[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6MFDTCZEE2G4SRJH7ZC4QTQ":3,"skill-same:01M6MFDTCZEE2G4SRJH7ZC4QTQ":71},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":38,"linkOnly":39,"markdownContent":40,"markdownBody":41,"frontMatter":42,"refName":51,"commitSha":52,"skillDirectory":53,"forksCount":13,"repositoryHtmlUrl":54,"repositoryHomepageUrl":6,"ownerHtmlUrl":55,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":56,"install":57,"resources":70},"01M6MFDTCZEE2G4SRJH7ZC4QTQ","chief-ai-officer-advisor",null,"Chief AI Officer advisory for startups: model build-vs-buy decisions (API vs fine-tune vs in-house), AI risk classification under EU AI Act + US state patchwork, AI cost economics (API-to-self-hosted breakeven), and AI team org evolution. Use when deciding whether to call an API or fine-tune, classifying AI use cases for regulatory risk, calculating when self-hosting pays off, sequencing AI hires, or when user mentions CAIO, AI strategy, model selection, foundation model, fine-tuning, EU AI Act, NIST AI RMF, AI governance, model risk, or AI economics. Strategic only — does not duplicate engineering AI\u002FML skills.","JiayiV\u002Fquestion_mixing","JiayiV","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F54466889?v=4","01M6JFJYYMRZ3QRGP9339JNQ2K","01M6JFJZ35KB0QVT50NAHPJA7G",0,"MIT","https:\u002F\u002Fgithub.com\u002FJiayiV\u002Fquestion_mixing\u002Fblob\u002Fmain\u002Fskills\u002Fpublic-02-claude-skills-main-gemini-skills-chief-ai-officer-advisor-skill-610b72d6\u002FSKILL.md",[],[18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37],"chief ai officer advisor","chief","advisor","risk classification","ai governance","economics","sequencing","act","leadership","self","ai team","selection","classifier","only","model","team","api","tuning","NIST","model selection","2026-11-10 10:12:30",false,"---\nname: \"chief-ai-officer-advisor\"\ndescription: \"Chief AI Officer advisory for startups: model build-vs-buy decisions (API vs fine-tune vs in-house), AI risk classification under EU AI Act + US state patchwork, AI cost economics (API-to-self-hosted breakeven), and AI team org evolution. Use when deciding whether to call an API or fine-tune, classifying AI use cases for regulatory risk, calculating when self-hosting pays off, sequencing AI hires, or when user mentions CAIO, AI strategy, model selection, foundation model, fine-tuning, EU AI Act, NIST AI RMF, AI governance, model risk, or AI economics. Strategic only — does not duplicate engineering AI\u002FML skills.\"\nlicense: MIT\nmetadata:\n  version: 1.0.0\n  author: Alireza Rezvani\n  category: c-level\n  domain: chief-ai-officer-leadership\n  updated: 2026-05-12\n  python-tools: model_buildvsbuy_calculator.py, ai_risk_classifier.py, ai_cost_economics.py\n  frameworks: model-buildvsbuy, ai-risk-governance, ai-economics, ai-team-org\n---\n\n# Chief AI Officer Advisor\n\nStrategic AI leadership for startup CAIOs and founders without one. **Four decisions, no AI hype:**\n\n1. **Should we use an API, fine-tune, or build our own?** — model build-vs-buy with 3-year TCO\n2. **Is this AI use case high-risk under regulation, and how do we govern it?** — EU AI Act + NIST AI RMF + US state patchwork\n3. **When do we switch from API to self-hosted, and at what cost?** — token economics with breakeven analysis\n4. **What AI role do we hire next?** — stage-to-role map (AI engineer ≠ ML engineer ≠ research scientist)\n\nThis skill does **not** cover tactical AI\u002FML engineering. For RAG implementation, agent design, prompt engineering, eval infrastructure, model deployment, or cost optimization, see `engineering\u002Frag-architect\u002F`, `engineering\u002Fagent-designer\u002F`, `engineering\u002Fprompt-governance\u002F`, `engineering\u002Fself-eval\u002F`, `engineering\u002Fllm-cost-optimizer\u002F`.\n\n## Keywords\n\nCAIO, chief AI officer, AI strategy, model selection, foundation model, fine-tuning, RLHF, DPO, LoRA, QLoRA, build vs buy, AI build-vs-buy, model risk tier, EU AI Act, AI Act Article 6, Article 9, Article 10, Annex III, prohibited AI, high-risk AI, NIST AI RMF, AI risk management framework, NYC Local Law 144, Colorado SB 21-169, Illinois HB 53, model card, eval set, eval harness, hallucination rate, jailbreak risk, prompt injection, AI red team, AI safety, alignment, model lifecycle, model registry, API-to-self-hosted breakeven, GPU economics, A100, H100, inference cost, fine-tuning cost, AI team, AI engineer, ML engineer, research scientist, MLOps, AI platform\n\n## Quick Start\n\n```bash\n# Decision A: API vs fine-tune vs build\npython scripts\u002Fmodel_buildvsbuy_calculator.py                          # embedded customer-support sample\npython scripts\u002Fmodel_buildvsbuy_calculator.py path\u002Fto\u002Fuse_case.json\n\n# Decision B: Risk classification under EU AI Act + US state laws\npython scripts\u002Fai_risk_classifier.py                                   # embedded hiring-AI sample\npython scripts\u002Fai_risk_classifier.py path\u002Fto\u002Fuse_case.json\n\n# Decision C: API vs self-hosted economics\npython scripts\u002Fai_cost_economics.py                                    # embedded 5M tokens\u002Fday sample\npython scripts\u002Fai_cost_economics.py path\u002Fto\u002Fworkload.json\n```\n\n## Key Questions (ask these first)\n\n- **What does this AI need to be good at, and how would you measure it?** (If no eval set, no ship.)\n- **What's the SLO on hallucination \u002F error rate?** (Without one, \"AI quality\" is a vibe.)\n- **What happens when the model is wrong?** (Fallback behavior, human-in-the-loop, blast radius.)\n- **What's the risk tier under EU AI Act, and is conformity assessment required?** (Determines product launch timeline.)\n- **At what monthly token volume does self-hosting beat API?** (Almost never below 100M tokens\u002Fmonth at frontier quality.)\n- **Are we hiring an AI engineer or an ML research scientist?** (Different jobs; founders confuse them.)\n\n## Core Responsibilities\n\n### 1. Model Build-vs-Buy\n\nThe decision is not \"use AI or not\" — it's **API vs fine-tune vs in-house** for each use case. Each path has a different TCO curve, latency profile, and capability ceiling.\n\n**Default path: API (frontier model)**\n- Use when: well-served by frontier (Claude, GPT, Gemini), QPS \u003C 100, latency budget > 1s, cost \u003C $50K\u002Fmonth\n- Why: frontier APIs are 10-100x more capable than what most teams can fine-tune in-house\n- Failure mode: API rate limits at scale, vendor lock-in, capability drift between model versions\n\n**Fine-tune a smaller model**\n- Use when: domain-specific behavior the API can't be prompted into (medical coding, legal redlining), high volume reducing API cost, latency budget \u003C 500ms, specific style\u002Fformat consistency required\n- Approaches: full fine-tune (rare), LoRA\u002FQLoRA (common), RLHF\u002FDPO (when alignment matters)\n- Failure mode: fine-tuned model lags frontier capability within 6-12 months; ongoing retraining cost\n\n**Build from scratch \u002F pre-train**\n- Use when: almost never. You're a foundation-model company, OR you have a unique data corpus, $50M+ funding, and 18+ month patience.\n- Failure mode: by the time you ship, frontier models have caught up and your sunk cost is unrecoverable\n\n**Run** `model_buildvsbuy_calculator.py` for a use-case-specific recommendation with 3-year TCO. See `references\u002Fmodel_buildvsbuy_strategy.md` for full decision tree.\n\n### 2. AI Risk Classification & Governance\n\nThe 2026 question every founder is facing: **does this AI use case trigger high-risk regulatory obligations?**\n\n**EU AI Act (in force 2026) tiers:**\n\n| Tier | Examples | Obligations |\n|---|---|---|\n| **Prohibited** | Social scoring, real-time biometric surveillance, manipulative AI | Cannot deploy in EU |\n| **High-risk** | Employment screening, credit scoring, education access, critical infrastructure, law enforcement, biometric ID | Conformity assessment, registration, post-market monitoring, transparency, human oversight |\n| **Limited-risk** | Chatbots, deepfakes, emotion recognition | Transparency: user must know they're interacting with AI |\n| **Minimal-risk** | Recommendation systems, spam filters, most B2B SaaS internals | No specific obligations |\n\n**Run** `ai_risk_classifier.py` to classify a use case and get the required-controls list.\n\n**US state patchwork (non-exhaustive):**\n\n- NYC LL 144 — Automated Employment Decision Tools (AEDTs) require annual bias audit + candidate notice\n- Colorado AI Act \u002F SB 21-169 — AI in consumer decisions (credit, insurance, employment, housing)\n- Illinois HB 53 — AI in interview\u002Fhiring\n- California SB 1001 — Bot disclosure\n- Texas TCPA — Biometric identifier capture\n- Federal NIST AI RMF — voluntary; increasingly referenced in contracts\n\n**Industry-specific overlays:**\n\n- Healthcare: FDA AI\u002FML guidance (2023), MDR (EU) for medical-device AI, 510(k) pathway for AI\u002FML-enabled medical devices\n- Financial: NYDFS Reg 23, FTC Section 5, ECOA for credit decisions\n- Insurance: NAIC model bulletin, state insurance commissioner rules\n\nSee `references\u002Fai_risk_governance.md` for the full regulatory landscape + governance program checklist.\n\n### 3. AI Cost Economics\n\n**The breakeven question:** at what monthly token volume does self-hosted inference beat API costs?\n\n**Key components:**\n\n- **API cost** — variable, per-token. Frontier models 2026: Claude Sonnet 4.6 ~$3\u002F$15 per M tokens (input\u002Foutput), GPT-4o ~$2.50\u002F$10, Gemini 2.5 ~$1.25\u002F$5\n- **Self-hosted cost** — fixed (GPU commitment) + variable (electricity). H100 spot ~$2-5\u002Fhour, A100 spot ~$1-3\u002Fhour. Llama 3.1 70B \u002F Qwen 2.5 72B: ~$0.50-2.00 per million output tokens at 70% utilization\n- **Hidden costs of self-hosting** — ops on-call, monitoring, model updates, scaling overhead, idle time penalty\n- **Hidden costs of API** — rate limits requiring multi-vendor failover, vendor lock-in, capability drift between versions, data residency\n\n**Typical breakeven (frontier-quality):** 100M–500M tokens\u002Fmonth, depending on model size and acceptable quality tradeoff. Below this, API wins. Above this, run the calculator.\n\n**Run** `ai_cost_economics.py` with workload characteristics for a breakeven point + sensitivity to GPU rates and model size.\n\nSee `references\u002Fai_cost_economics.md` for the full economics model and operational considerations.\n\n### 4. AI Team Org Evolution\n\n**The wrong question:** \"Should we hire an ML engineer or a research scientist?\"\n**The right question:** \"What's the next AI capability we need to ship, and what role unblocks that?\"\n\nStage-to-role map:\n\n| Stage | First AI hire | Then | Then |\n|---|---|---|---|\n| Pre-PMF | Founder + 1 ML-curious engineer playing with prompts | — | — |\n| Series A | **AI engineer** (applied, full-stack; owns prompts\u002Fevals\u002Fdeployment) | Second AI engineer for evals\u002Fquality | — |\n| Series B | AI\u002FML platform engineer (inference, evals, observability) | Third AI engineer for production reliability | Data scientist if model is core IP |\n| Series C | Manager of AI | ML research scientist (only if model IS the product) | AI safety \u002F red team (if customer-facing AI) |\n| Late-stage | Head of AI → CAIO | Multiple research scientists, platform team, safety\u002Fred team | Federated AI leads per business unit |\n\n**Critical distinctions:**\n\n- **AI engineer** ≠ **ML engineer** ≠ **research scientist**\n  - AI engineer: full-stack + prompts + evals + deployment. Most startups need this, not the others.\n  - ML engineer: production deployment, monitoring, retraining infrastructure. Hire after data engineer.\n  - Research scientist: model invention, novel architectures. Only at Series C+ if model is core IP.\n\n**Centralize-vs-embed for AI:** AI starts centralized (one team) and stays there longer than data team, because the surface area is smaller. Embed only when AI is being deployed in 4+ product surfaces.\n\nSee `references\u002Fai_team_org_evolution.md`.\n\n## Workflows\n\n### Workflow 1: Model Selection Decision (1 hour)\n**Goal:** Decide whether a specific use case should use API, fine-tune, or build.\n\n```bash\n# 1. Define use_case.json (volume, latency, accuracy, team size, budget)\npython scripts\u002Fmodel_buildvsbuy_calculator.py use_case.json\n# 2. Review 3-year TCO + breakeven\n# 3. Cross-check with cs-cfo-advisor on budget commitment\n# 4. Cross-check with cs-cto-advisor on engineering capacity (esp. for fine-tune)\n# 5. Log via \u002Fcs:decide; consider \u002Fcs:freeze 60 on multi-year vendor commitment\n```\n\n### Workflow 2: AI Risk Classification (2-4 hours)\n**Goal:** Classify a use case under EU AI Act + US state laws, identify required controls.\n\n```bash\n# 1. Define use_case.json (decisions affected, users, geography, sector)\npython scripts\u002Fai_risk_classifier.py use_case.json\n# 2. For HIGH-RISK: budget conformity assessment + registration\n# 3. For LIMITED-RISK: implement transparency requirements\n# 4. Cross-check with cs-general-counsel-advisor on contractual implications\n# 5. Cross-check with cs-ciso-advisor on technical safeguards\n# 6. Log via \u002Fcs:decide\n```\n\n### Workflow 3: API-to-Self-Hosted Breakeven (1 day)\n**Goal:** Decide when (and whether) to migrate from API to self-hosted inference.\n\n```bash\n# 1. Build workload.json (tokens\u002Fday, model size, latency, quality tolerance)\npython scripts\u002Fai_cost_economics.py workload.json\n# 2. Run sensitivity scenarios (low\u002Fmid\u002Fhigh GPU rates)\n# 3. Estimate migration cost (engineering time + risk)\n# 4. Cross-check with cs-cfo-advisor on capex commitment\n# 5. Cross-check with cs-cto-advisor on platform readiness\n# 6. Log via \u002Fcs:decide; pair with \u002Fcs:freeze if signing GPU commitment\n```\n\n### Workflow 4: AI Team Roadmap (1 week)\n**Goal:** Sequence next 18 months of AI hires aligned to capabilities to ship.\n\n1. List top 5 AI capabilities the product needs in 12 months\n2. Map each capability to the role that ships it (see `ai_team_org_evolution.md`)\n3. Sequence hires (one role at a time, ramp before next)\n4. Cross-check with cs-chro-advisor on comp + leveling\n5. Identify the centralize-vs-embed trigger\n\n## Output Standards\n\n```\n**Bottom Line:** [one sentence — decision and rationale]\n**The Decision:** [one of: model selection | risk classification | economics | next hire]\n**The Evidence:** [numbers from the tool, not adjectives]\n**How to Act:** [3 concrete next steps]\n**Your Decision:** [the call only the founder can make]\n```\n\n## Adjacent Skills\n\n- `c-level-advisor\u002Fskills\u002Fchief-data-officer-advisor\u002F` — Training data rights, data product strategy (chains directly to model decisions)\n- `c-level-advisor\u002Fskills\u002Fcto-advisor\u002F` — Architecture capacity, scaling cliffs (esp. for self-hosted inference)\n- `c-level-advisor\u002Fskills\u002Fciso-advisor\u002F` — Threat modeling for AI (prompt injection, jailbreak, training data poisoning)\n- `c-level-advisor\u002Fskills\u002Fgeneral-counsel-advisor\u002F` — AI contracts (vendor liability, output ownership, training-data licensing)\n- `c-level-advisor\u002Fskills\u002Fcfo-advisor\u002F` — Build-vs-buy TCO math, multi-year vendor commitments\n- `c-level-advisor\u002Fskills\u002Fchro-advisor\u002F` — AI team hiring + comp\n- `engineering\u002Fskills\u002Frag-architect\u002F` — Tactical RAG implementation\n- `engineering\u002Fskills\u002Fagent-designer\u002F` — Tactical agent architecture\n- `engineering\u002Fprompt-governance\u002F` — Tactical prompt management\n- `engineering\u002Fskills\u002Fself-eval\u002F` — Tactical eval infrastructure\n- `engineering\u002Fllm-cost-optimizer\u002F` — Tactical inference cost optimization\n\n## References\n\n- [model_buildvsbuy_strategy.md](references\u002Fmodel_buildvsbuy_strategy.md) — Full decision tree + 3-year TCO components + when each path fails\n- [ai_risk_governance.md](references\u002Fai_risk_governance.md) — EU AI Act + NIST AI RMF + US state patchwork + industry overlays + governance program\n- [ai_cost_economics.md](references\u002Fai_cost_economics.md) — API pricing 2026 + GPU rental economics + utilization realities + migration cost\n- [ai_team_org_evolution.md](references\u002Fai_team_org_evolution.md) — Stage-to-role map + role definitions (AI engineer ≠ ML engineer ≠ scientist) + anti-patterns\n\n---\n\n**Version:** 1.0.0\n**Status:** Production Ready\n**Disclaimer:** AI regulation is evolving rapidly. This skill surfaces decisions and tradeoffs as of 2026 but cannot replace qualified AI counsel for binding compliance decisions, especially under EU AI Act conformity assessments.\n","# Chief AI Officer Advisor\n\nStrategic AI leadership for startup CAIOs and founders without one. **Four decisions, no AI hype:**\n\n1. **Should we use an API, fine-tune, or build our own?** — model build-vs-buy with 3-year TCO\n2. **Is this AI use case high-risk under regulation, and how do we govern it?** — EU AI Act + NIST AI RMF + US state patchwork\n3. **When do we switch from API to self-hosted, and at what cost?** — token economics with breakeven analysis\n4. **What AI role do we hire next?** — stage-to-role map (AI engineer ≠ ML engineer ≠ research scientist)\n\nThis skill does **not** cover tactical AI\u002FML engineering. For RAG implementation, agent design, prompt engineering, eval infrastructure, model deployment, or cost optimization, see `engineering\u002Frag-architect\u002F`, `engineering\u002Fagent-designer\u002F`, `engineering\u002Fprompt-governance\u002F`, `engineering\u002Fself-eval\u002F`, `engineering\u002Fllm-cost-optimizer\u002F`.\n\n## Keywords\n\nCAIO, chief AI officer, AI strategy, model selection, foundation model, fine-tuning, RLHF, DPO, LoRA, QLoRA, build vs buy, AI build-vs-buy, model risk tier, EU AI Act, AI Act Article 6, Article 9, Article 10, Annex III, prohibited AI, high-risk AI, NIST AI RMF, AI risk management framework, NYC Local Law 144, Colorado SB 21-169, Illinois HB 53, model card, eval set, eval harness, hallucination rate, jailbreak risk, prompt injection, AI red team, AI safety, alignment, model lifecycle, model registry, API-to-self-hosted breakeven, GPU economics, A100, H100, inference cost, fine-tuning cost, AI team, AI engineer, ML engineer, research scientist, MLOps, AI platform\n\n## Quick Start\n\n```bash\n# Decision A: API vs fine-tune vs build\npython scripts\u002Fmodel_buildvsbuy_calculator.py                          # embedded customer-support sample\npython scripts\u002Fmodel_buildvsbuy_calculator.py path\u002Fto\u002Fuse_case.json\n\n# Decision B: Risk classification under EU AI Act + US state laws\npython scripts\u002Fai_risk_classifier.py                                   # embedded hiring-AI sample\npython scripts\u002Fai_risk_classifier.py path\u002Fto\u002Fuse_case.json\n\n# Decision C: API vs self-hosted economics\npython scripts\u002Fai_cost_economics.py                                    # embedded 5M tokens\u002Fday sample\npython scripts\u002Fai_cost_economics.py path\u002Fto\u002Fworkload.json\n```\n\n## Key Questions (ask these first)\n\n- **What does this AI need to be good at, and how would you measure it?** (If no eval set, no ship.)\n- **What's the SLO on hallucination \u002F error rate?** (Without one, \"AI quality\" is a vibe.)\n- **What happens when the model is wrong?** (Fallback behavior, human-in-the-loop, blast radius.)\n- **What's the risk tier under EU AI Act, and is conformity assessment required?** (Determines product launch timeline.)\n- **At what monthly token volume does self-hosting beat API?** (Almost never below 100M tokens\u002Fmonth at frontier quality.)\n- **Are we hiring an AI engineer or an ML research scientist?** (Different jobs; founders confuse them.)\n\n## Core Responsibilities\n\n### 1. Model Build-vs-Buy\n\nThe decision is not \"use AI or not\" — it's **API vs fine-tune vs in-house** for each use case. Each path has a different TCO curve, latency profile, and capability ceiling.\n\n**Default path: API (frontier model)**\n- Use when: well-served by frontier (Claude, GPT, Gemini), QPS \u003C 100, latency budget > 1s, cost \u003C $50K\u002Fmonth\n- Why: frontier APIs are 10-100x more capable than what most teams can fine-tune in-house\n- Failure mode: API rate limits at scale, vendor lock-in, capability drift between model versions\n\n**Fine-tune a smaller model**\n- Use when: domain-specific behavior the API can't be prompted into (medical coding, legal redlining), high volume reducing API cost, latency budget \u003C 500ms, specific style\u002Fformat consistency required\n- Approaches: full fine-tune (rare), LoRA\u002FQLoRA (common), RLHF\u002FDPO (when alignment matters)\n- Failure mode: fine-tuned model lags frontier capability within 6-12 months; ongoing retraining cost\n\n**Build from scratch \u002F pre-train**\n- Use when: almost never. You're a foundation-model company, OR you have a unique data corpus, $50M+ funding, and 18+ month patience.\n- Failure mode: by the time you ship, frontier models have caught up and your sunk cost is unrecoverable\n\n**Run** `model_buildvsbuy_calculator.py` for a use-case-specific recommendation with 3-year TCO. See `references\u002Fmodel_buildvsbuy_strategy.md` for full decision tree.\n\n### 2. AI Risk Classification & Governance\n\nThe 2026 question every founder is facing: **does this AI use case trigger high-risk regulatory obligations?**\n\n**EU AI Act (in force 2026) tiers:**\n\n| Tier | Examples | Obligations |\n|---|---|---|\n| **Prohibited** | Social scoring, real-time biometric surveillance, manipulative AI | Cannot deploy in EU |\n| **High-risk** | Employment screening, credit scoring, education access, critical infrastructure, law enforcement, biometric ID | Conformity assessment, registration, post-market monitoring, transparency, human oversight |\n| **Limited-risk** | Chatbots, deepfakes, emotion recognition | Transparency: user must know they're interacting with AI |\n| **Minimal-risk** | Recommendation systems, spam filters, most B2B SaaS internals | No specific obligations |\n\n**Run** `ai_risk_classifier.py` to classify a use case and get the required-controls list.\n\n**US state patchwork (non-exhaustive):**\n\n- NYC LL 144 — Automated Employment Decision Tools (AEDTs) require annual bias audit + candidate notice\n- Colorado AI Act \u002F SB 21-169 — AI in consumer decisions (credit, insurance, employment, housing)\n- Illinois HB 53 — AI in interview\u002Fhiring\n- California SB 1001 — Bot disclosure\n- Texas TCPA — Biometric identifier capture\n- Federal NIST AI RMF — voluntary; increasingly referenced in contracts\n\n**Industry-specific overlays:**\n\n- Healthcare: FDA AI\u002FML guidance (2023), MDR (EU) for medical-device AI, 510(k) pathway for AI\u002FML-enabled medical devices\n- Financial: NYDFS Reg 23, FTC Section 5, ECOA for credit decisions\n- Insurance: NAIC model bulletin, state insurance commissioner rules\n\nSee `references\u002Fai_risk_governance.md` for the full regulatory landscape + governance program checklist.\n\n### 3. AI Cost Economics\n\n**The breakeven question:** at what monthly token volume does self-hosted inference beat API costs?\n\n**Key components:**\n\n- **API cost** — variable, per-token. Frontier models 2026: Claude Sonnet 4.6 ~$3\u002F$15 per M tokens (input\u002Foutput), GPT-4o ~$2.50\u002F$10, Gemini 2.5 ~$1.25\u002F$5\n- **Self-hosted cost** — fixed (GPU commitment) + variable (electricity). H100 spot ~$2-5\u002Fhour, A100 spot ~$1-3\u002Fhour. Llama 3.1 70B \u002F Qwen 2.5 72B: ~$0.50-2.00 per million output tokens at 70% utilization\n- **Hidden costs of self-hosting** — ops on-call, monitoring, model updates, scaling overhead, idle time penalty\n- **Hidden costs of API** — rate limits requiring multi-vendor failover, vendor lock-in, capability drift between versions, data residency\n\n**Typical breakeven (frontier-quality):** 100M–500M tokens\u002Fmonth, depending on model size and acceptable quality tradeoff. Below this, API wins. Above this, run the calculator.\n\n**Run** `ai_cost_economics.py` with workload characteristics for a breakeven point + sensitivity to GPU rates and model size.\n\nSee `references\u002Fai_cost_economics.md` for the full economics model and operational considerations.\n\n### 4. AI Team Org Evolution\n\n**The wrong question:** \"Should we hire an ML engineer or a research scientist?\"\n**The right question:** \"What's the next AI capability we need to ship, and what role unblocks that?\"\n\nStage-to-role map:\n\n| Stage | First AI hire | Then | Then |\n|---|---|---|---|\n| Pre-PMF | Founder + 1 ML-curious engineer playing with prompts | — | — |\n| Series A | **AI engineer** (applied, full-stack; owns prompts\u002Fevals\u002Fdeployment) | Second AI engineer for evals\u002Fquality | — |\n| Series B | AI\u002FML platform engineer (inference, evals, observability) | Third AI engineer for production reliability | Data scientist if model is core IP |\n| Series C | Manager of AI | ML research scientist (only if model IS the product) | AI safety \u002F red team (if customer-facing AI) |\n| Late-stage | Head of AI → CAIO | Multiple research scientists, platform team, safety\u002Fred team | Federated AI leads per business unit |\n\n**Critical distinctions:**\n\n- **AI engineer** ≠ **ML engineer** ≠ **research scientist**\n  - AI engineer: full-stack + prompts + evals + deployment. Most startups need this, not the others.\n  - ML engineer: production deployment, monitoring, retraining infrastructure. Hire after data engineer.\n  - Research scientist: model invention, novel architectures. Only at Series C+ if model is core IP.\n\n**Centralize-vs-embed for AI:** AI starts centralized (one team) and stays there longer than data team, because the surface area is smaller. Embed only when AI is being deployed in 4+ product surfaces.\n\nSee `references\u002Fai_team_org_evolution.md`.\n\n## Workflows\n\n### Workflow 1: Model Selection Decision (1 hour)\n**Goal:** Decide whether a specific use case should use API, fine-tune, or build.\n\n```bash\n# 1. Define use_case.json (volume, latency, accuracy, team size, budget)\npython scripts\u002Fmodel_buildvsbuy_calculator.py use_case.json\n# 2. Review 3-year TCO + breakeven\n# 3. Cross-check with cs-cfo-advisor on budget commitment\n# 4. Cross-check with cs-cto-advisor on engineering capacity (esp. for fine-tune)\n# 5. Log via \u002Fcs:decide; consider \u002Fcs:freeze 60 on multi-year vendor commitment\n```\n\n### Workflow 2: AI Risk Classification (2-4 hours)\n**Goal:** Classify a use case under EU AI Act + US state laws, identify required controls.\n\n```bash\n# 1. Define use_case.json (decisions affected, users, geography, sector)\npython scripts\u002Fai_risk_classifier.py use_case.json\n# 2. For HIGH-RISK: budget conformity assessment + registration\n# 3. For LIMITED-RISK: implement transparency requirements\n# 4. Cross-check with cs-general-counsel-advisor on contractual implications\n# 5. Cross-check with cs-ciso-advisor on technical safeguards\n# 6. Log via \u002Fcs:decide\n```\n\n### Workflow 3: API-to-Self-Hosted Breakeven (1 day)\n**Goal:** Decide when (and whether) to migrate from API to self-hosted inference.\n\n```bash\n# 1. Build workload.json (tokens\u002Fday, model size, latency, quality tolerance)\npython scripts\u002Fai_cost_economics.py workload.json\n# 2. Run sensitivity scenarios (low\u002Fmid\u002Fhigh GPU rates)\n# 3. Estimate migration cost (engineering time + risk)\n# 4. Cross-check with cs-cfo-advisor on capex commitment\n# 5. Cross-check with cs-cto-advisor on platform readiness\n# 6. Log via \u002Fcs:decide; pair with \u002Fcs:freeze if signing GPU commitment\n```\n\n### Workflow 4: AI Team Roadmap (1 week)\n**Goal:** Sequence next 18 months of AI hires aligned to capabilities to ship.\n\n1. List top 5 AI capabilities the product needs in 12 months\n2. Map each capability to the role that ships it (see `ai_team_org_evolution.md`)\n3. Sequence hires (one role at a time, ramp before next)\n4. Cross-check with cs-chro-advisor on comp + leveling\n5. Identify the centralize-vs-embed trigger\n\n## Output Standards\n\n```\n**Bottom Line:** [one sentence — decision and rationale]\n**The Decision:** [one of: model selection | risk classification | economics | next hire]\n**The Evidence:** [numbers from the tool, not adjectives]\n**How to Act:** [3 concrete next steps]\n**Your Decision:** [the call only the founder can make]\n```\n\n## Adjacent Skills\n\n- `c-level-advisor\u002Fskills\u002Fchief-data-officer-advisor\u002F` — Training data rights, data product strategy (chains directly to model decisions)\n- `c-level-advisor\u002Fskills\u002Fcto-advisor\u002F` — Architecture capacity, scaling cliffs (esp. for self-hosted inference)\n- `c-level-advisor\u002Fskills\u002Fciso-advisor\u002F` — Threat modeling for AI (prompt injection, jailbreak, training data poisoning)\n- `c-level-advisor\u002Fskills\u002Fgeneral-counsel-advisor\u002F` — AI contracts (vendor liability, output ownership, training-data licensing)\n- `c-level-advisor\u002Fskills\u002Fcfo-advisor\u002F` — Build-vs-buy TCO math, multi-year vendor commitments\n- `c-level-advisor\u002Fskills\u002Fchro-advisor\u002F` — AI team hiring + comp\n- `engineering\u002Fskills\u002Frag-architect\u002F` — Tactical RAG implementation\n- `engineering\u002Fskills\u002Fagent-designer\u002F` — Tactical agent architecture\n- `engineering\u002Fprompt-governance\u002F` — Tactical prompt management\n- `engineering\u002Fskills\u002Fself-eval\u002F` — Tactical eval infrastructure\n- `engineering\u002Fllm-cost-optimizer\u002F` — Tactical inference cost optimization\n\n## References\n\n- [model_buildvsbuy_strategy.md](references\u002Fmodel_buildvsbuy_strategy.md) — Full decision tree + 3-year TCO components + when each path fails\n- [ai_risk_governance.md](references\u002Fai_risk_governance.md) — EU AI Act + NIST AI RMF + US state patchwork + industry overlays + governance program\n- [ai_cost_economics.md](references\u002Fai_cost_economics.md) — API pricing 2026 + GPU rental economics + utilization realities + migration cost\n- [ai_team_org_evolution.md](references\u002Fai_team_org_evolution.md) — Stage-to-role map + role definitions (AI engineer ≠ ML engineer ≠ scientist) + anti-patterns\n\n---\n\n**Version:** 1.0.0\n**Status:** Production Ready\n**Disclaimer:** AI regulation is evolving rapidly. This skill surfaces decisions and tradeoffs as of 2026 but cannot replace qualified AI counsel for binding compliance decisions, especially under EU AI Act conformity assessments.\n",{"name":5,"license":14,"metadata":43,"description":7},{"author":44,"domain":45,"updated":46,"version":47,"category":48,"frameworks":49,"python-tools":50},"Alireza Rezvani","chief-ai-officer-leadership",1778544000000,"1.0.0","c-level","model-buildvsbuy, ai-risk-governance, ai-economics, ai-team-org","model_buildvsbuy_calculator.py, ai_risk_classifier.py, ai_cost_economics.py","main","068c2486ccc0f8d200a3db4d21ec137064eda3b4","skills\u002Fpublic-02-claude-skills-main-gemini-skills-chief-ai-officer-advisor-skill-610b72d6","https:\u002F\u002Fgithub.com\u002FJiayiV\u002Fquestion_mixing","https:\u002F\u002Fgithub.com\u002FJiayiV","2026-08-25 16:30:52",{"source":8,"skillName":5,"commands":58,"prompt":68,"zipAvailable":69},[59,62,65],{"tool":60,"command":61},"npx","npx skills add JiayiV\u002Fquestion_mixing --skill chief-ai-officer-advisor",{"tool":63,"command":64},"bunx","bunx skills add JiayiV\u002Fquestion_mixing --skill chief-ai-officer-advisor",{"tool":66,"command":67},"pnpm","pnpm dlx skills add JiayiV\u002Fquestion_mixing --skill chief-ai-officer-advisor","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FJiayiV\u002Fquestion_mixing\u002Fblob\u002Fmain\u002Fskills\u002Fpublic-02-claude-skills-main-gemini-skills-chief-ai-officer-advisor-skill-610b72d6\u002FSKILL.md",true,[],[72,98,126,159,186,221],{"skillNo":73,"name":74,"nameZh":6,"description":75,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":76,"categories":77,"keywords":78,"publishedAt":38},"01M6NN9PJS9NHWPDG4QP93BP8F","webmcpify","Make a web app agent-ready — propose a WebMCP tool manifest, integrate, verify in a real browser, heal; unrelated code stays untouched. Use for \"webmcpify\", \"add WebMCP\", or \"expose app actions to AI agents\".","https:\u002F\u002Fgithub.com\u002FJiayiV\u002Fquestion_mixing\u002Fblob\u002Fmain\u002Fskills\u002Fwebmcpify\u002FSKILL.md",[],[74,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95,96,97],"github","ai agents","web","notes","code","browser","https","actions","manifest","inventory","verify","tool","use","ready","unrelated","app","scope","for","webmcp",{"skillNo":99,"name":100,"nameZh":6,"description":101,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":102,"licenseFileUrl":6,"sourceHtmlUrl":103,"categories":104,"keywords":105,"publishedAt":38},"01M6NN6KZE5RBN6NGHN2WXV1RE","vardoger-analyze","Use when the user asks to personalize the GitHub Copilot CLI assistant, adapt Copilot to their style, use vardoger, or analyze their Copilot CLI conversation history. Reads the local session directory at `~\u002F.copilot\u002Fsession-state\u002F`, extracts recurring preferences and conventions, and writes a fenced personalization block into `~\u002F.copilot\u002Fcopilot-instructions.md`. Runs entirely on the user's machine via the local `vardoger` CLI (`pipx install vardoger`); no network calls and no uploads. Triggers: 'personalize my copilot', 'analyze my copilot history', 'tailor copilot to me', 'run vardoger', 'update my copilot instructions from history', 'make copilot learn my style'.","Apache-2.0","https:\u002F\u002Fgithub.com\u002FJiayiV\u002Fquestion_mixing\u002Fblob\u002Fmain\u002Fskills\u002Fvardoger-analyze\u002FSKILL.md",[],[106,107,108,109,110,111,112,113,114,115,116,117,118,119,120,121,122,123,124,125],"vardoger analyze","analyze","directory","cli","github copilot","copilot","run","update","learn","network","my style","install","conversation","me","local","github copilot cli","pipx","my","at","copilot instructions",{"skillNo":127,"name":128,"nameZh":6,"description":129,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":130,"categories":131,"keywords":140,"publishedAt":38},"01M6NN3CZGQP3XAJ04BM1NNYGT","transloadit-media-processing","Process media files (video, audio, images, documents) using Transloadit. Use when asked to encode video to HLS\u002FMP4, generate thumbnails, resize or watermark images, extract audio, concatenate clips, add subtitles, OCR documents, or run any media processing pipeline. Covers 86+ processing robots for file transformation at scale.","https:\u002F\u002Fgithub.com\u002FJiayiV\u002Fquestion_mixing\u002Fblob\u002Fmain\u002Fskills\u002Ftransloadit-media-processing\u002FSKILL.md",[132,137],{"categoryNo":133,"categoryName":134,"taxonomyType":135,"taxonomyTypeDesc":136},"01JBSEEDCAT000000000000016","图像与多媒体","SCENARIO","使用场景",{"categoryNo":138,"categoryName":139,"taxonomyType":135,"taxonomyTypeDesc":136},"01JBSEEDCAT000000000000023","Skill与智能体开发",[141,142,143,144,145,146,147,148,149,150,151,152,153,91,154,155,156,157,96,158],"transloadit media processing","Media","media processing","processing","MCP","mcp server","documents","subtitles","pipeline","ocr","Video","Audio","mp4","images","files","transformation","free","hls",{"skillNo":160,"name":161,"nameZh":6,"description":162,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":163,"categories":164,"keywords":168,"publishedAt":38},"01M6NN28E92F9GH4WFYQ2H1Y15","terraform-azurerm-set-diff-analyzer","Analyze Terraform plan JSON output for AzureRM Provider to distinguish between false-positive diffs (order-only changes in Set-type attributes) and actual resource changes. Use when reviewing terraform plan output for Azure resources like Application Gateway, Load Balancer, Firewall, Front Door, NSG, and other resources with Set-type attributes that cause spurious diffs due to internal ordering changes.","https:\u002F\u002Fgithub.com\u002FJiayiV\u002Fquestion_mixing\u002Fblob\u002Fmain\u002Fskills\u002Fterraform-azurerm-set-diff-analyzer\u002FSKILL.md",[165],{"categoryNo":166,"categoryName":167,"taxonomyType":135,"taxonomyTypeDesc":136},"01JBSEEDCAT000000000000011","运维与部署",[169,170,171,172,173,174,175,176,177,91,31,178,179,180,181,182,183,184,185,96],"terraform","SET","terraform azurerm set diff analyzer","diff","analyzer","diffs","azure","gateway","firewall","resource changes","provider","resources","load balancer","attributes","front","load","balancer",{"skillNo":187,"name":188,"nameZh":6,"description":189,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":190,"categories":191,"keywords":203,"publishedAt":38},"01M6NN1QMH2TDYWTH3XY3N5XPV","technical-job-search","Use this skill when a software engineer asks for help with job search tasks: parsing or analyzing a job description, tailoring a CV\u002Fresume, writing a cover letter, evaluating a job offer, or drafting a post-interview follow-up email. Do not activate for general career advice unrelated to an active job search action.","https:\u002F\u002Fgithub.com\u002FJiayiV\u002Fquestion_mixing\u002Fblob\u002Fmain\u002Fskills\u002Ftechnical-job-search\u002FSKILL.md",[192,195,200],{"categoryNo":193,"categoryName":194,"taxonomyType":135,"taxonomyTypeDesc":136},"01JBSEEDCAT000000000000018","写作与文案",{"categoryNo":196,"categoryName":197,"taxonomyType":198,"taxonomyTypeDesc":199},"01JBSEEDCAT000000000000036","内容创作者","PROFESSION","适用职业",{"categoryNo":201,"categoryName":202,"taxonomyType":198,"taxonomyTypeDesc":199},"01JBSEEDCAT000000000000038","人力资源",[204,205,206,207,208,209,210,211,107,212,91,213,214,215,216,217,218,219,96,220],"search","Job Search","technical job search","technical","writing","tasks","email","parsing","interview","up","engineer","general","activate","do","task","cv","letter",{"skillNo":222,"name":223,"nameZh":6,"description":224,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":225,"categories":226,"keywords":227,"publishedAt":38},"01M6NMZFA67XYS1GS09WT42235","steno-mode","Shorthand-first response compression that cuts ~40% of response tokens while preserving technical precision and exact literals. Use when the user says \"steno mode\", \"shorthand mode\", \"compressed responses\", \"token reduction\", \"brief structured output\", or invokes \u002Fsteno. Supports four compression levels: lite, brief, court, machine. Do not trigger for requests needing polished prose such as onboarding\u002Ftutorial content, stakeholder or customer-facing copy, or teaching-focused explanations.","https:\u002F\u002Fgithub.com\u002FJiayiV\u002Fquestion_mixing\u002Fblob\u002Fmain\u002Fskills\u002Fsteno-mode\u002FSKILL.md",[],[228,229,230,231,232,207,233,91,217,234,235,236,237,238,239,240,241,242,96,243],"steno mode","mode","customer","brief","first","tutorial","teaching","tokens","as","trigger","structured output","machine","output","court","requests","response compression"]