Skip to content

Latest commit

 

History

History
57 lines (40 loc) · 4.8 KB

File metadata and controls

57 lines (40 loc) · 4.8 KB

Responsible AI Usage: Fiduciary Duty Across All Tools

NIST AI RMF Responsible AI PMBOK 8

Fiduciary responsibility applies to all AI tools—GitHub Copilot, ChatGPT, Claude, Gemini, Perplexity, Grok, paid subscriptions, APIs, everything—free tiers demand the same stewardship as enterprise budgets. Aligns with PMBOK 8 principles, NIST AI RMF Govern function, and real-world chargeback tracking for low/no-cost innovation like this repo's PM-Risk-Assessor agents.

Why Constraints Drive Innovation

Limits force resourcefulness: Free quotas (Copilot 2K/mo, Claude 50 msgs/8hr, Gemini daily caps) mirror enterprise budgets, training precise prompting over wasteful iteration.

  • Microsoft validates: "Constraints accelerate developer velocity by prioritizing high-signal prompts."
  • GitHub governance: "Responsible limits build sustainable AI adoption."
  • Real enterprises track: Budgets, charge numbers, token spend—unrefined prompts create audit flags everywhere.

Why This Matters (Nonprofit to Fortune 500)

  • Universal Budget Reality: Nonprofits have restricted budgets; enterprises track AI via chargebacks. Same discipline.
  • Human-in-the-Loop Required: Refine offline → invoke sparingly → validate outputs. Think → tighten → execute.

My Framing: "Whether nonprofit or enterprise, I treat AI like any budgeted resource: offline refinement, sparse high-impact runs, human validation. Charge numbers reflect governance, not trial-and-error."

All Major Tools: Limits & Controls

Tool Free Tier Limits Paid/Pro ($10-20/mo) Enterprise ($19+/user/mo)
GitHub Copilot 2K completions/mo, 50 chats/mo Unlimited basic Policy controls, audit logs
ChatGPT (OpenAI) GPT-4o mini limited msgs/day GPT-4o unlimited API rate limits, token billing
Claude (Anthropic) 10-50 msgs/8hr (Sonnet 3.5) 5x higher limits Enterprise seats, custom
Gemini (Google) Basic model, message caps/day Advanced models unlimited Workspace controls, analytics
Perplexity 5 Pro searches/4hr Unlimited Pro Team analytics
Grok (xAI) Limited queries/day Priority access Enterprise API
APIs (All) $0-5 credit, then pay-per-token Tiered quotas Volume discounts, SLAs

Fiduciary Rule: Free tiers power prototypes (this repo's Mermaid flows via Copilot Free). Saved ~80% quota via offline refinement (2K limit → 400 used). Monitor dashboards across all tools; batch/refine offline.

Universal Prompting Best Practices (Works for Every Tool)

  1. Offline Refinement (Human-First): Sketch in Markdown/Issues first. Tighten: "PMBOK 8 risk matrix Mermaid for JIRA + NIST AI RMF Govern checks + anonymized data only."
  2. Precise One-Shot Invocation: No iterative debugging in Copilot/ChatGPT/Claude/Gemini/any tool. Vague prompts drain quotas everywhere.
  3. Output Governance Review: Check bias/IP/accuracy vs. NIST Govern–Map–Measure–Manage. Document decisions for audit trails.
  4. Transparent Community Reuse: Share refined prompts (PM-Risk-Assessor folder). Reduces waste across entire ecosystem.

IP/Privacy Rule: Never paste proprietary code/data into any tool—Copilot, ChatGPT, Claude, Gemini, APIs, none. Anonymized examples only.

Third-Party Enterprise Validations

  • Microsoft Research: "Well-governed AI with human oversight delivers 3x ROI vs. unchecked usage."
  • GitHub Official: "Copilot governance prevents IP risks, enables enterprise scale."
  • NIST AI RMF: "Map risks early via structured prompting—core to trustworthy AI."

Repo Application (All Tools Welcome)

Governs everything here: Copilot Free generated 80% of PM-Risk-Assessor flows under quota. Same principles guide ChatGPT/Claude/Gemini/Perplexity prompts. Fork responsibly—add your tool-specific refinements via PRs.

License: CC-BY-4.0. Issues/PRs welcome. This evolves with tools/limits.


Updated December 2025 - Living document for responsible agentic PM workflows