Job Posting Autopsy — The Org Chart Edition: What a Properly Staffed AI Engineering and Applied Science Org Actually Looks Like

Series: Weekly Data Science Job Posting Review

The last several entries in this series have been autopsies — diagnosing job postings that compress five or six engineering disciplines into a single mid-level hire. The pattern is consistent enough across industries that it’s worth flipping the exercise around. Instead of pointing at one broken posting, this piece builds the thing that should exist instead: a properly leveled AI organization at a large retailer — both the AI Engineering ladder and the parallel Applied/Research Science ladder it partners with — written in the same format the postings themselves use — position summary, essential functions, required tools, experience bar, independent judgment, and compensation.

This isn’t about any single company. It’s a reference architecture, built by cross-referencing how AI engineering actually gets leveled at large, mature organizations — Target, Walmart, and Microsoft all publish enough detail in live postings and compensation data to reverse-engineer a real pattern. The goal is to give hiring managers and candidates alike a default to check a posting against, instead of inventing the scope from scratch every time a role opens.

A Note on Experience Requirements: Most of This Tooling Doesn’t Have a 7-Year History

One thing worth naming before any of the levels below: the specific tools and disciplines this ladder is built around — production RAG pipelines, agentic orchestration frameworks like LangGraph and AutoGen, PromptOps as a named practice, LLM-specific observability and eval tooling — are mostly two to three years old as disciplines with any real production track record. ChatGPT’s public launch was late 2022. RAG as a standard enterprise pattern, agent frameworks with real adoption, and dedicated eval tooling like Ragas or DeepEval are newer still, mostly 2023 and later.

That matters directly for how the experience requirements in this ladder should be read. “5+ years in production LLM/RAG systems” and “7+ years in agentic orchestration” are not physically obtainable as stated — nobody has seven years of agentic-systems experience in a field that’s existed for two. What these requirements actually mean in practice, and how they should be evaluated in a screen, is total engineering experience (which can absolutely run 7–10+ years) combined with however much of the last two to three years that person spent specifically on LLM/RAG/agentic work. A Staff-level candidate with 9 years of general software and ML engineering experience, the last 2.5 of which were deep, hands-on production LLM work, is a perfectly legitimate Staff candidate — arguably a stronger one than someone who has nominally touched these tools for “5 years” by stretching the definition back to 2020-era pre-ChatGPT NLP work that doesn’t actually transfer.

The compensation bands and experience bars below should be read with that adjustment built in: weight the recency and depth of the AI-specific work heavily, and treat total years of experience as the secondary signal, not the primary one. A posting that demands a high number on both simultaneously — many years of general experience and many years specifically in tooling that’s barely three years old — is asking for a candidate profile that mostly doesn’t exist yet at scale, which is itself one of the more common and least examined scoping errors in postings like the ones this series autopsies.

A Note Before the Ladder: Who Should Own What

One question worth answering before listing roles: does a retailer this size need to hire Data Engineering, MLOps, and AI Governance as part of the AI org, or does that work come from elsewhere? And separately — where does research and applied science fit relative to engineering?

At companies of meaningful scale, the honest answer for the first question is mostly elsewhere. Data Engineering typically serves the entire enterprise — finance, merchandising, supply chain, and AI alike — not just the AI team, and stands up as its own org with its own ladder. AI Governance is most often a Legal/Compliance function with an AI-specific liaison, not a headcount line inside engineering. MLOps/Platform is the one exception that genuinely varies: at some companies it’s a shared Platform Engineering function; at companies early in AI maturity, it’s frequently absorbed into the AI engineering team by default, simply because no dedicated platform team exists yet to hand it to.

Target’s own job board is a useful real-world check on this. Its Brooklyn Park, MN office simultaneously lists a Principal AI Engineer, a Senior Engineering Manager for Advanced AI, a Lead Machine Learning Engineer focused specifically on ML Ops, a Senior Applied Data Scientist for Search and Browse, a Principal Data Scientist for Advanced Machine Learning, and a separate Director of Enterprise Data Governance — six distinct roles across at least four different ladders, not one bundled hire. That’s the structural pattern this piece is built around: AI Engineering owns the IC ladder below, and partners with — rather than absorbs — Data Engineering and Governance as shared functions. MLOps is included as a partner function below for the same reason: it’s the one most often missing at companies early in their AI build-out, which is exactly when it gets dumped onto the AI Engineer by default.

The second question — where research and applied science fit — is worth getting precisely right rather than assuming. The relationship is not a hierarchy where research hands off a finding and engineering takes it from there. It’s closer to integrated co-development. A recent academic survey of AI organizational roles describes Research Scientist, Research Engineer, Applied Scientist, and ML Engineer as existing on a continuum with bidirectional connections and porous boundaries — not as departments separated by a handoff wall. Target’s own postings confirm this concretely: its Applied and Senior Data Scientists are described as collaborating with AI/ML engineers on peer teams to build and productionize generative AI solutions together — not “we research, they ship.” Target’s own Principal Data Scientist posting goes further, listing “adopt modular architecture and good software development/engineering practices” as a core function of the research role itself — an engineering competency embedded inside the job description, not a skill that lives across an org boundary. Target runs Data Sciences and Advanced AI Engineering as separate, parallel organizations for leveling and hiring-bar purposes — different skill emphasis, different academic requirements — but the day-to-day work is co-developed on shared initiatives, not handed off in sequence.

That makes Applied/Research Science the right model for this org chart: a second peer ladder, parallel to AI Engineering, both reporting up to the same Director, integrated on shared initiatives rather than connected by a one-way handoff. The org boundary exists for career-ladder and hiring-bar reasons — research roles screen more heavily on academic depth and methodological originality, engineering roles screen more heavily on production delivery track record — not because the work itself splits cleanly into “research phase” and “build phase.” The chart below shows the reporting structure; further down, after both ladders are defined, a second diagram shows how the two tracks actually co-develop on a shared initiative. Below is the org chart, followed by the Applied/Research Science ladder and the AI Engineering ladder it co-develops with.

Applied / Research Science Track

This track is where causal inference, experimentation design, predictive modeling, and original methodological work live — the questions of what should we measure, what’s actually causal versus correlated, and what’s the right model for this business problem — as distinct from the AI Engineering track’s focus on how do we build, deploy, and operate the system that uses that model. Both are essential; conflating them, the way many overscoped AI Engineer postings do, is the same scoping error this series has flagged repeatedly, just from the other direction.

Applied Scientist I / II

Position Summary

Builds and validates models and analyses under the direction of a Senior or Staff Applied Scientist, typically embedded with a specific business domain (marketing, supply chain, pricing, personalization). Translates a defined business question into a modeling approach and executes it; does not yet set methodology or framework for the team.

Essential Functions

• Perform exploratory data analysis, identify anomalies, and validate data quality before modeling
• Build and evaluate predictive or causal models against a defined business question, under Senior/Staff guidance on method selection
• Design and analyze A/B tests and quasi-experiments, including identifying double-counting and grain mismatches in retail data
• Communicate findings to business partners with clear narrative, visualization, and a stated recommendation
• Write production-quality, tested, documented code suitable for handoff to AI/ML engineering for productionization
• Partner directly with AI Engineer I/II counterparts on specific feature initiatives

Tools / Technology Required

Python (pandas, scikit-learn, statsmodels); SQL; experimentation/A/B testing frameworks; data visualization tooling (Tableau, Looker, or equivalent); foundational statistics — linear algebra, probability, optimization. Familiarity with one causal inference method (difference-in-differences, propensity matching, or instrumental variables) is increasingly expected even at this level, not just at Senior.

Experience / Qualifications

0–4 years of professional experience, or a master’s/PhD in a quantitative field (Statistics, Economics, Applied Math, Computer Science) with 0–2 years industry experience. Comparable large-retailer Sr Data Scientist postings specify a master’s or PhD as a baseline qualification even at the II/Senior boundary — a meaningfully higher academic bar than the AI Engineering ladder, which screens primarily on production project work.

Independent Judgment

Executes modeling work within a defined business question and established methodology. Flags ambiguity in the question itself rather than resolving it independently.

Compensation

$95,000–$130,000 base, COL-adjusted for a Midwest HQ market. National range $105,000–$150,000.

Senior Applied Scientist

Position Summary

Owns the modeling and causal-analysis approach for a business domain end-to-end — designing the experiment, building the model, and translating results into a recommended action — and is the first level expected to push back on a poorly framed business question rather than just answer it as asked.

Essential Functions

• Translate ambiguous business problems into a well-specified modeling or experimentation approach, including identifying what’s causal versus correlated
• Design quality, flexible, performant data science systems and services at scale for their domain
• Architect solutions that capture, process, and mine large retail datasets, partnering with Data Engineering on what “analysis-ready” data requires
• Collaborate directly with AI/ML engineering peers to productionize models, including handoff of evaluation criteria and retraining triggers
• Lead experimentation strategy for their domain: test design, statistical power, guardrail metrics
• Mentor Applied Scientist I/II team members on methodology and code quality
• Present findings and recommendations directly to business stakeholders and product partners

Tools / Technology Required

Everything at the Applied Scientist II level, plus: production-scale data tooling (PySpark, Hive, or equivalent); deeper causal inference methodology; recommender systems, reinforcement learning, or search/ranking expertise depending on domain; working knowledge of cloud AI/ML platforms and deployment practices, even though deployment itself is typically owned by an AI engineering partner.

Experience / Qualifications

5+ years of professional experience in applied statistics, econometrics, or data science, including a demonstrated record of translating analysis into business decisions. Comparable large-retailer Sr Data Scientist postings list 3+ years of production ML implementation as a floor alongside a master’s or PhD — confirming this level expects both the academic depth and the production track record, not one or the other.

Independent Judgment

Exercises independent judgment on modeling approach and experiment design within their domain. Pushes back on the business question itself when the data doesn’t support it as framed. Escalates only decisions with cross-domain or company-wide measurement implications.

Compensation

$140,000–$175,000 base, COL-adjusted for a Midwest HQ market. National range $150,000–$190,000.

Staff Applied Scientist

Position Summary

Owns the research and methodology direction for a research domain spanning multiple business areas, and sets the evaluation standards other scientists on the team are accountable to — the applied-science equivalent of the Staff AI Engineer’s platform-wide scope.

Essential Functions

• Set methodology standards for causal inference, experimentation, and model evaluation used across the Applied Science team
• Lead research on emerging methods (causal ML, advanced experimentation design, novel optimization approaches) and translate them into team practice
• Partner with Staff/Principal AI Engineers on which research outputs are ready for productionization and what that handoff requires
• Lead cross-domain initiatives where the same methodological question shows up in multiple business areas (e.g., attribution methodology used in both marketing and merchandising)
• Mentor Senior Applied Scientists and serve as the team’s methodological escalation point
• Represent Applied Science in architecture reviews for major AI initiatives, ensuring the modeling approach is sound before engineering investment begins

Tools / Technology Required

Everything at the Senior level, with org-wide methodology depth: advanced causal inference (synthetic control, structural causal models); experimentation platform design; research literature fluency sufficient to evaluate which new methods are production-ready versus still academic.

Experience / Qualifications

7–10+ years of professional experience in applied research, econometrics, or data science, typically including a PhD or equivalent demonstrated research depth (publications, original methodology, or a comparable independent research record).

Independent Judgment

Sets methodological standards and precedent other scientists build against. Makes recommendations that affect how the company measures success across multiple domains. Escalates only decisions with company-wide measurement or research-direction implications.

Compensation

$165,000–$205,000 base, COL-adjusted for a Midwest HQ market. National range $180,000–$225,000.

Principal Applied Scientist

Position Summary

The senior-most individual contributor on the research track, setting multi-year research direction and serving as the company’s deepest authority on what’s measurable, what’s causal, and what new methodology is worth investing in. Target’s own Principal Data Scientist posting for its Advanced Machine Learning team describes contributing to the research, design, and implementation of leading-edge generative AI techniques while guiding other team members on modular architecture and engineering practice — a real-world description of a research role that still stays close enough to production to shape what gets built, without being the one who builds it.

Essential Functions

• Set multi-year research direction across the Applied Science org, identifying which methodological investments will matter in two to three years
• Serve as final authority on causal claims underlying major business decisions
• Partner directly with the Principal AI Engineer and Director on which research initiatives become engineering investments, and in what sequence
• Represent Applied Science in executive-level conversations about what the company can and cannot validly claim from its data
• Identify organizational gaps in research capability before they limit what the company can credibly measure
• Mentor Staff Applied Scientists and contribute to the research hiring bar and leveling decisions

Tools / Technology Required

Everything at the Staff level, at company-wide scope, with the expectation of original methodological contribution — not just applying known methods, but adapting or developing new ones for problems the existing literature doesn’t cleanly cover.

Experience / Qualifications

10+ years of professional experience in applied research or data science, with a PhD or equivalent research record strongly preferred. A demonstrated track record of original published work, open-source research tooling, or comparable independent research contribution is a meaningful differentiator at this level — the kind of evidence that a resume listing only internal project work typically can’t provide.

Independent Judgment

Final authority on research direction and methodological soundness company-wide. Operates with minimal oversight on highly ambiguous measurement and causal-inference problems. Decisions routinely affect what the company can credibly claim about its own data.

Compensation

$195,000–$240,000 base, COL-adjusted for a Midwest HQ market. National range $210,000–$270,000. Target’s own market data shows Principal-level technical roles in this AI org landing between roughly $168,000 and $303,000 depending on level and location, which brackets this band reasonably on the national side.

With both peer ladders defined, the diagram below shows what that integration actually looks like in practice — not a handoff from research to engineering, but a shared workflow where both tracks frame, build, and ship together, and production results close the loop back to validate the original causal claim.

A concrete example makes this less abstract. Take an LLM-based RAG assistant shipping a groundedness eval as a release gate — already a named responsibility in the Senior AI Engineer role below. AI Engineering owns building that gate: wiring the eval harness into the deployment pipeline, setting the threshold that blocks a release, instrumenting the pipeline so a failed eval actually stops a bad deploy rather than just logging a warning. Applied Science owns whether that gate is measuring the right thing: whether the eval’s test set is representative of real usage, whether the groundedness score actually correlates with hallucination rather than just plausible-sounding text, and — once the feature ships — whether a reported lift in conversion or engagement is a causal effect of the assistant or a correlation confounded by something else running in the business that week. An engineer can build a technically correct eval gate that’s measuring the wrong thing and pass every release anyway. A scientist can design a methodologically airtight eval that never gets wired into the pipeline as an actual gate and never blocks a bad deploy. Both failure modes are real, and avoiding both is why this has to be co-development rather than a handoff in either direction.

Here is the AI Engineering ladder itself.

AI Engineer I

Position Summary

The AI Engineer I is the entry point to the AI engineering ladder. The role works under direct mentorship from Senior and Staff engineers on well-scoped, clearly defined tasks within an established codebase and architecture. This is a learning-and-building role, not a decision-making one — independent architectural judgment is explicitly not expected at this level.

Essential Functions

• Implement features and bug fixes within an existing, established codebase under active code review
• Write unit and integration tests for components they own
• Assist with data preparation and transformation tasks scoped by Senior or Staff engineers
• Build and maintain prompt templates within an established prompt library, with PromptOps review from a Senior engineer before anything ships
• Monitor production dashboards and escalate anomalies according to documented runbooks
• Participate in peer code review as both reviewer and reviewee
• Document completed work clearly enough that a Senior engineer can review it asynchronously

Tools / Technology Required

Python; SQL; Git; one major cloud platform (AWS, Azure, or GCP — pick the company’s primary, not all three); a foundation-model API (OpenAI, Anthropic, or equivalent); basic familiarity with a RAG framework (LangChain or LlamaIndex) is a plus but not required at this level.

Experience / Qualifications

Bachelor’s degree in Computer Science, Data Science, or related field, or equivalent practical experience. 0–2 years of professional experience. Portfolio or coursework demonstrating applied ML or LLM project work is valued more than credentials alone — most entry-level postings in this space now screen on demonstrated project work first.

Independent Judgment

Performs duties within close supervision. Follows established patterns and documented guidance. Escalates ambiguity rather than resolving it independently.

Compensation

$100,000–$125,000 base, COL-adjusted for a Midwest HQ market. National data is consistent enough across sources to anchor this with confidence: Glassdoor reports an entry-level AI engineer average of $128,000, with a 25th–75th percentile band of $96,000–$178,000; separate 2026 market guides put entry-level (0–2 years) base at $100,000–$130,000 nationally, with major-tech-hub offers starting at $115,000–$135,000 before equity. This is meaningfully higher than a generic entry-level software engineering band — the specialization premium shows up immediately, even at the entry level, which is itself worth noting for any company benchmarking this role against general SWE bands instead of AI-specific ones.

AI Engineer II

Position Summary

The AI Engineer II operates independently within a scoped problem space, owning individual features or pipeline components end-to-end under architectural direction set by Senior or Staff engineers. This is the level most postings titled “AI Engineer” should actually be written against — and, per this series, almost never are.

Essential Functions

• Design, build, and deploy individual LLM or RAG features within an established system architecture
• Write and maintain prompt templates; contribute new patterns to a shared, version-controlled prompt library
• Build and run evaluation harnesses for the features they own, using existing eval frameworks rather than building new ones from scratch
• Own data pipeline components feeding their features, including basic data quality checks
• Participate in an on-call rotation for the specific services they own, with Senior-level backup available
• Collaborate directly with one or two business stakeholders on requirements for the features they’re building
• Contribute to internal documentation and best-practice guides

Tools / Technology Required

Python; SQL; one cloud platform; a RAG framework (LangChain, LlamaIndex, or equivalent); a vector database (Pinecone, Weaviate, Milvus, or pgvector); Docker; basic CI/CD familiarity; a RAG evaluation framework (Ragas or DeepEval) is increasingly expected at this level, not just at Senior — comparable roles at companies like Acubed by Airbus already list hands-on experience with vector databases and RAG evaluation frameworks as a II-level expectation, not a reach requirement.

Experience / Qualifications

2–4 years of professional experience, with at least 1 year of hands-on production LLM/RAG work. Comparable production-focused LLM roles in the broader market set the bar at 4+ years of software engineering experience with at least 2 years specifically in production LLM application development — at the high end, this maps closer to a strong II or an early Senior, which is itself a useful calibration check for any posting using “3–5 years” language while listing Staff-level scope.

Independent Judgment

Operates independently within a clearly scoped problem. Makes implementation-level decisions; does not set architecture or precedent for other engineers.

Compensation

$130,000–$160,000 base, COL-adjusted for a Midwest HQ market. National range $140,000–$175,000. This sits below the “mid-level AI engineer” band several staffing-firm sources report for 2026 — $160,000–$210,000 base for three-to-five-year, LLM/RAG-capable engineers nationally — because that band is better calibrated to the Senior level below, not II; it’s a useful reminder that market guides aggregating “mid-level” loosely tend to blend what this ladder splits into II and Senior. A comparable AI application engineering role in Santa Clara, requiring 4+ years and RAG pipeline ownership, posted a salary range of $130,000 to $170,000 — useful context for how much a major coastal market premium adds on top of an otherwise similar scope, and a reasonable national ceiling for a strong II-level hire.

Senior AI Engineer

Position Summary

The Senior AI Engineer owns a full AI feature area end-to-end — for example, a customer-facing RAG assistant or an internal agentic automation suite — and makes independent technical decisions within that scope. This is the first level where “independent judgment” language genuinely describes the job, rather than being aspirational language attached to a mid-level role.

Essential Functions

• Architect and lead delivery of a full AI feature area, from data sourcing through deployment and monitoring
• Define evaluation criteria and groundedness/safety release gates; own PromptOps practices for their area, including prompt versioning and structured-output guardrails
• Design agentic workflows including step limits, kill switches, feature flags, and human-in-the-loop approval points for high-impact actions
• Own model lifecycle decisions for their area: experiment tracking, model registry, retraining cadence, embedding/vector-index versioning, rollback strategy
• Set service-level objectives for systems they own and lead incident response when those objectives are breached
• Mentor AI Engineer I and II team members through code review and design feedback
• Represent the AI team in cross-functional planning with product, data, and security stakeholders
• Propose new automation opportunities to stakeholders with cost and ROI framing attached

Tools / Technology Required

Everything at the II level, plus: agent orchestration frameworks (LangGraph, AutoGen, or equivalent); MLOps tooling for experiment tracking and model registry (MLflow or Weights & Biases); observability tooling for token/cost monitoring and tracing; containerization and orchestration (Docker, Kubernetes) at an operate-independently level, not just familiarity. The 2026 default production stack across comparable postings is consistently MLflow plus Docker plus Kubernetes plus one cloud platform — a Senior-level candidate should be fluent in that exact combination, not adjacent tools.

Experience / Qualifications

5+ years of professional experience, with 3+ years specifically in production LLM, RAG, or agentic systems. Comparable Principal-level postings at large retailers set the bar even higher — 7+ years of hands-on software engineering and applied AI/ML development, including production deployment of LLM-based or agentic systems — which is a useful signal that this Senior band genuinely sits a full step below Principal, not adjacent to it.

Independent Judgment

Exercises independent judgment within their feature area. Makes recommendations that influence team practices and patterns. Escalates only decisions with cross-team or company-wide impact.

Compensation

$155,000–$195,000 base, COL-adjusted for a Midwest HQ market. National range $165,000–$210,000. This aligns closely with what staffing-firm 2026 data reports for AI-specialized mid-to-senior engineers nationally — $160,000–$210,000 base for three-to-five-year LLM/RAG-capable engineers, with another 10–15% added for confirmed generative AI depth — which, mapped against this ladder’s experience bar of 5+ years with 3+ specifically in production LLM/RAG work, lands squarely on this Senior level rather than on II. Microsoft’s published IC5 band for a comparable RAG-focused Member of Technical Staff role — designing large-scale retrieval and personalization systems — runs $139,900 to $274,800 nationally, underscoring how wide the spread gets once a candidate pool includes frontier-lab compensation; a Midwest retailer won’t and shouldn’t try to match the top of that range, but the bottom of it is a useful floor check.

Staff AI Engineer

Position Summary

The Staff AI Engineer owns a platform domain spanning multiple product surfaces and multiple engineers. This is the first level expected to set technical direction that other engineers are accountable to, rather than just executing within direction set by someone else.

Essential Functions

• Define reference architecture and platform standards for AI engineering across the org, not just within their own feature area
• Lead evaluation and selection of foundation models, embedding strategies, vector retrieval architecture, and inference infrastructure at the platform level
• Own PromptOps and LLMOps tooling strategy used by every Senior, II, and I engineer on the team — prompt versioning systems, eval pipelines, release-gate frameworks
• Design the observability strategy for AI systems org-wide: tracing, hallucination detection, token/cost monitoring, alerting standards
• Lead architecture reviews for major new initiatives before implementation begins
• Mentor Senior engineers and serve as the team’s technical escalation point
• Drive build-vs-buy decisions on AI infrastructure components
• Partner with the Director on technical roadmap and capacity planning

Tools / Technology Required

Everything at the Senior level, with platform-level (not just feature-level) depth: multi-model evaluation and routing strategy across foundation model providers; infrastructure-as-code for AI platform components; cost-optimization tooling for inference at scale; cross-cloud or multi-region deployment patterns where applicable.

Experience / Qualifications

7–10+ years of professional experience, with deep production experience across at least two of: LLM/RAG systems, agentic orchestration, MLOps/LLMOps tooling, large-scale data platforms. This maps closely to how comparable large-retailer Principal-level postings are actually scoped — 7+ years hands-on, extensive experience with agent development frameworks, evaluation and observability tooling, cloud ML platforms, and production deployment of LLM-based and agentic systems, which is worth noting: at several large retailers, “Staff” and “Principal” titling conventions overlap, so this band and the next one are sometimes the same role under different naming.

Independent Judgment

Sets technical direction and precedent that other engineers build against. Makes architecture-level recommendations that affect engineering standards org-wide. Escalates only decisions with company-wide budget or risk implications.

Compensation

$175,000–$215,000 base, COL-adjusted for a Midwest HQ market. National range $190,000–$240,000.

Principal AI Engineer

Position Summary

The Principal AI Engineer is the senior-most individual contributor on the ladder, operating at the intersection of multiple platform domains and setting multi-year technical direction. Target’s own Principal AI Engineer posting for its centralized Advanced AI team describes the role as providing deep technical leadership across enterprise AI initiatives, partnering with engineering, product, and business teams to shape scalable solution patterns, and influencing technical direction and engineering standards across AI initiatives — a real-world description that maps closely to how this level should be scoped at any large retailer building serious AI capability.

Essential Functions

• Set multi-year technical vision for the AI platform, spanning data infrastructure, model serving, agentic systems, and governance touchpoints
• Make final architectural calls on company-wide AI infrastructure decisions
• Architect agentic and LLM-powered systems end to end, evaluate and select the right models and frameworks, and establish observability, evaluation, and engineering practices that ensure solutions are production-ready, maintainable, and well-documented
• Represent AI engineering in executive-level conversations about AI strategy, capability, and risk
• Identify organizational gaps in skills, tooling, or process before they become delivery blockers
• Lead the response to high-ambiguity technical problems with no existing internal playbook
• Stay current on emerging technologies, recommend tooling and architectural approaches, and publish learnings for the team and broader organization
• Mentor Staff engineers and contribute to org-wide hiring bar and leveling decisions

Tools / Technology Required

Everything at the Staff level, at company-wide scope. Comparable Principal-level postings require extensive experience with AI engineering tooling and platforms — agent development frameworks, model APIs, evaluation and observability frameworks, cloud ML platforms, containers, and orchestration technologies — the expectation is fluency across the full stack, not specialization in one slice of it.

Experience / Qualifications

10+ years of professional experience, with a track record of having built and scaled at least one production AI platform from early stage to enterprise scale. Comparable postings specify 7+ years of demonstrated hands-on experience in software engineering and applied AI/ML development as a floor, with most real-world Principal hires landing meaningfully above that minimum. A master’s or PhD in a related technical field is commonly listed as preferred, not required.

Independent Judgment

Final technical authority on company-wide AI architecture decisions. Operates with minimal oversight on highly ambiguous, high-impact problems. Decisions routinely affect engineering standards and technical direction across the entire organization.

Compensation

$210,000–$260,000 base, COL-adjusted for a Midwest HQ market. National range $230,000–$290,000+. Median total compensation for Machine Learning Engineer roles broadly is reported at $270,000, and Walmart’s own reported Machine Learning Engineer compensation tops out at a yearly total of $416,600 at the senior end, with a median of $330,000 — though it’s worth flagging directly that figures like Walmart’s include meaningful stock compensation, which a privately held company cannot offer. At a private retailer, base salary has to do more of that work, which is exactly the dynamic worth naming explicitly in any Principal-level offer at a company without public equity.

Director of AI Engineering

Position Summary

The Director owns delivery across the entire AI engineering org — the full IC ladder above, plus dotted-line coordination with partner functions like Data Engineering, Platform/MLOps, and Governance. This is the first role on the chart where formal people management is the primary function, not a side effect of technical seniority.

Essential Functions

• Own AI engineering headcount planning, leveling, and hiring bar across all IC levels
• Translate AI strategy set by executive leadership into a delivery roadmap and team structure
• Manage Staff and Principal engineers directly; oversee the full ladder through skip-level engagement
• Own the engineering budget for the AI org: headcount, infrastructure spend, tooling licenses
• Serve as the accountable owner when production AI incidents carry business impact
• Partner with Legal, Security, and Compliance on AI governance policy
• Report team health, delivery status, and technical risk to executive leadership

Tools / Technology Required

Working fluency (not hands-on daily use) across the full AI engineering stack described above, sufficient to evaluate technical tradeoffs presented by Staff and Principal engineers and to credibly represent the team’s technical position to non-technical executives.

Experience / Qualifications

10+ years of professional experience including prior technical leadership or people-management experience, ideally including direct management of Senior or Staff-level engineers. A comparable large-retailer organization staffs this exact function as a dedicated Senior Engineering Manager role sitting alongside, not above, its Principal AI Engineer individual contributor track — confirming this is a distinct discipline from the Principal IC role above it on this chart, not a natural promotion path from it.

Independent Judgment

Owns org-level decisions on hiring, budget, and delivery priorities within the boundaries set by executive AI strategy. Escalates only decisions requiring budget reallocation outside their approved scope or company-wide policy change.

Compensation

$200,000–$255,000 base, COL-adjusted for a Midwest HQ market. National range $220,000–$280,000, typically with a meaningful annual bonus and, at public companies, long-term equity incentive — a lever a privately held retailer does not have available and should account for in base pay accordingly.

What This Investment Actually Costs, and What It’s Realistic to Expect Back

Everything above is a leveling and scoping framework. The question a CFO will actually ask is simpler: what does properly staffing this cost, and what do we get for it? Both halves deserve real numbers, not aspirational ones — and the honest answer to the second half is less triumphant than most vendor pitches suggest, which is itself useful information for anyone building the business case.

What it costs

Phase one — a minimum viable team to ship a real production AI capability, not the full org chart: one Senior AI Engineer, one Senior Applied Scientist, and one AI Engineer II, with the Director role filled by an existing engineering leader taking on the mandate rather than a dedicated new hire. At the COL-adjusted Midwest bands above, that’s roughly $155,000–$195,000 plus $140,000–$175,000 plus $130,000–$160,000 in base salary — $425,000 to $530,000 in base compensation, before benefits load (typically adds 20–30% on top, so realistically $510,000–$690,000 fully loaded) and before infrastructure or tooling spend.

Full org chart — both ladders at mature staffing (Director, Principal, Staff, two Seniors, two Engineer I/II on the engineering side; Principal, Staff, two Seniors, two Applied Scientist I/II on the research side): roughly 12 people, with base compensation summing to $2.1–$2.6 million annually, or $2.5–$3.4 million fully loaded with benefits. That figure sits inside the range this series would expect: it’s well below the $7–11.6 million average enterprise AI spend reported for 2026, because that aggregate figure typically includes infrastructure, licensing, and vendor spend on top of headcount — the people cost is usually the smaller half of the total bill, not the larger one.

What the return actually looks like — and why most companies aren’t seeing it

This is the part worth being honest about rather than optimistic. The 2026 data on enterprise AI ROI is unusually consistent across independent sources, and the headline number is not encouraging: an MIT NANDA study of 300 public AI deployments found that 95% of generative AI pilots delivered no measurable profit-and-loss impact. PwC’s 2026 Global CEO Survey found 56% of CEOs report no revenue or cost benefit from AI at all. Deloitte’s research found only 6% of organizations achieve payback in under a year.

The more useful finding sits underneath that headline number, though, and it’s the entire argument for why staffing matters: the failures cluster around the same root causes. Workday’s research found that “time saved” by AI tools often isn’t converted into business value, because employees spend the saved hours correcting errors and verifying outputs — what researchers are calling a new productivity paradox. Other research points to the same pattern from a different angle: the top 5% of companies generating real value at scale produce 1.7x revenue growth and 3.6x total shareholder return versus laggards, and the distinguishing factor isn’t access to better models — it’s whether the company has the organizational discipline (dedicated teams, outcome-based metrics, clear ownership) to convert AI capability into a measured business outcome.

That’s the direct case for the org chart this piece describes. A single overscoped Engineer II hire, with no Applied Science partner to validate what’s actually causal versus correlated and no governance partner to keep deployment compliant, is structurally the same failure mode MIT, PwC, and Deloitte are describing at scale — just compressed into one person instead of one company. Proper staffing doesn’t guarantee the 1.7x outcome. But understaffing according to this pattern is close to a guarantee of landing in the 95% that don’t show measurable return, because the org never had the capacity to do the measurement, validation, and iteration work that separates a pilot from a production capability with a real number attached to it.

Where real, validated returns do show up, financial services leads at a reported 4.2x return per dollar invested, with media and telecom close behind — both sectors with mature data infrastructure and dedicated applied science functions, not single-engineer AI initiatives. Retail’s reported adoption rate (64%) trails financial services and technology, which is consistent with the thesis running through this whole series: the postings, and the org structures behind them, are less mature.

The token-cost and build-vs-buy question, answered honestly

This deserves direct treatment because it’s a common assumption worth correcting: building and self-hosting frontier models internally is not, for the large majority of enterprise workloads, a cost-saving move. Multiple independent 2026 analyses converge on the same finding: managed APIs remain cheaper than self-hosting open-weight models for roughly 87–95% of production workloads, with the breakeven point sitting somewhere between 100 million and 11 billion tokens per month depending on the model and provider compared. Below that volume, the fully loaded cost of self-hosting — GPU rental or purchase, the 20–30% of a senior engineer’s time required to maintain the inference stack, model update cycles, and idle-capacity waste — runs 3 to 5 times higher than the equivalent API spend in real-world comparisons, including one documented case where a company’s self-hosted deployment cost 5.6 times more than the API alternative it was meant to replace.

The real driver for running models on owned infrastructure isn’t compute economics — it’s data residency and regulatory exposure. A retailer processing payment-card data, loyalty-program PII, and consumer purchase history has a genuine compliance case for routing sensitive workloads through a private endpoint or self-hosted gateway, particularly as the EU AI Act’s August 2026 enforcement date and tightening state-level privacy law raise the cost of getting this wrong. The pattern emerging across regulated industries in 2026 isn’t “build everything in-house” — it’s a hybrid architecture: route the bulk of general-purpose AI traffic through managed APIs (cheaper, zero infrastructure burden, instant access to frontier model improvements), and reserve self-hosted or private-endpoint infrastructure specifically for the workloads touching regulated data, where the requirement is sovereignty and auditability rather than cost savings.

This is precisely the kind of architectural judgment call this org chart is built to support. It’s a Staff or Principal AI Engineer decision — build-vs-buy, batch-vs-real-time, which workloads warrant the operational overhead of self-hosting — not a decision a mid-level Engineer II should be making alone, and not a decision that gets made well without an Applied Scientist partner who can validate whether the in-house model’s accuracy actually holds up against the frontier API it’s replacing. Getting this architecture wrong in either direction — over-building expensive infrastructure nobody needed, or under-protecting data that needed a private endpoint — is itself a measurable cost, and it’s exactly the kind of cost a properly staffed team is positioned to avoid that an understaffed one typically isn’t.

What This Solves

Map any of the overscoped postings this series has examined against this structure, and the diagnosis stops being subjective. A posting asking for LLM/RAG development, PromptOps, agentic workflow design, causal modeling, experimentation design, data pipeline ownership, model lifecycle management, observability, governance touchpoints, and on-call — bundled into one mid-level title with no disclosed salary — decomposes cleanly into a Senior AI Engineer role (LLM/RAG, PromptOps, agentic workflows), a Senior Applied Scientist role (causal modeling, experimentation design) on the parallel research ladder, plus a meaningful slice of partner-function work that belongs to Data Engineering and Platform/MLOps, not to either ladder.

That’s three or four distinct functions doing the work, not one person. A hiring manager who needs all of that output and only has budget for a single req has a budget and org-design problem, not a job-posting problem — and no amount of rewriting the bullets fixes that. The companies that get this right at scale, like Target, simply staff it as separate roles from the start: a Principal AI Engineer focused on architecture and technical direction, a Principal Data Scientist focused on research and methodology, a separate engineering management track for people leadership, a dedicated ML Ops-focused engineer for platform reliability, and a fully separate Data Governance function — distinct ladders that partner, not one person doing all of it.

For Candidates

If you’re evaluating a posting that looks like the ones this series autopsies, use this structure as a gut check. Ask in the screen which of the bullets map to AI Engineering versus Applied/Research Science versus Data Engineering versus Platform/MLOps versus Governance. A recruiter or hiring manager who can answer that cleanly has scoped the role with an org chart in mind, even if the title or pay still needs adjustment. One who can’t has a posting assembled from a wish list, not a team design. This distinction matters especially for candidates with a research or applied-science background: a posting titled “AI Engineer” that’s actually asking for causal inference, experimentation design, and methodology work is mistitled in the other direction — and worth flagging in the screen just as much as the reverse.

For Employers

The fix is rarely “write a better job posting.” It’s “decide who owns what before the req goes live.” A single posting with twenty bullets spanning AI engineering, data engineering, platform operations, and governance isn’t an aggressive ask — it’s a sign the org chart, and the partnership with whatever shared functions already exist at the company, was never drawn.

Job Posting Autopsy is a weekly series that applies structured analysis to data science and AI job listings. The goal is not to mock employers but to help candidates make informed decisions — and to push the industry toward more honest role descriptions and team structures. Have a posting you’d like reviewed? Drop it in the comments.

About the author

Brian Curry is a Kansas City–based data scientist and AI strategist with 20+ years across enterprise analytics, applied AI, and digital strategy, including roles at Vail Resorts, Tractor Supply, Koch Industries, Hallmark, Garmin, and AT&T. His work spans predictive modeling, causal inference, marketing economics, and the practical realities of building AI capability inside large organizations — the same lens behind this series. He is the founder of Vector1 Research, an independent lab focused on AI economics and causal systems, and the creator of Papilon, an open-source Python library for marketing analytics and complex-systems simulation.