Twelve-month plan to convert existing private depth into externally legible signal. Targeted at consulting organizations and significant employers staffing agentic-AI expert roles.
- Schema-first, governance-only. Carry the md-util stance across every new AI artifact — no row movement, no live DDL, blast radius bounded.
- Ship one public artifact per quarter. Private depth doesn’t recruit; the goal is the public surface, not new private depth (which is already adequate).
- Earn credentials that close gaps the work already covers — Databricks and Anthropic builder partner. Skip the rest.
- Hold the deterministic / non-deterministic line. It’s the defining judgment call and the strongest differentiator. Do not blur it for fashion.
| Shape | Why this candidate fits | What still needs to be true on paper |
|---|
| Principal / Staff Agentic Engineer (frontier-AI consultancy or product co) | 10,879-LOC SHACL governance CLI + 67 Claude Code skills + Nate B. Jones 5-level pass + HDD–SDD–BDD methodology | One public artifact; one talk; Anthropic credential |
| AI Adoption / Enablement Architect (large consultancy: McKinsey QuantumBlack, Deloitte AI, BCG X) | Hyperion Master Curriculum + AI 5-section program + 30y enterprise DW depth + agentic ops library | Case study; reference letters; Databricks cert |
| Field / Solutions Architect — Anthropic / similar | Production Claude Code at platform scale; custom MCP; 92-ticket Linear cycle | Anthropic builder partner; reference customer story |
| Independent agentic-consulting practice (Metro Decisions) | Already operating; portfolio in place | Public artifact + talk drive inbound; rate card defensible |
2026 Q3 (Jul–Sep) — Externalize
Link to heading
| Item | Outcome |
|---|
Publish redacted SHACL prompt-vocabulary OSS drop (mdu-vocab) | Pullable artifact, GitHub README, one-page rationale |
| Conference / meetup talk submission: “Schema-first agentic data governance” | Submitted to ODSC, Data Council, or DataEngBytes |
Build mdu eval prompt-regression rig on existing 10-dataset batch | New mdu subcommand; first eval report committed |
| Anthropic builder partner / certified-builder application | Application in flight |
| Write up Nate B. Jones 5-level pass as a one-page proof | Add to portfolio site |
2026 Q4 (Oct–Dec) — Credential & speak
Link to heading
| Item | Outcome |
|---|
| Finish Databricks certification track | Cert listed on CV |
| Deliver Q3 conference talk | Slides + recording link added to resume |
| Publish HDD–SDD–BDD write-up | First long-form post; series outline |
Cross-model eval results in mdu eval | Claude / GPT / Llama / locally trained nanoGPT side-by-side |
| Open-source the prompt library + XML schema | “Prompt-as-code” repo |
2027 Q1 (Jan–Mar) — Anchor frontier credential
Link to heading
| Item | Outcome |
|---|
| Anthropic builder partner status (or equivalent) | Logo on CV |
| Second talk OR client case study from a 2026 engagement | Citation-grade reference |
| Open-source one MCP server (e.g. msgvault-MCP companion) | Shareable repo |
| Expand eval rig to RAG + tool-use benchmarks | Eval-engineering line is concrete |
| LMTRAIN-2 — train a domain-tuned small model on practice-docs corpus | “Trained a domain model in service of an agentic data product” line |
2027 Q2 (Apr–Jun) — Convert to revenue / role
Link to heading
| Item | Outcome |
|---|
| Active pursuit of target role shapes | Defined pipeline |
practice-docs book outline → drafted Part I | Public draft, leveraged in interviews |
| Logos (or successor product) — paid pilot or grant | Revenue or funded R&D line |
| One paid speaking engagement | External validation + rate card support |
Skill build queue (additive per quarter)
Link to heading
| Quarter | Add skill |
|---|
| 2026 Q3 | Formal eval-rig vocabulary (Inspect / Promptfoo terminology) |
| 2026 Q3 | Anthropic prompt-caching, extended thinking, batch API patterns |
| 2026 Q4 | Cross-model orchestration via Claude Code (agent({model:…})) at scale |
| 2026 Q4 | RAG hardening (Qdrant production patterns, hybrid retrieval) |
| 2027 Q1 | Multi-agent fan-out patterns (Workflow library + adversarial verify) |
| 2027 Q1 | Tool-use eval (tool-call accuracy, planning depth) |
| 2027 Q2 | Production cost / latency engineering (cache-hit math, model-tier selection) |