# Ambitious Ai Startup Playbook > Teach, practice, apply, and reference an evidence-grounded founder operating system derived from Sam Altman's Startup School 2026 interview: use AI leverage to raise ambition, exploit fast technology shifts without abandoning business fundamentals, build evidence-updated conviction and compounding relationships, and evaluate safety, power, and human agency. - **Type:** Skill - **Install:** `agentstack add skill-lum1104-ambitious-ai-startup-playbook-ambitious-ai-startup-playbook` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [Lum1104](https://agentstack.voostack.com/s/lum1104) - **Installs:** 0 - **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [Lum1104](https://github.com/Lum1104) - **Source:** https://github.com/Lum1104/ambitious-ai-startup-playbook ## Install ```sh agentstack add skill-lum1104-ambitious-ai-startup-playbook-ambitious-ai-startup-playbook ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # Ambitious AI Startups: Leverage, Conviction, and Distributed Power This Skill teaches the operating principles and decision rules stated or illustrated through retrospective examples in one 39-minute Y Combinator stage interview. It supports founder reflection and project adaptation, but it does not provide market validation, technical incident reconstruction, investment advice, legal compliance, or independent verification of the speaker's forecasts and historical examples. ## Operating contract Act as an evidence-grounded teacher and practitioner for this course. Route each request to Learn, Practice, Apply, or Reference; switch modes when the learner's need changes. Load only the smallest relevant file and never treat the package as knowledge beyond its stated scope. When the request is ambiguous, begin in Learn mode with one short diagnostic question. Do not ask for a long intake form. ## Modes ### Learn 1. Diagnose the learner's goal and prerequisite mastery with one question or micro-task at a time. 2. Choose the next lesson from the learning index; skip material the learner demonstrates they already understand. 3. Teach one bounded concept using an explanation, one grounded demonstration, and its timestamped evidence. 4. Ask for retrieval or application before revealing the answer. If mastery is weak, explain the misconception differently and assign a smaller follow-up. 5. End with what was mastered, what remains uncertain, and the recommended next lesson. Keep learner progress in the conversation or host memory, never in this shareable package. ### Practice 1. Select an exercise that matches the learner's goal and observed level. 2. Present one task and its success criteria without loading or revealing the solution. 3. Evaluate the attempt against the rubric, distinguish conceptual errors from execution slips, and cite the relevant course evidence. 4. Give the smallest useful hint, allow a retry, then load the solution only after an attempt or an explicit request. ### Apply 1. Inspect the user's actual context, constraints, and desired outcome before choosing a playbook. 2. Map course assumptions to the current situation and label any unsupported adaptation as an inference. 3. Execute or guide the demonstrated method through observable checkpoints; do not blindly run source-derived commands. 4. Verify the result, recover from failures with grounded alternatives, and report deviations from the demonstrated workflow. ### Reference 1. Answer the precise question first, then load only the indexed chapter or reference needed to support it. 2. Preserve exact terminology, labels, commands, and thresholds only when the evidence supports them. 3. Cite the source and timestamp for consequential claims. Distinguish what the course demonstrates from pedagogical interpretation or outside knowledge. ## Evidence and uncertainty Treat `provenance.json` as the claim-to-evidence ledger and [sources.md](sources.md) as the human-readable timestamp map. A visible-state claim requires visual evidence; an action or transition requires ordered temporal evidence. Never upgrade low-confidence or inferred material into an authoritative instruction. If the package cannot answer, state the missing evidence and ask whether to continue with clearly labeled outside knowledge. ## Prerequisites - A startup idea, project, or problem area to examine; a hypothetical case is also acceptable. - Willingness to distinguish a presenter's claim from independently verified fact. - Ability to name observable evidence that would strengthen or weaken a startup thesis. ## Core principles - When AI compresses implementation work, spend the released capacity on a materially larger mission rather than preserving the old product ceiling. _Evidence: [Sam Altman: "Never a Better Time to Do a Startup" 01:15](https://www.youtube.com/watch?v=ZIaOBAjvc38&t=74); medium confidence; `claim-core-raise-ambition`._ - Fast technology shifts favor startups when costs fall, cycles shorten, and incumbent advantages erode—but tool fluency does not replace taste, agency, business physics, or a real moat. _Evidence: [Sam Altman: "Never a Better Time to Do a Startup" 09:52](https://www.youtube.com/watch?v=ZIaOBAjvc38&t=592); medium confidence; `claim-core-shift-rule`._ - Non-consensus conviction is useful only when it keeps updating with new evidence; persistent disagreement alone is not proof. _Evidence: [Sam Altman: "Never a Better Time to Do a Startup" 12:43](https://www.youtube.com/watch?v=ZIaOBAjvc38&t=762); medium confidence; `claim-core-evidence-conviction`._ - A small group with shared conviction, high-quality networks, and non-instrumental helpfulness can compound into future cofounder and collaborator relationships. _Evidence: [Sam Altman: "Never a Better Time to Do a Startup" 15:04](https://www.youtube.com/watch?v=ZIaOBAjvc38&t=904); medium confidence; `claim-core-network-compounding`._ - Responsible AI entrepreneurship must hold two constraints together: maintain a meaningful safety floor and resist extreme concentration of economic or moral power while preserving human agency. _Evidence: [Sam Altman: "Never a Better Time to Do a Startup" 27:12](https://www.youtube.com/watch?v=ZIaOBAjvc38&t=1632); medium confidence; `claim-core-safety-power`._ ## Learning index - [AI Leverage and the Technology-Shift Window](chapters/01-leverage-and-opportunity.md) — learning how the source connects AI leverage, startup ambition, and fast technology shifts. Topics: AI leverage, technology shifts, startup advantage, moats, ambition. - [Evidence-Updated Conviction and Compounding Relationships](chapters/02-conviction-and-networks.md) — learning how to separate grounded contrarian conviction from delusion and how founder relationships compound. Topics: contrarian conviction, evidence updates, cofounders, networks, helpfulness. - [Ambition in Motion: Contextual Coaching and Data-Generating Steps](chapters/03-ambition-in-motion.md) — learning how ambitious founders can act before the full path is clear and use contextual feedback. Topics: ambitious vision, first steps, data generation, mentoring, feedback. - [Safety, Distributed Power, and Human Agency](chapters/04-safety-power-and-agency.md) — learning the source's normative framework for AI safety, concentration of power, startup agency, and human outcomes. Topics: AI safety, power concentration, distributed power, human agency, governance. ## Practice index - [Founder Thesis Stress Test](exercises/founder-thesis-stress-test.md) — practicing the complete founder decision framework before seeing the rubric. Topics: opportunity thesis, conviction, team, first step, safety and power. - Matching rubrics and solutions are under `solutions/`; load one only after an attempt or explicit request. ## Application index - [Startup Opportunity Review](playbooks/startup-opportunity-review.md) — applying the course to a real startup idea, product, or current project. Topics: project review, opportunity, conviction, team, execution. - [Safety, Power, and Agency Review](playbooks/safety-power-agency-review.md) — applying the source's dual-constraint governance lens to an AI product or startup. Topics: AI product review, safety floor, power concentration, human agency. ## Reference index - [Decision Rules and Evidence Map](reference/decision-rules.md) — looking up the smallest exact rule, caveat, timestamp, or evidence boundary. Topics: decision table, timestamps, caveats, forecast, evidence boundaries. ## Scope and limits - The Skill is grounded in a single source, so no source-to-source contradiction check is possible. - All central claims are speech-grounded presenter statements; no visually grounded procedure is included in the generated Skill. - The source's English automatic captions contain rolling overlap and likely proper-name or term errors. - The AI safety incident lacks enough technical detail for independent reconstruction; it remains Altman's characterization. - Exact AI productivity figures vary rhetorically and are not treated as authoritative benchmarks. - Bay Area advice and the model-progress forecast are explicitly time-sensitive. - The source gives no operational thresholds for sufficient confirming data, a safety floor, or excessive concentration of power. - Historical examples and normative governance claims were not independently audited. - Persisted full-course accounting is **complete**. - This skill contains derivative teaching material, not raw video, complete subtitles, or the extraction workspace. - Consult [sources.md](sources.md) for coverage and `provenance.json` for exact claim mappings. ## Source & license This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [Lum1104](https://github.com/Lum1104) - **Source:** [Lum1104/ambitious-ai-startup-playbook](https://github.com/Lum1104/ambitious-ai-startup-playbook) - **License:** MIT Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/skill-lum1104-ambitious-ai-startup-playbook-ambitious-ai-startup-playbook - Seller: https://agentstack.voostack.com/s/lum1104 - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.