Labor category structures were built for a delivery world with clean role boundaries. A software engineer writes code. A data engineer moves and models data. A solution architect designs the system. A security engineer secures it. Rate cards, qualification criteria, and price evaluations all rest on the assumption that those boundaries hold.
The engineer who delivers an agentic system does all four, often within the same sprint. They design the context and retrieval architecture, which is data engineering. They implement the orchestration and tool layer, which is software engineering. They scope the agent's authority against an entitlement model, which is security architecture. They sit with the client's process owner and determine what the system should actually do, which is consulting. The role is genuinely composite, and it does not decompose neatly onto a rate card written before 2023.
Two ways this goes wrong
The first is mapping low. The capability is slotted into a senior software engineer category because the work involves writing code and that category is defensible on paper. The ceiling on that category then makes it impossible to pay market compensation for the profile, so the seat is filled with someone the category fits and the work does not. The engagement absorbs the difference as rework, schedule slip, and eventually a difficult conversation with the client.
The second is mapping high. The capability is slotted into a solution architect, subject matter expert, or principal consultant category where the rate is adequate. Two exposures follow. The candidate may fail the category's stated qualification criteria — many senior categories carry education and total-years-of-experience thresholds that a genuinely capable agentic engineer with six years of deep, current experience will not satisfy qualification thresholds on all contract vehicles. And on a competitive procurement, a high-category-heavy staffing plan invites a price evaluation problem that has nothing to do with whether the plan is correct.
Both failure modes originate in the same place: treating the labor category as fixed and the person as the variable.
What works instead
Drive the requirement, not the mapping. Where you have influence on the requirement document — and on a task order under an existing vehicle you frequently do — write the capability into the performance work statement explicitly. Evaluation design, agent authorization architecture, and inference cost engineering are describable requirements. When they appear in the requirement, the category selection follows from them defensibly rather than being reverse-engineered afterward under scrutiny.
Use the substitution provisions you already have. Most vehicles include experience-for-education substitution and related flexibility. These provisions are consistently underused, largely because contracts and delivery teams review them at proposal time and not at staffing time.
Build the crosswalk before you need it. A short, evidenced document mapping the agentic capability to the labor categories on each vehicle, with the reasoning and the qualifying evidence attached, is an hour of work in a quiet week and a week of work under a deadline. It also survives personnel turnover in a way that institutional memory does not.
Defend the rate on composition, not novelty. No contracting officer is persuaded by a premium attached to a technology term. The argument that does land is arithmetic: this staffing plan delivers the scope with fewer full-time equivalents over a shorter period of performance, with less rework and a lower defect rate reaching production. Bring the comparison.
The evidence problem sitting behind all of it
Even where the category mapping is correct, someone has to substantiate that the proposed individual qualifies — and the conventional evidence for that is a résumé narrative plus a degree, both of which are weak instruments for this profile.
The engineers who are genuinely good at agentic delivery frequently arrived from adjacent disciplines within the last few years. Their strongest evidence is what they have built and what they can demonstrate under examination, not the linear progression a qualification matrix was written to recognize. Firms that solve this build a structured, documented technical assessment and attach its results to the staffing narrative — a defensible artifact showing what was measured, against what standard, with what outcome.
That artifact does two jobs. It supports the category mapping during evaluation, and it gives the delivery organization an honest internal signal about whether the person can actually do the work. Most firms currently have neither, which is why capability claims in staffing plans have become difficult for buyers to distinguish from one another.
The commercial side has the same problem wearing different clothes
Master service agreements and negotiated rate cards produce an identical mismatch without the compliance exposure. A rate card agreed in 2022 contains no line that describes this work, and amending it mid-relationship is a commercial negotiation most account teams would rather avoid. The result is the same: the work is billed under a category that understates it, and the margin absorbs the difference quietly until someone runs the engagement profitability report.
The practical recommendation
Audit your active vehicles now, before the task order lands. For each one, identify which categories can credibly carry agentic delivery work, what the qualification criteria actually require, what substitution flexibility exists, and where the rate ceilings sit relative to current market compensation for the profile.
This is a few days of work for a contracts analyst working alongside someone who understands the technical role. It is the difference between responding to a task order in the time available and discovering, seventy-two hours before submission, that your vehicle cannot price the team you intended to propose.