An open-source Agent Skill that helps you define what "Agents as a Service" means for your specific business.
At GTC 2026, Jensen Huang stated that every company needs an Agents as a Service strategy. The term is gaining traction, but no clear operational definition exists yet. What distinguishes AaaS from SaaS with AI features? From managed services? From API-based AI? These are open questions.
This skill is an attempt to explore that uncharted territory. It starts from your business reality — not from abstract definitions — and derives what AaaS means specifically for you.
The skill runs a working session that:
- Understands your business — you provide context (website, docs, description), the agent restates what you do and how you deliver value. You confirm or correct.
- Identifies your business category — anchors the research in the right domain and scale.
- Researches agentic solutions across three layers — NVIDIA platforms, complementary vendors, and open source. Prioritizes technical documentation and GitHub repos over marketing material. Matches solutions to the company's size, resources, and industry.
- Proposes three user stories — one human, one agent-to-agent, one hybrid. Each includes current state, future state, and a concrete number.
- Derives a custom AaaS definition — from the approved ideas, not the other way around. Genus, differentia, boundaries, observable criteria.
- Maps uncertainty — what you know (with evidence), what you assume (and what would break it), what you can't define yet.
The skill checks outputs for logic degradation — smart-sounding language where the conclusion doesn't follow from the premises. If an output could describe any company, it fails and gets rewritten.
The human is the quality gate. The agent does the work.
Session output (MD + JSON) is structured so it can serve as input for subsequent work — hypothesis formation, data architecture, prototyping.
git clone https://github.com/rafaelknuthLLM/aaas-definition.git
cp -r aaas-definition ~/.claude/skills//aaas-definition <your business, website, or product>
Manually invoked only. Does not trigger automatically.
Follows the Open Agent Skills standard. Works with Claude (Code, Desktop, Cowork), OpenAI Codex, Gemini CLI, GitHub Copilot, Cursor, and local LLMs via instavm/open-skills.
aaas-definition does one thing: derive a business-specific AaaS definition grounded in concrete examples. The following are designed but not yet built as standalone skills. Each one's output feeds the next.
aaas-architecture — Define entities, relationships, and state transitions underlying your AaaS offering. If you can't define the data structure, you don't yet understand the problem.
aaas-hypothesis — Turn strategic claims into falsifiable hypotheses with measurable predictions, kill conditions, and timelines.
aaas-patterns — Questions drawn from Linux, cloud, and SaaS transitions — organized by strategic concern. Checks whether you're walking into structurally repeatable mistakes.
aaas-principles — Decision filters grounded in engineering and systems thinking (Hamilton, Torvalds, Goldratt, McIlroy/Kim), under the umbrella of Engineering Thinking as defined by the Royal Academy of Engineering.
aaas-synthesis — Cross-phase consistency check, gap map, readiness assessment, and one concrete next move.
A brief note on choices made in building this skill, for anyone considering forking or modifying it.
No persona. The skill does not role-play ("you are a strategy advisor"). It enforces rigor through structure — logic checks, source tagging, required outputs.
Flexible flow, strict standards. The skill does not enforce a rigid sequence. Users can provide context in any order. The skill steers toward the target (a grounded, business-specific definition) while enforcing quality standards (logic check, uncertainty mapping, specificity) throughout.
Bottom-up, not top-down. The definition is derived from specific examples and user stories, not imposed as an abstract framework. This reflects the consistent finding that starting from business reality produces better results than starting from theory.
Source tagging. Every claim is tagged [DOC], [WEB], or [UNSOURCED] so you can trace where things came from.
Three user stories, three perspectives. Every session produces exactly three user stories: human (direct interaction with agent), agent (agent-to-agent, like Stripe serving SaaS via API), and hybrid (human oversees, agent executes). This forces the user to think across all three counterparty types.
User-provided context only. The skill uses only the information the user provides about their business. Web search is for researching agentic solutions and the business category — not for gathering additional company information.
Source quality over quantity. Technical docs, GitHub repos, and developer guides are prioritized. Marketing material is flagged as [WEB-MKTG]. Solutions are matched to the company's scale — no hyperscaler recommendations for small firms, no undersized solutions for enterprises.
Spatial check. Before presenting output, the skill checks for redundancies, unnecessary detail, jargon that adds no precision, and disproportionate sections.
If you use this skill and find it useful, or find it lacking, open an issue. If you fork it and take it in a different direction, we'd like to hear about it.
Apache 2.0
Developed March 2026 as part of ongoing work exploring AaaS strategy. Built by Rafael Knuth.