# SKILL.md: AIC, the AI Corporation **Skill name:** AIC, AI Corporation Guidelines, Strategies, and Executions **Version:** 1.0 **Primary purpose:** Use this skill to spin up, evaluate, operate, and scale an AI-native company around a core idea. **Core formula:** Purpose + Agents + Interfaces + Dashboards + Human Review + Community + Governance + Unit Economics. --- ## 0. The Job of This Skill This skill turns a raw idea into an AI-native operating company. Use it when the user wants to: - Turn a business idea into a real company structure. - Build an AI-first startup, agency, platform, product, or service. - Convert an existing company into an AI-operated company. - Create an agentic operating model around a customer problem. - Design the team, agents, workflows, dashboards, governance, and launch plan. - Create a repeatable company-building playbook based on current AI operating research. The goal is not to write another brainstorm document. The goal is to define the company as an executable system. Every output should move from idea to operating model. --- ## 1. Operating Philosophy ### 1.1 The AIC thesis An AI Corporation is not a normal company that uses AI tools. An AI Corporation is a lean human core surrounded by agents, workflows, dashboards, human reviewers, community inputs, external talent, leveraged assets, and governed execution systems. The AIC model assumes that the next company is: - Small at the core. - Large at the edge. - Smart in the middle. - Fast in experimentation. - Careful in governance. - Relentless about unit economics. - Built around a purpose strong enough to coordinate humans and machines. ### 1.2 Model foundation The foundational formula: **MTP + SCALE + IDEAS = Exponential Organization** Where: - **MTP:** Massive Transformative Purpose. - **SCALE:** Staff on Demand, Community and Crowd, AI and Algorithms, Leveraged Assets, Engagement. - **IDEAS:** Interfaces, Dashboards, Experimentation, Autonomy, Social Technologies. AIC extends that model for the agentic AI era: **AIC = MTP Constitution + Agent Fleet + Workflow Interfaces + Live Dashboards + Human Review + Community Loops + Governance + FinOps** ### 1.3 The AIC upgrade The old organization was built for scarcity. The exponential organization is built for abundance. AIC updates the source of abundance. The old abundance was: - Cloud. - Mobile. - Platforms. - APIs. - Gig labor. - Social networks. - Crowds. - Shared assets. The new abundance is: - On-demand intelligence. - Tool-using AI agents. - Human specialist clouds. - Automated research and execution. - Machine-readable knowledge systems. - Real-time dashboards. - AI-assisted experimentation. - AI-governed workflows. - Low-cost creative, analytical, and operational capacity. The scarce resource is no longer information. The scarce resource is trusted coordination. --- ## 2. Hard Rules ### 2.1 Do not overbuild Do not recommend building a full platform before proving the workflow. Default path: 1. Manual service. 2. AI-assisted service. 3. Repeatable workflow. 4. Internal tool. 5. Client dashboard. 6. Productized platform. ### 2.2 Do not sell vague AI Do not define the company as “AI-powered” without specifying: - What job it performs. - For whom. - With what inputs. - Through what workflow. - With what human review. - At what cost. - With what measurable outcome. ### 2.3 Do not remove human judgment where trust matters Human review is required for: - Legal, medical, financial, or regulated advice. - Client-facing strategic recommendations. - Public claims. - Brand-sensitive output. - Safety-sensitive output. - High-cost decisions. - Hiring, firing, compensation, credit, insurance, housing, or other consequential decisions. ### 2.4 Do not invent market proof If a claim needs current evidence, research it. Mark all assumptions as assumptions until verified. ### 2.5 Do not ignore unit economics Every AI workflow must eventually answer: - Cost per run. - Cost per accepted output. - Cost per client. - Gross margin. - Human review time. - Error rate. - Rework rate. - Time saved. - Revenue generated or protected. ### 2.6 Do not confuse autonomy with chaos Autonomy must be bounded by: - Purpose. - Roles. - Authority limits. - Escalation rules. - Audit logs. - Dashboard visibility. - Human approval gates. - Kill switches. ### 2.7 Do not let agents run blind Every agent needs: - A clear mission. - Allowed tools. - Forbidden actions. - Required inputs. - Required outputs. - Review requirements. - Evidence requirements. - Memory rules. - Cost limits. - Failure handling. --- ## 3. Required Inputs When possible, gather these inputs before building the AIC plan. Do not stall if some are missing. Make grounded assumptions and flag them. ### 3.1 Core idea - What is the idea? - What problem does it solve? - Who has the problem? - Why now? - Why AI-native? - What would be 10x better, faster, cheaper, more trusted, or more accessible? ### 3.2 Customer - Primary customer. - Economic buyer. - Daily user. - Urgency level. - Current alternatives. - Budget owner. - Buying trigger. - Cost of doing nothing. ### 3.3 Outcome - What result does the customer want? - What metric proves the result? - What would make them renew? - What would make them refer? - What would make them switch from current options? ### 3.4 Existing assets - Domain expertise. - Brand credibility. - Audience. - Data. - Workflows. - Client relationships. - Software. - Past projects. - Templates. - Human network. - Capital constraints. ### 3.5 Constraints - Budget. - Timeline. - Technical ability. - Regulatory risk. - Data sensitivity. - Customer access. - Human review availability. - Market skepticism. - Competitive pressure. --- ## 4. Default Output Set When asked to spin up an AIC, produce some or all of these outputs: 1. **AIC Brief** The company concept, customer, problem, promise, wedge, and business model. 2. **MTP Constitution** A purpose statement plus operating principles, allowed behaviors, forbidden behaviors, evidence standards, and brand posture. 3. **Market Sensing Brief** Current signals, customer pains, competitors, substitutes, category dynamics, and timing. 4. **Offer Architecture** Entry offer, core offer, premium offer, productized service, recurring model, and platform path. 5. **Agent Fleet Map** The agent roles needed to operate the company. 6. **Workflow Operating System** Intake, routing, execution, review, delivery, feedback, learning, and reporting. 7. **Dashboard Plan** Metrics, instrumentation, cadence, alerts, and ownership. 8. **Human Review Plan** Which outputs require human approval, who reviews them, and what standards apply. 9. **Governance and Risk Register** Risks, controls, escalation paths, audit logs, privacy boundaries, compliance needs, and kill switches. 10. **Experiment Backlog** Testable assumptions, minimum viable tests, success thresholds, and learning cadence. 11. **Launch Roadmap** 7-day, 30-day, 60-day, and 90-day execution plan. 12. **Scale Plan** What to automate, what to delegate, what to productize, what to dashboard, and what to keep human. --- ## 5. AIC Company-Building Process Use this process as the default path from idea to operating company. --- # Phase 0: Idea Intake and Triage ## Goal Determine whether the core idea deserves an AI Corporation build. ## Actions 1. Restate the idea in one sentence. 2. Identify the customer. 3. Identify the pain. 4. Identify the measurable outcome. 5. Identify why AI changes the economics or experience. 6. Identify the fastest test. 7. Identify the main risk. 8. Decide whether the idea is: - Strong enough to build. - Worth testing first. - Too broad. - Too regulated. - Better as a service than software. - Better as an internal workflow first. ## Output template ```md ## AIC Idea Triage **Core idea:** **Customer:** **Pain:** **Current alternative:** **AI advantage:** **10x possibility:** **Fastest test:** **Main risk:** **Initial verdict:** ``` --- # Phase 1: MTP Constitution ## Goal Create the operating soul of the company. This is the MTP. In AIC terms, it becomes a constitution because agents and humans both need machine-readable behavioral guidance. ## The MTP must be - Massive enough to attract energy. - Transformative enough to matter. - Purposeful enough to guide decisions. - Specific enough to prevent generic fluff. - Operational enough to become policies, prompts, workflows, and review standards. ## MTP Constitution components 1. **Massive Transformative Purpose** 2. **Customer promise** 3. **Enemy or friction to defeat** 4. **Operating principles** 5. **Evidence standards** 6. **Voice and brand standards** 7. **Agent behavior rules** 8. **Human review principles** 9. **Forbidden behaviors** 10. **Decision filters** 11. **Long-term compounding advantage** ## MTP Constitution template ```md # MTP Constitution ## Massive Transformative Purpose [One clear sentence.] ## Customer Promise We help [customer] achieve [outcome] by [mechanism], without [pain or waste]. ## The Enemy We are fighting [waste, friction, opacity, delay, complexity, bad information, bad incentives, etc.]. ## Operating Principles 1. [Principle] 2. [Principle] 3. [Principle] 4. [Principle] 5. [Principle] ## Evidence Standard - Confirmed facts require credible sources or primary data. - Strategic assumptions must be labeled. - AI-generated outputs must be reviewed before client delivery when the output affects reputation, spend, compliance, or business direction. ## Agent Rules Agents may: - [Allowed action] - [Allowed action] Agents may not: - [Forbidden action] - [Forbidden action] ## Human Review Gates Human approval is required before: - [Output or decision type] - [Output or decision type] ## Decision Filters Before we act, ask: 1. Does this serve the MTP? 2. Does it improve the customer outcome? 3. Does it reduce friction or increase trust? 4. Can it be tested cheaply? 5. Can it scale without losing quality? ``` --- # Phase 2: Market Sensing and Timing ## Goal Build a live understanding of why this idea matters now. This replaces static market research with a sensing layer. ## Research requirements For any current market, use current sources. Prioritize: - Customer conversations. - Competitor websites. - Analyst reports. - Product reviews. - Forums and communities. - Job postings. - Investor materials. - Regulatory sources. - Technical documentation. - Pricing pages. - Case studies. - News from the past 12 to 18 months when recency matters. ## Questions to answer 1. What has changed in the market? 2. What pain is increasing? 3. What tools are customers already using? 4. Where are current tools failing? 5. What are competitors overpromising? 6. What is becoming cheaper because of AI? 7. What is becoming more valuable because of trust? 8. What jobs are being created or eliminated? 9. What workflows are still manual? 10. What data is available but underused? 11. What regulation or compliance pressure exists? 12. What wedge is available now that was not available before? ## Output template ```md # Market Sensing Brief ## Category ## Why Now ## Customer Pain Signals ## Current Alternatives ## Competitor Patterns ## Open Gaps ## AI-Enabled Advantage ## Trust or Compliance Constraints ## Best Wedge ## Key Sources ``` --- # Phase 3: Customer, Job, and Wedge ## Goal Narrow the company from a broad idea to a sharp beachhead. ## Required definitions ### Customer The specific person or organization that has the pain. ### Job-to-be-done The progress the customer is trying to make. ### Wedge The small, urgent, valuable use case that gets the company into the market. ### Expansion path The logical next problems to solve after the wedge works. ## Wedge rules A good wedge is: - Painful now. - Easy to explain. - Measurable. - Repeated often. - Expensive to do manually. - Made better by AI. - Not too risky for early automation. - Attached to a buyer with budget. ## Output template ```md # Customer and Wedge ## Primary Customer ## Economic Buyer ## Daily User ## Job-to-be-Done ## Current Workflow ## Failure Points ## Wedge Use Case ## Why This Wedge Wins ## Expansion Path ``` --- # Phase 4: AIC Stack Design ## Goal Define the company as an operating system. ## The AIC Stack ### 1. MTP Constitution The purpose, rules, principles, decision filters, and evidence standards. ### 2. Sensing Layer Market, customer, competitor, regulatory, and technology signals. ### 3. Intake Layer Forms, emails, calls, uploads, APIs, CRM entries, chat, or internal requests. ### 4. Knowledge Layer Company memory, customer data, source library, style guides, SOPs, past work, policies, and structured facts. ### 5. Agent Layer Specialized AI agents with defined roles, tools, authority, and outputs. ### 6. Interface Layer Workflows, routers, APIs, MCP servers, approvals, automations, and handoffs. ### 7. Human Review Layer Editors, strategists, subject experts, QA, legal review when needed, and final approvers. ### 8. Dashboard Layer Live metrics for work, quality, cost, speed, revenue, risk, and learning. ### 9. Experiment Layer Hypotheses, tests, variants, success thresholds, and learning archives. ### 10. Community Layer Customers, contributors, expert network, advisors, creators, testers, and fans. ### 11. Governance Layer Risk controls, permissions, audits, privacy rules, security boundaries, and escalation paths. ### 12. FinOps Layer AI spend, cloud spend, tool spend, human review cost, cost per output, gross margin, and utilization. ## Output template ```md # AIC Stack | Layer | Purpose | First version | Owner | Tooling | Metric | |---|---|---|---|---|---| | MTP Constitution | | | | | | | Sensing | | | | | | | Intake | | | | | | | Knowledge | | | | | | | Agent | | | | | | | Interface | | | | | | | Human Review | | | | | | | Dashboard | | | | | | | Experiment | | | | | | | Community | | | | | | | Governance | | | | | | | FinOps | | | | | | ``` --- # Phase 5: Offer Architecture ## Goal Turn the operating model into something customers can buy. ## Default progression ### 1. Diagnostic offer A fast, low-friction entry product that reveals the customer’s problem and creates trust. Examples: - AI readiness audit. - Growth audit. - Workflow audit. - Brand voice audit. - SEO audit. - Data quality audit. - Agent opportunity scan. ### 2. Implementation offer A done-for-you or done-with-you service that fixes the highest-value problem. Examples: - Build the first workflow. - Deploy the first agent team. - Create the knowledge base. - Set up dashboards. - Automate intake and reporting. ### 3. Managed operating offer A recurring service that runs the system. Examples: - Monthly AI operations management. - Agent monitoring and optimization. - Reporting and experiments. - Continuous content, SEO, support, analysis, or sales ops. ### 4. Platform path Only after repeatable demand exists. Examples: - Client portal. - Workflow library. - Agent marketplace. - Benchmark database. - Self-service dashboard. - Subscription platform. ## Offer evaluation rules A good offer has: - Clear buyer. - Clear pain. - Clear deliverable. - Clear before and after. - Clear timeline. - Clear proof. - Clear pricing logic. - Clear path to recurring revenue. ## Output template ```md # Offer Architecture ## Entry Offer - Name: - Price: - Buyer: - Promise: - Deliverables: - Timeline: - Required inputs: - Human review: - Success metric: ## Core Offer ## Recurring Offer ## Platform Path ``` --- # Phase 6: Agent Fleet Design ## Goal Define the AI workforce. Agents are not random chatbots. They are role-specific operating units with defined authority, tools, memory, and review rules. ## Default agent types ### 1. Strategy Agent Turns goals into plans, tradeoffs, briefs, and roadmaps. ### 2. Research Agent Gathers evidence, sources claims, tracks competitors, and monitors market signals. ### 3. Customer Agent Analyzes customer pain, feedback, personas, journeys, objections, and support patterns. ### 4. Product Agent Defines product specs, feature priorities, MVPs, user stories, and acceptance criteria. ### 5. Workflow Agent Maps operations, handoffs, automations, bottlenecks, and standard operating procedures. ### 6. Data Agent Handles data structure, schemas, data quality, source lineage, and reporting inputs. ### 7. Creative Agent Creates copy, concepts, visual direction, campaign ideas, and brand expression. ### 8. Engineering Agent Drafts technical plans, code scaffolds, API plans, data flows, and integration specs. ### 9. QA Agent Checks outputs against requirements, facts, brand rules, risk rules, and acceptance criteria. ### 10. Governance Agent Checks policy, permissions, sensitive data, compliance risks, audit needs, and escalation triggers. ### 11. Sales Agent Builds outreach, proposals, objections, qualification logic, CRM updates, and follow-up sequences. ### 12. Finance Agent Tracks costs, revenue, margins, forecasts, pricing, usage, and AI spend. ## Agent charter template ```md # Agent Charter ## Agent Name ## Mission ## Primary Inputs ## Primary Outputs ## Tools Allowed ## Data Allowed ## Authority Level - Recommend only - Draft only - Execute with approval - Execute within limits - Execute autonomously ## Forbidden Actions ## Evidence Standard ## Review Requirements ## Escalation Triggers ## Success Metrics ## Failure Modes ## Logging Requirements ``` ## Authority levels ### Level 0: No agency The system only answers questions. ### Level 1: Drafting agency The agent can draft outputs but cannot send, publish, spend, or modify systems. ### Level 2: Recommendation agency The agent can recommend actions and provide reasoning, but humans decide. ### Level 3: Controlled execution The agent can execute bounded tasks after approval. ### Level 4: Limited autonomy The agent can execute low-risk tasks within predefined limits. ### Level 5: Full bounded autonomy The agent can act continuously within a governed domain, with logs, alerts, and kill switches. Default new agents to Level 1 or Level 2 until proven safe. --- # Phase 7: Workflow Operating System ## Goal Design how work moves through the company. AIC companies win by converting messy requests into repeatable workflows. ## Standard workflow pattern 1. Intake. 2. Clarification. 3. Source gathering. 4. Routing. 5. Agent execution. 6. Human review. 7. Revision. 8. Delivery. 9. Feedback. 10. Dashboard update. 11. Learning capture. 12. Template improvement. ## Workflow spec template ```md # Workflow Spec ## Workflow Name ## Customer or Internal User ## Trigger ## Inputs Required ## Intake Method ## Routing Logic ## Agents Involved ## Human Roles Involved ## Tools and Systems ## Output ## Review Gates ## Acceptance Criteria ## Timing Standard ## Cost Standard ## Failure Handling ## Dashboard Metrics ## Learning Capture ``` ## Workflow design rules - Design for the most common path first. - Add exceptions after the main path works. - Make status visible. - Keep handoffs explicit. - Log every important decision. - Make review standards clear. - Capture reusable learning. - Improve the workflow after every run. --- # Phase 8: Knowledge Layer ## Goal Build the shared memory of the company. An AIC without a knowledge layer becomes a pile of disconnected prompts. ## Knowledge types ### 1. Source library Verified sources, research, reports, customer materials, policies, and references. ### 2. Company memory Decisions, strategy, principles, lessons learned, customer preferences, and approved language. ### 3. Workflow memory SOPs, examples, templates, accepted outputs, rejected outputs, and recurring errors. ### 4. Customer memory Customer goals, constraints, brand voice, approvals, preferences, and past work. ### 5. Agent memory Agent instructions, role definitions, tool rules, prompts, examples, and evaluation rubrics. ## Knowledge rules - Store only what should be reused. - Label source quality. - Distinguish fact, assumption, opinion, and decision. - Preserve source links. - Keep client data separated. - Do not mix confidential customer data into general memory. - Make knowledge retrievable by workflow and agent. - Retire stale knowledge. ## Knowledge entry template ```md # Knowledge Entry ## Title ## Type - Fact - Assumption - Decision - Preference - Source - Template - Lesson - Policy ## Summary ## Source or Origin ## Date Added ## Owner ## Applies To ## Confidence ## Expiration or Review Date ## Related Workflows ``` --- # Phase 9: Interfaces ## Goal Make abundance usable. Interfaces are the control plane between external inputs and internal execution. ## Interface types - Customer intake forms. - Internal request forms. - Upload portals. - APIs. - MCP servers. - Webhooks. - CRM integrations. - Project management workflows. - Agent tool calls. - Approval queues. - Notification systems. - Delivery portals. ## Interface design principles - Self-service where possible. - Standardized inputs. - Clear required fields. - Automated routing. - Human escalation when ambiguity is high. - Status visibility. - Error handling. - Permission checks. - Auditability. ## Interface template ```md # Interface Spec ## Interface Name ## User ## Purpose ## Inputs ## Validation Rules ## Routing Rules ## Connected Systems ## Permissions ## Human Escalation Rules ## Output ## Error States ## Metrics ``` --- # Phase 10: Dashboards ## Goal Make the company visible to itself. Dashboards are not decoration. They are the management system. ## Dashboard categories ### 1. Company dashboard - Revenue. - Pipeline. - Cash. - Gross margin. - Active clients. - Retention. - Delivery status. - Customer satisfaction. ### 2. Workflow dashboard - Requests received. - Requests completed. - Cycle time. - Bottlenecks. - Error rate. - Rework rate. - Human review time. - SLA performance. ### 3. Agent dashboard - Agent runs. - Output acceptance rate. - Error rate. - Escalations. - Cost per run. - Tool failures. - Review notes. ### 4. Experiment dashboard - Active tests. - Hypotheses. - Success thresholds. - Results. - Learnings. - Decisions made. ### 5. Trust and governance dashboard - Approval exceptions. - Risk flags. - Policy violations. - Sensitive data events. - Audit gaps. - Escalations. ### 6. FinOps dashboard - AI spend. - Cloud spend. - Tool subscriptions. - Cost per workflow. - Cost per accepted deliverable. - Margin by offer. - Waste or unused capacity. ## Dashboard rules - Start with decisions, not vanity metrics. - Every metric needs an owner. - Every dashboard needs a review cadence. - Every alert needs an action. - Every metric should lead to a decision or behavior change. ## Dashboard template ```md # Dashboard Plan ## Dashboard Name ## Decision It Supports ## Audience ## Metrics ## Data Sources ## Update Frequency ## Alert Rules ## Owner ## Review Cadence ## Actions Triggered ``` --- # Phase 11: Human Review Layer ## Goal Use humans for judgment, trust, taste, relationships, and accountability. Human review is not a weakness. It is the trust layer. ## Review categories ### 1. Factual review Checks facts, sources, claims, dates, numbers, and citations. ### 2. Strategic review Checks whether the output supports the business goal. ### 3. Brand review Checks tone, style, positioning, and customer expectations. ### 4. Technical review Checks code, data, integrations, security, and functionality. ### 5. Legal or compliance review Checks regulated claims, contractual obligations, privacy, and risk. ### 6. Customer approval Confirms that the buyer or stakeholder accepts the output. ## Review gate template ```md # Review Gate ## Output Type ## Reviewer ## Review Standard ## Required Checks ## Pass Criteria ## Common Failure Modes ## Escalation Rule ## Approval Record ``` --- # Phase 12: Governance and Risk ## Goal Keep speed from becoming danger. AIC companies need governance from day one because agents multiply execution capacity. ## Risk categories - Data privacy. - Security. - Hallucination. - Bad advice. - Brand damage. - Regulatory claims. - Biased outputs. - IP misuse. - Unauthorized spending. - Tool misuse. - Customer data leakage. - Incorrect automation. - Model drift. - Prompt injection. - Vendor dependency. - Cost sprawl. ## Governance components 1. Role-based access. 2. Tool permissions. 3. Sensitive data rules. 4. Approval gates. 5. Audit logs. 6. Escalation paths. 7. Incident response. 8. Kill switches. 9. Vendor review. 10. Data retention rules. 11. Model evaluation. 12. Customer disclosure rules when needed. ## Risk register template ```md # AIC Risk Register | Risk | Likelihood | Impact | Owner | Control | Escalation | Status | |---|---:|---:|---|---|---|---| | | | | | | | | ``` ## Agent governance rule No agent should have more authority than the company can monitor. --- # Phase 13: Experiment Factory ## Goal Make learning continuous. The AIC should not rely on big bets. It should run controlled tests. ## Experiment types - Customer discovery. - Landing page tests. - Offer tests. - Pricing tests. - Workflow tests. - Prompt tests. - Agent evaluation tests. - Channel tests. - Sales script tests. - Product feature tests. - Retention tests. - Onboarding tests. ## Experiment rules - State the hypothesis. - Define the smallest test. - Define success before running. - Use evidence, not vibes. - Capture learning. - Decide what changes. - Kill weak ideas quickly. - Scale what works. ## Experiment card template ```md # Experiment Card ## Hypothesis ## Why It Matters ## Test Design ## Audience ## Success Metric ## Minimum Success Threshold ## Cost ## Timeframe ## Risks ## Result ## Decision - Scale - Revise - Kill - Retest ## Learning Captured ``` --- # Phase 14: Community and Crowd ## Goal Create a compounding edge outside the company. Community is not a newsletter list. It is a living contribution system. ## Community types - Customers. - Advisors. - Experts. - Beta testers. - Creators. - Developers. - Fans. - Operators. - Freelancers. - Industry insiders. - Local partners. ## Community loops ### 1. Learning loop Community shares problems, insights, and feedback. ### 2. Contribution loop Community helps improve the product, data, workflows, or content. ### 3. Trust loop Community proves credibility through testimonials, case studies, and referrals. ### 4. Talent loop Community becomes a source of staff on demand. ### 5. Distribution loop Community spreads useful work because it benefits them. ## Community design rules - Lead with value. - Do not ask before giving. - Make contribution easy. - Recognize useful contributors. - Build around the MTP. - Avoid gimmick engagement. - Tie community activity to product learning. ## Community plan template ```md # Community Plan ## Who Belongs ## Why They Care ## Value We Give First ## Contribution We Invite ## Recognition System ## Feedback Capture ## Community to Product Loop ## Community to Talent Loop ## Community to Sales Loop ``` --- # Phase 15: Staff on Demand and Human Specialist Cloud ## Goal Use external human talent without building a bloated company. ## Talent categories - Senior strategist. - Subject matter expert. - Researcher. - Copywriter. - Designer. - Developer. - Automation specialist. - Data analyst. - QA reviewer. - Compliance reviewer. - Sales closer. - Customer success specialist. ## Rules - Keep the core team small. - Document every repeatable process. - Use freelancers for specialized or burst capacity. - Build a vetted bench. - Use trial projects before critical work. - Assign clear standards and review gates. - Convert the best repeat contributors into preferred partners. ## Talent bench template ```md # Staff on Demand Bench | Role | Person/Vendor | Specialty | Rate | Availability | Quality Score | Best Use | Backup | |---|---|---|---:|---|---:|---|---| | | | | | | | | | ``` --- # Phase 16: Leveraged Assets and FinOps ## Goal Stay asset-light without losing control of cost or capability. ## Leverage by default Prefer renting, subscribing, integrating, or partnering for: - Cloud hosting. - AI models. - APIs. - Databases. - Design tools. - Automation platforms. - Project management. - CRM. - Analytics. - Communication. - Customer support. - Payments. - Security monitoring. ## Own or control Own or strongly control: - Brand. - Customer relationship. - Core knowledge base. - Workflow IP. - Agent instructions. - Evaluation rubrics. - Proprietary data where lawful and valuable. - Customer insights. - Distribution channel when possible. ## Dependency ledger template ```md # Dependency Ledger | Asset or Vendor | Purpose | Monthly Cost | Switching Cost | Risk | Backup | Owner | |---|---|---:|---:|---|---|---| | | | | | | | | ``` ## FinOps rules - Track AI cost by workflow. - Track AI cost by client. - Track cost per accepted output. - Track human review cost. - Kill tools that do not justify cost. - Avoid model overkill. - Match model power to task risk and complexity. - Watch for hidden platform sprawl. --- # Phase 17: Launch and Growth ## Goal Get the first users, prove the outcome, and create the growth loop. ## Launch sequence 1. Define the wedge. 2. Build the simplest credible offer. 3. Recruit 3 to 5 design partners or sandbox clients. 4. Deliver manually with AI assistance. 5. Instrument the workflow. 6. Capture before and after proof. 7. Turn proof into case studies. 8. Package the workflow. 9. Sell the next 10 customers. 10. Automate the repeated parts. 11. Create recurring revenue. 12. Build platform elements only after repetition. ## Growth channels Choose based on customer behavior. Possible channels: - Founder-led outreach. - LinkedIn authority content. - Targeted email. - Partner referrals. - Expert community. - Webinars. - Audits. - Reports. - Benchmarks. - Templates. - Tools. - Case studies. - Search content. - Marketplace listings. ## Growth rule Do not scale acquisition until delivery works. --- # Phase 18: Scale Path ## Goal Move from service to system to platform. ## Scale stages ### Stage 1: Founder-led service The founder delivers most work with AI assistance. ### Stage 2: Documented service The workflow is documented and repeatable. ### Stage 3: Agent-assisted service Agents handle research, drafts, routing, QA, and reporting. ### Stage 4: Managed system The company manages workflows for clients on a recurring basis. ### Stage 5: Client dashboard Clients can see status, recommendations, and results. ### Stage 6: Productized platform The workflow becomes software. ### Stage 7: Ecosystem Partners, experts, contributors, and customers extend the system. ## Scale decision rule Only automate what has repeated successfully at least three times and has clear economic value. --- ## 6. SCALE and IDEAS Translation for AIC ### SCALE 1: Staff on Demand AIC translation: human specialist cloud plus agent labor. What to build: - Vetted freelancer bench. - Agent role bench. - Review standards. - Talent routing. - Contributor scorecards. - Contract templates. Next move: Use agents for first drafts and humans for taste, judgment, expertise, and accountability. ### SCALE 2: Community and Crowd AIC translation: contribution engine. What to build: - Customer advisory loop. - Beta tester community. - Expert contributor network. - Feedback capture. - Public learning assets. - Case study loop. Next move: Turn users, experts, and fans into signal sources, testers, referrers, and contributors. ### SCALE 3: AI and Algorithms AIC translation: agent fleet. What to build: - Narrow agents. - Tool permissions. - Evaluation rubrics. - Human approval gates. - Model selection rules. - Agent dashboards. Next move: Stop thinking “one AI assistant.” Build a governed set of specialized agents. ### SCALE 4: Leveraged Assets AIC translation: asset-light infrastructure plus FinOps. What to build: - Tool stack. - Vendor ledger. - Switching plan. - Cost dashboard. - Data ownership map. - Backup paths. Next move: Rent infrastructure, own intelligence, track cost ruthlessly. ### SCALE 5: Engagement AIC translation: behavior loops that make customers, contributors, and agents improve the system. What to build: - Progress visibility. - Recognition. - Referrals. - Benchmarks. - Personalized recommendations. - Feedback loops. - Achievement paths. Next move: Design engagement around useful progress, not gimmicks. ### IDEAS 1: Interfaces AIC translation: control plane. What to build: - Intake forms. - APIs. - MCP connectors. - Approval queues. - Workflow routers. - Delivery portals. - Escalation paths. Next move: Make every repeat request enter through a structured interface. ### IDEAS 2: Dashboards AIC translation: mission control. What to build: - Company dashboard. - Workflow dashboard. - Agent dashboard. - Experiment dashboard. - Governance dashboard. - FinOps dashboard. Next move: Make performance visible before trying to scale. ### IDEAS 3: Experimentation AIC translation: test factory. What to build: - Hypothesis backlog. - Minimum viable tests. - Prompt tests. - Offer tests. - Workflow tests. - Success thresholds. - Learning archive. Next move: Replace opinions with cheap tests. ### IDEAS 4: Autonomy AIC translation: bounded authority for humans and agents. What to build: - Role charters. - Agent authority levels. - Decision rights. - Escalation rules. - Audit logs. - Kill switches. Next move: Let people and agents move fast inside explicit boundaries. ### IDEAS 5: Social Technologies AIC translation: shared work graph. What to build: - Team chat. - Project spaces. - Knowledge base. - Activity stream. - Decision log. - Customer history. - Agent output archive. Next move: Make company memory searchable, transparent, and connected to work. --- ## 7. Strategic Patterns ### 7.1 The best first company is usually a service Most AIC ideas should begin as a productized service, not software. Why: - Faster to sell. - Faster to learn. - Easier to customize. - Lower build cost. - Better customer insight. - Easier to prove ROI. Software comes after the workflow repeats. ### 7.2 The moat is the workflow, not the model Models will keep changing. The defensible layer is: - Customer knowledge. - Workflow design. - Proprietary examples. - Evaluation criteria. - Human expert review. - Distribution. - Trust. - Data feedback loops. - Community. ### 7.3 The dashboard is the manager In an AIC, management happens through visible systems. Use dashboards to reduce meetings, reduce confusion, and increase autonomy. ### 7.4 Human taste becomes more valuable As AI increases output volume, the scarce skill becomes selection. Humans should focus on: - Taste. - Strategy. - Trust. - Original judgment. - Customer understanding. - Relationship. - Final approval. - Meaning. ### 7.5 Governance is a product feature Trustworthy AI operation is not back-office compliance. It is part of the offer. Customers need to know: - What AI does. - What humans review. - What data is protected. - What claims are verified. - What happens when something goes wrong. ### 7.6 Narrow agents beat magical agents Do not create giant undefined agents. Create narrow agents that do one workflow well. ### 7.7 Every run should improve the system Each project should create: - New examples. - Better prompts. - Better templates. - Better workflows. - Better dashboard metrics. - Better customer understanding. - Better agent evaluations. --- ## 8. First 30 Days Default Plan Use this when the user wants to start now. ### Week 1: Define and focus - Write the MTP Constitution. - Choose one customer. - Choose one painful workflow. - Write the entry offer. - Define the success metric. - Build the risk register. ### Week 2: Build the minimum viable workflow - Create intake form. - Create knowledge base folder. - Create agent charters. - Create review checklist. - Create delivery template. - Create dashboard sheet. ### Week 3: Run sandbox tests - Recruit 3 to 5 test users or clients. - Run the workflow manually with AI assistance. - Track time, cost, output quality, and feedback. - Capture common errors. - Improve prompts and checklists. ### Week 4: Package and sell - Refine the offer. - Write case-study-style proof. - Create a simple landing page. - Build outreach list. - Run first sales test. - Decide whether to continue, revise, or kill. --- ## 9. 90-Day AIC Roadmap ### Days 1 to 15: Clarity - Define MTP Constitution. - Define customer. - Define wedge. - Define offer. - Define success metrics. - Define governance rules. ### Days 16 to 30: Workflow - Build the first workflow. - Create agent charters. - Create intake and delivery templates. - Create dashboard v1. - Run internal tests. ### Days 31 to 45: Sandbox delivery - Recruit design partners. - Deliver real outputs. - Track cost, time, quality, and satisfaction. - Improve review gates. - Capture examples. ### Days 46 to 60: Package - Productize the workflow. - Set pricing. - Build basic sales materials. - Create proof assets. - Create onboarding process. ### Days 61 to 75: Sell - Run outbound tests. - Run content tests. - Run referral tests. - Close first paid clients. - Refine the promise. ### Days 76 to 90: Systemize - Automate repeated steps. - Add dashboard depth. - Add staff on demand. - Improve agent evaluations. - Decide platform path. --- ## 10. Evaluation Rubric Score each AIC idea from 1 to 5. | Criterion | 1 | 3 | 5 | |---|---|---|---| | Pain intensity | Mild | Annoying | Urgent and expensive | | Buyer clarity | Unknown | Likely buyer | Obvious budget owner | | AI advantage | Cosmetic | Helpful | Changes cost or speed dramatically | | Trust feasibility | Risky | Manageable | Strong human review path | | Workflow repeatability | Custom every time | Partially repeatable | Highly repeatable | | Data availability | Weak | Some data | Strong inputs and feedback | | Speed to first test | Months | Weeks | Days | | Revenue path | Unclear | Plausible | Immediate paid wedge | | Moat potential | None | Some workflow learning | Data, trust, workflow, community | | Cost control | Unknown | Trackable | Strong unit economics | ### Interpretation - **40 to 50:** Strong AIC candidate. - **30 to 39:** Test before building. - **20 to 29:** Narrow the wedge. - **Below 20:** Do not build yet. --- ## 11. Common Failure Modes ### 11.1 Platform fantasy Building software before proving anyone wants the workflow. Fix: Start as a service. ### 11.2 Agent sprawl Creating too many agents with overlapping jobs. Fix: Map the workflow first, then assign agents. ### 11.3 No review gate Letting AI output reach customers without accountability. Fix: Define review gates before delivery. ### 11.4 No dashboard Running the company by vibes. Fix: Track cycle time, quality, cost, revenue, and learning from day one. ### 11.5 No buyer Building for a “market” instead of a person with a budget. Fix: Define the economic buyer and buying trigger. ### 11.6 No unit economics Ignoring AI, cloud, tool, and human review costs. Fix: Track cost per accepted output. ### 11.7 Generic positioning Saying “AI-powered” instead of naming the business result. Fix: Sell the outcome, not the technology. ### 11.8 Unbounded autonomy Letting humans or agents act without clear authority. Fix: Use authority levels and escalation rules. --- ## 12. AIC Master Prompt Use this when beginning a new AIC build. ```md You are operating under the AIC, AI Corporation Skill. Take the following core idea and turn it into an AI-native company operating model. Core idea: [PASTE IDEA] Build the response in this order: 1. Idea triage. 2. MTP Constitution. 3. Customer and wedge definition. 4. Market sensing needs. 5. Offer architecture. 6. AIC stack. 7. Agent fleet map. 8. First workflow spec. 9. Dashboard plan. 10. Human review plan. 11. Governance and risk register. 12. Experiment backlog. 13. 30-day execution plan. 14. 90-day roadmap. Be specific. Do not give generic AI startup advice. Make assumptions where necessary and label them clearly. Prefer a service-first path unless there is strong evidence that software should be built immediately. ``` --- ## 13. Minimal AIC Build Kit For the first version, build only this: 1. One MTP Constitution. 2. One customer profile. 3. One wedge offer. 4. One intake form. 5. One workflow spec. 6. Three agent charters. 7. One review checklist. 8. One delivery template. 9. One dashboard. 10. One experiment backlog. 11. One risk register. 12. One sales page or outreach script. If those twelve pieces are not working, do not build more. --- ## 14. Final AIC Principle Do not build an AI company around a tool. Build it around a job that matters, a purpose that coordinates people, a workflow that repeats, agents that reduce cost or increase speed, humans who provide judgment, dashboards that reveal truth, governance that creates trust, and economics that improve every time the system runs.