move on to a practitioner whose language is already legible.
- 01If a buyer in your target market read your three most recent public artifacts, could they state in one sentence what problem you solve and for whom?
- 02Does your public language use the vocabulary buyers use when they describe the problem — even when that vocabulary is less precise than your preferred technical language?
- 03Is there a gap between what you call what you do and what buyers call the problem they have?
- 04Could a non-technical buyer in your target vertical understand the value delivered, even if they cannot evaluate the technical method?
No to any: the position exists but is not legible. The next move is vocabulary alignment — not dumbing down, but translating. The technical precision belongs in the artifact. The value language belongs in the positioning statement. These are different documents.
Every practitioner eventually develops more than one area of depth. The question is whether the second area was chosen strategically — because it compounds the value of the first — or incidentally, because it was interesting. Strategic adjacency produces a position that is defensible on two dimensions simultaneously. Incidental adjacency produces a practitioner who is mediocre at two things rather than excellent at one.
- 01Can you articulate why your second area of depth makes the first more valuable — specifically, not in general?
- 02Does the combination of your two areas of depth produce capability that neither area produces independently?
- 03Are the buyers who need your primary area of depth likely to also need your secondary area — or are they different buyers?
- 04Is the secondary area chosen because it strengthens the position, or because it was the next available learning opportunity?
No to any: the adjacency is incidental rather than strategic. This is not urgent — incidental adjacency is better than no second area. The move is to examine whether the current secondary area can be reframed to serve the primary position, or whether a different secondary area would compound more effectively.
Position early, compound long
The positioning decision is one of the highest-leverage decisions a technical practitioner makes, and it is almost universally deferred. The practitioner takes available work rather than positioned work. The available work builds capability in whatever direction clients pull. The capability accumulates without concentrating. Years pass before the practitioner recognizes that they have depth in several areas and a position in none.
The cost of this deferral is asymmetric. The practitioner who positions early pays a short-term cost: they decline some work that falls outside the position, they invest in proof of work that takes months to generate returns, they commit to a direction before the market has fully validated it. The practitioner who defers pays a compounding cost: every year of undirected depth accumulation is a year that could have been spent deepening a position that would have been generating inbound by year three.
The interaction with the proof-of-work layer is direct. The Proof-of-Work framework describes how artifacts compound — but only if they are produced within a consistent position. Artifacts produced across multiple unrelated positions do not compound; they demonstrate range without demonstrating depth. The positioning decision determines whether the artifact production effort compounds or disperses. It is the multiplier on all downstream investment.
The practitioner who has chosen a position, built proof of work in it, and communicated it in buyer language is in a structurally different situation than the practitioner who has not — not because they are more capable, but because their capability is legible, findable, and concentrated in a surface that buyers can select. That legibility is worth more, economically, than an equivalent amount of additional technical depth in a position that no buyer can find.
END · CRR/01
Continue Reading