Founder Experiments

Set up a simple form using Typeform or Tally that captures name, phone number, and inquiry type.

Connect the form to a webhook using Make.com or n8n. When a form is submitted, the webhook fires.

Use an AI code tool (Claude, Cursor, or Replit) to write a script that triggers a call via Twilio or a voice AI provider like Thoughtly, ElevenLabs, or Bland AI. Pass the prospect's name and inquiry type into the opening script.

Prompt your AI engineer to write a call script with a natural opener, two qualifying questions, and a calendar booking link delivered via SMS at the end.

Run 10 test calls on warm leads. Measure hangup rate, conversation length, and booking rate against your current follow-up method.

Pick one government procurement portal relevant to your industry — SAM.gov for federal contracts, your state's purchasing portal, or USASpending.gov for award history.

Use Claude or Cursor to write a Python script that pulls new contract opportunities matching three keyword filters you define (product category, agency type, minimum value).

Pipe the results into a Google Sheet that auto-refreshes daily. Add a column for AI-generated fit score: prompt Claude to rate each opportunity 1-5 based on your capability description.

Set a Make.com automation to Slack you any opportunity rated 4 or 5, with a one-paragraph summary of why it fits and the submission deadline.

Submit a response to one opportunity this week — even a no-bid letter. The process of responding teaches you the procurement language faster than any research.

Pick your next client proposal. Instead of scoping by hours, define the engagement as three 2-week sprints with a fixed deliverable and acceptance criteria for each.

Price each sprint at 1.5x what you would have charged hourly for the same scope. Add a clause: if the sprint deliverable is approved in under 5 days, the next sprint begins immediately.

At sprint kickoff, run a 30-minute outcome interview: what does success look like? What metrics change? This replaces the scope document and creates shared accountability.

Track actual hours spent per sprint for your own records only. Do not share this with the client. Compare your effective hourly rate to your previous model.

At project close, ask the client: did paying per sprint change how you engaged with the work? Their answer is your best marketing copy for the next proposal.

Identify one internal tool you are currently paying for or evaluating. Give it three non-negotiable requirements.

Open Bloom and describe the tool in one or two sentences. Add your three requirements as constraints.

Let the agent build. Don't touch the code — just describe what's missing or wrong and iterate.

Share the result with one teammate via the App Clip link. Watch them open a native app from a single tap.

Note: how long did it take? What does this app do that the off-the-shelf version doesn't? How does it feel to own the data?

Gather your 20 most important internal documents: SOPs, onboarding guides, product specs, FAQs. Export them as plain text or markdown files.

Use Claude or Cursor to scaffold a simple Python RAG app using LlamaIndex or LangChain with a local or low-cost vector store (ChromaDB is free, Qdrant has a generous free tier).

Point the ingestion script at your documents folder. Run it. Your knowledge base is now queryable in natural language.

Add a basic chat interface — Streamlit or Gradio gets you there in under 30 minutes. Deploy to a free Render or Railway instance.

Share the URL with your team. Track which questions they ask. The most frequent questions reveal your biggest documentation gaps.

Export your last three months of SaaS billing. Build a simple table: tool name, monthly cost, last login date, and primary use case.

For each tool, ask Claude to suggest two free or cheaper alternatives that cover 80% of the use case. Add those to the table.

Score each tool 1-3: 1 = could eliminate today, 2 = could replace in a month, 3 = genuinely irreplaceable. Eliminate or cancel all tools scored 1 immediately.

For every tool scored 2, create a 30-day migration task with a specific replacement in mind. Add it to your project management system.

Total the potential monthly savings. If it exceeds $500, that is the budget for one AI tool that actually moves the needle.

Pick one trade niche where you have a contact or relationship: HVAC, roofing, landscaping, plumbing, electrical, cleaning. Interview one operator for 30 minutes about their weekly admin pain.

Map the five tasks they hate most. For most trade businesses, it is: quoting, follow-up, scheduling, invoicing, and Google reviews.

Sign up for GoHighLevel (14-day trial). Build a sub-account with: a lead capture form, a 3-step SMS follow-up sequence, a calendar booking page, an invoice template, and an automated Google review request triggered by job close.

Record a 5-minute Loom walking through the system. Send it to your contact. Ask: would you pay $200/month for someone to set this up and manage it for you?

If yes, you have a business. If no, ask what they would pay and what they actually need. That feedback is worth more than the build.

Write a one-paragraph business requirement in plain English: who uses this, what problem it solves, what they do with it, and what success looks like. No technical specs.

Open Woz or your AI builder of choice. Paste only the business requirement — do not describe the technical architecture. Let the AI infer the stack.

Review the first build for business logic accuracy, not code quality. Ask: does this do what the user actually needs? Iterate on requirements, not code.

Share a link with one real potential user within 24 hours of starting. Do not explain it. Just watch them try to use it.

After watching the first user session, write three sentences about what confused them. Those three sentences become your next build prompt.

Open your most-used product screen. Catalog every visual decision: background color, text color, button color, border radius, font size, spacing unit. Write each one down.

Use Claude to generate a JSON design tokens file from your catalog. Each value should have a semantic name (e.g., 'color.interactive.primary') not a raw value ('#3B82F6').

Ask Claude to generate a simple React or Tailwind component library from those tokens: button, input, card, badge. Each component should only use token names, never raw values.

Swap one page of your product to use only the new component library. Note every place where the page broke — those are your undocumented design decisions.

Share the component file with your designer or a trusted peer. Ask them to build one new screen using only those components. Their questions reveal your token gaps.

Pick one industry where you have domain knowledge. Search for recent legislation (last 3 years) that added a new compliance requirement for businesses in that space.

Talk to five operators in that industry. Ask: how are you handling this compliance requirement today? Spreadsheet, paper, nothing, or a tool?

If the answer is spreadsheet or nothing for three or more of them, you have a product idea. Describe the workflow they need in plain English.

Build a minimal version using a no-code tool (Glide, Softr, or Bubble for data apps) or ask Claude to scaffold a Next.js app. The first version only needs to do one thing: track the compliance requirement.

Charge $99/month. If five people pay it in the first month, raise it to $199. Keep raising until the first no.

Open Claude or your preferred AI coding tool. Prompt it: 'Build me a single-page pitch scoring web app with three sections: Founder, Business, and Gut Check. Each criterion scored 0–3. At the end, show me a summary of weak spots and generate three investor questions I should prepare for based on my lowest scores.'

Score yourself honestly in the Founder section: storytelling, emotional intelligence, learning agility, domain expertise, conviction, and coachability.

Score honestly in the Business section: defensibility, ICP clarity, business model strength, TAM credibility, and tech differentiation.

Complete the Gut Check: moat type (community, data, brand, or none) and whether you would invest your own money.

Give the tool to a trusted advisor. Have them score you independently. Compare the gap between your self-assessment and theirs.

Write down the three things you would do with the raise. Be specific: hire X, build Y, spend Z on marketing. Now ask: which of these can I do at 10% of that cost to validate the assumption first?

Calculate the minimum ARR at which you would have real negotiating leverage with investors. Denver Ventures' threshold: $500K ARR is where the conversation changes.

Map your current runway to that target. If you can reach $500K ARR before running out of money, you should try. If you cannot, you need to raise — but know why.

List every investor meeting you are considering. For each one, write the specific thing you want from them beyond money: intro network, domain expertise, operational help. If the answer is only money, deprioritize them.

Write a one-page bootstrapped product plan: what you build, who pays first, at what price, and what that first cohort proves. This is your leverage document even if you raise.

Ask Claude to generate a 20-question exit readiness assessment covering: clean cap table, documented processes, revenue predictability, customer concentration risk, IP ownership, team dependency on founders, and data room completeness.

Complete the assessment honestly. Score each item 0 (not started), 1 (in progress), or 2 (complete). Any item scored 0 that relates to revenue, IP, or cap table is a blocker — address it first.

Build a simple data room folder structure today even if it is empty: financials, legal, customer contracts, IP documentation, team org chart, product roadmap. Populate what you can in one hour.

Identify the one person at your most likely acquirer who would lead the deal. Find them on LinkedIn. Note the kind of content they engage with and the language they use to describe acquisitions.

Write a one-paragraph acquisition narrative from their perspective: why does buying your company make their life easier? What problem do you solve for their portfolio or strategy?

Pick one high-stakes conversation relevant to your industry: a customer calling to cancel, a job candidate receiving a rejection, a patient asking about a diagnosis. Write a one-paragraph brief: persona, emotional state, what they want.

Use Claude as the simulation engine. System prompt: 'Play this persona, respond dynamically, track tone and empathy markers, and note whether the user is validating or dismissing the persona's concerns.'

Set a turn limit of 8–10 exchanges. After the conversation ends, send the transcript back to Claude with a new prompt: score across emotional validation, clarity of communication, and de-escalation effectiveness. Return as JSON with written explanation.

Use Cursor, Replit, or Bloom to wrap it in a basic chat UI with a post-session report screen. The full stack should be runnable in under 48 hours.

Run two people through the same scenario. Compare their transcripts and scores. Note the variance — that gap is the training opportunity.

Define your product's 'first value moment' — the exact action that tells you a user understands why they signed up. Be specific: not 'created an account' but 'ran their first report' or 'sent their first message.'

Pull data on your last 10 sign-ups. For each user, find the timestamp of sign-up and the timestamp of that first value moment. Calculate the gap in hours.

Map every step between sign-up and first value moment. Write each one on a sticky note or doc. Count them. Anything above 5 steps is almost certainly killing activation.

Identify the single step with the highest abandonment. Use Hotjar, FullStory, or a session recording tool. Watch 5 recordings of users who dropped off at that step.

Remove or automate that step. If you cannot remove it, write a better instruction. Measure the activation rate on the next 10 sign-ups.

Think back to your last three build sessions that stalled. Write down exactly where you got stuck: the error message, the concept you didn't understand, the tool that behaved unexpectedly.

Categorize each failure: (1) AI generated wrong code, (2) missing domain knowledge, (3) wrong tool for the job, (4) unclear requirements, (5) environment setup issues. Most failures fall into one of these.

For each category, write a one-paragraph recovery protocol: the first three things you do when you hit this type of wall, before giving up.

Build a personal 'unstuck' doc. Pin it in your browser. The moment you feel flow leaving your body, open it.

Share the recovery protocol with one other builder. Ask them to add their own failure modes. The combined doc is now a team knowledge base.

Build a script (Node.js or Python) that takes a supplier name and country as inputs. Use a search API — Perplexity, Tavily, or Exa — to pull publicly available information about that supplier.

Feed that information into a Claude API call with this system prompt: 'You are a supply qualification analyst for a vetted marketplace. Produce a structured brief covering: (1) verifiable credentials, (2) named practitioners and qualifications, (3) red flags including review anomalies, (4) community sentiment, and (5) an overall trust signal score from 1–10 with reasoning.'

The output becomes a pre-visit research brief any team member can use before committing to an in-person site visit.

Add a simple UI with Cursor or Replit — an input form, a loader, and a formatted output card. Entire build should take under 48 hours with AI-assisted coding.

Log your actual vetting decisions against the AI brief scores over time. Use that data to refine the scoring prompt and improve calibration.

Open your AI coding tool of choice (Claude, Cursor, Replit). Prompt it: 'Build me a simple Python chatbot using a JSON knowledge base. When a user asks a question the bot cannot answer, log the question with a timestamp to a CSV file called unanswered.csv. Give me a basic web interface using Flask so my team can use it in a browser.'

Populate the JSON knowledge base with your 20 most common customer or team questions and answers. Your support inbox, Slack history, and FAQ page are gold mines for this.

Run the bot for five business days. Have your team use it instead of emailing you or each other for those common questions.

Open unanswered.csv. Sort by frequency. The questions that appear most often are your highest-value automation targets and your roadmap for what to build or buy next.

Create a GitHub repo and add your marketing copy, product spec, engineering spec, and QA docs as individual markdown files.

Open Claude Code and prompt it: 'Build a skill that reads all markdown files in this repo and outputs a structured list of contradictions, omissions, and misalignments between them. Run it on a schedule and flag any drift whenever a document changes.'

Write a constraints file — a markdown doc that tells the agent what consistent, on-brand output looks like. Ask Claude what constraints it needs to produce reliable output.

Have the agent reference the constraints file every time it runs. When any document changes, the agent checks for drift before it ships.

To go deeper: feed your reading sessions into a JSON blob, then have an agent run entity analysis, topic analysis, and sentiment scoring automatically.

Pick one recurring task that takes you 15 minutes or more per day. Write it down: the inputs, the decision you make, and the output you produce.

Define the agent's scope in one sentence. It does exactly one thing. No fallbacks, no related tasks, no scope creep.

Build it in under 4 hours using Claude Code, n8n, or Make.com. If it takes longer, the scope is too broad — cut it in half.

Run it for 5 business days. Track: time saved per day, error rate, and one instance where it got something wrong. Document the failure.

At the end of day 5, make a binary decision: keep it, kill it, or redesign it. James Everingham's rule: fast failure with a specific lesson is worth more than slow success with no data.

List every AI tool your team has used in the last 90 days. For each one, note what data you submitted: documents, code, customer data, internal processes, financials.

Pull the Terms of Service for each tool. Ask Claude to summarize the data retention, training, and ownership clauses for each one in plain English. Note any tool that claims rights to your inputs.

Score each tool on a 3-point risk scale: 1 = input used for training, 2 = data retained beyond session, 3 = third-party data sharing. Cancel or replace any tool scored 3 immediately.

For your highest-risk tools, define a data hygiene protocol: what can be submitted, what must be redacted, and who is allowed to use the tool for sensitive work.

Document your findings in a one-page AI Data Policy. This is the foundation of your security posture when a larger customer asks about your AI usage.

Define the one measurable outcome your product delivers for customers. Be specific: not 'improves efficiency' but 'reduces X by Y% within Z weeks.'

Identify the current market baseline for that outcome. What does that metric look like without your product? Find data — from industry reports, customer benchmarks, or your own case studies.

Write an outcome-aligned pricing proposal: your fee is a percentage of the measured improvement above baseline. Present this to your next two prospects alongside your standard pricing.

Build a simple measurement framework: how will you and the customer jointly measure the outcome at 30, 60, and 90 days? Get agreement on the measurement method before pricing is discussed.

Track which customers prefer outcome pricing vs. flat fee. Outcome pricing self-selects for customers who believe in the product and are willing to be held accountable for usage.

Write down your product's core value proposition as it exists today. Now write what it needs to look like in 10 years to still be relevant. Note every assumption that changes between those two versions.

Map the AI capabilities that currently exist at a research or demo level but are not yet production-ready. Pick the one most likely to be production-ready in 3 years. How does it change your product if it is?

Identify the one workflow in your product that would benefit most from agentic AI — not a chatbot, but an agent that takes action. Define what that agent would do, what inputs it needs, and what oversight it requires.

Build a constraints document: what can your AI agents never do without human approval? This is your governance layer, and it is faster to define now than after an incident.

Present your 10-year map to a customer. Ask them: which of these would make you switch from our competitor today? Their answer tells you your 12-month roadmap.

Ask ChatGPT, Claude, and Perplexity the top three questions your ideal customer asks before buying a product like yours. Note whether your company or content appears in any answer.

For each query where you do not appear, find the source that does — it is almost always a piece of content: a blog post, a Reddit thread, a LinkedIn article, or a review site.

Create one piece of content specifically designed to answer the top question in your category. Publish it on your own domain. Format it as a direct, declarative answer — AI models prefer structured, authoritative responses.

Submit your content to any aggregators or directories that appeared in the AI answers. Being cited in those sources increases your probability of appearing in AI-generated answers.

Re-run the same three queries in 30 days. Track whether your content or brand now appears. This is your AI search share of voice baseline.

Define your category in one sentence as a customer would describe it — not how you would. Use that language in every query.

Open ChatGPT, Claude, Perplexity, and Google AI Overviews. Ask each the same five questions your best customers asked before buying from you. Screenshot every response.

Score each response: does your brand appear? Does a competitor appear? Does a third-party review or comparison site appear? Build a simple tracker.

For every query where a competitor appears and you do not, find the specific content asset that earned them that mention. Reverse-engineer its structure.

Publish one content asset this week that directly targets a query where you are invisible. Measure AI citation rate monthly.

List the 10 people in your network who connect seemingly unrelated worlds: the healthcare operator who also advises startups, the developer who has deep relationships in real estate, the marketer who spent 5 years in logistics.

For each person, write the intersection they represent: two domains that are rarely connected but could unlock outsized value if combined. That intersection is a potential business or partnership.

Pick the most interesting intersection. Write a one-paragraph thesis for why an AI product built at that crossing would be hard to replicate — who would build it, and why don't they?

Send a message to the connector: 'I've been thinking about X and Y. Do you know anyone building at that intersection?' Their answer either validates your thesis or saves you months of research.

Follow Josh Furstoss's protocol: the best AI companies are not being built by the people you'd expect, in the markets you'd expect. Write down the unexpected builder for your thesis — then become that person.

Write down every assumption your AI feature makes about the environment it will run in: network conditions, hardware, data quality, user behavior, lighting, noise, or edge cases.

Design three adversarial test scenarios that violate those assumptions. For Marcin's robotics team, this was a real airport at 3am with moving crowds and unpredictable surfaces.

Run your feature in each adversarial scenario before any user sees it. Log every failure mode with a timestamp and exact failure description.

For each failure, classify it: (A) unacceptable in production, (B) acceptable with a warning, or (C) acceptable as-is. Fix every A before launch.

Publish your failure gauntlet results internally. The team that knows how their product breaks in the real world ships better products than the team that only knows how it works in demos.

Identify five people who match your ideal customer profile. They do not need to be warm leads — cold outreach works if your ask is small and specific.

Send each person a single question: 'What is the most frustrating part of [the problem you're solving]? I'm trying to understand it before building anything.' Do not describe your solution.

After the response, ask one follow-up only: 'What have you tried to solve it?' The answer tells you the graveyard of solutions that already failed.

Compile the five responses. Highlight every phrase that appears in two or more answers — those are your real requirements. Anything that only appears once is a nice-to-have.

Write a one-paragraph product description using only language from the five interviews. No marketing language. If you can't describe it in their words, you do not understand the problem yet.

Pick one industry you know deeply. List every data source an expert in that field uses to make decisions. Be exhaustive: tools, spreadsheets, reports, emails, verbal briefings.

For each data source, ask: does it talk to the others? Score it: 0 = fully isolated, 1 = manual export/import, 2 = partial integration, 3 = real-time sync. If most are 0 or 1, you have found a fragmentation problem.

Interview three experts in that field. Ask them: 'What decision do you make weekly that requires you to pull from more than one of these sources?' That decision is your product's core use case.

Write a one-paragraph description of the fragmentation problem from the expert's perspective, not the technology perspective. Share it with the three people you interviewed. Ask: is this accurate?

The product that solves their fragmentation problem is worth more than the one that adds a new data source. Resh burned $30M finding this. Your interview costs nothing.

Before writing your next sprint's tickets, identify three customers who represent your ideal buyer. Schedule a 20-minute call with each this week.

In each call, ask only: 'What did you try to do in our product last week that didn't go the way you expected?' Do not ask what features they want. Ask what failed.

After all three calls, write down every failure they described. If a failure appears in two or more calls, it is a P0 — it goes to the top of the sprint. No exceptions.

Cut one item from your existing sprint backlog for every P0 you add. The sprint length does not change. The priorities do.

At sprint close, review: did fixing the real failures change any of your roadmap assumptions? Document what changed. This is your calibration record.

Before writing your next feature spec or dev brief, open Codalio or Claude. Describe your product, your user, and the problem you are trying to solve. Then stop typing.

Let the AI interview you: it will ask clarifying questions about user journey, edge cases, success metrics, and technical constraints. Answer each one honestly.

After the interview, ask the AI to generate a structured requirements document: user story, acceptance criteria, technical constraints, and out-of-scope items. Review it for accuracy.

Share the generated brief with your developer before any code is written. Ask them: is anything here ambiguous? Their questions reveal missing requirements you can resolve in 10 minutes instead of a two-week rework.

Compare the time spent on this brief vs. your previous manual spec process. Note whether the resulting build better matched your intent.

Open a blank document. Write your honest answers to five questions: (1) What is the one thing blocking your growth right now that you have been avoiding? (2) Which of your team's current workflows will an AI agent replace in 12 months? (3) What assumption about your business model has never been tested? (4) What would you cut today if you had to survive on 50% of your current revenue? (5) What are you building that your best customer has never asked for?

Share your answers with one person who will tell you the truth — not your co-founder, not an investor. A peer founder who has no stake in your answer.

For each answer, write one action that costs nothing but time and could change the answer in 30 days. These are your highest-leverage experiments.

Set a 30-day calendar reminder for each action. At the 30-day mark, write one sentence on what changed.

Apply Ran Aroussi's 30-year coder principle: the best diagnostics are the ones that surface what you already know but have not acted on. The audit does not reveal new information — it removes the excuses for avoiding what you already see.

Write your idea in one sentence: who it helps, what it does, and why now. Do not describe the technology.

Open BrainGrid or Claude. Paste your one sentence and prompt: 'Generate a product requirements document for this idea. Include: user personas, core use cases, success metrics, technical constraints, out-of-scope items, and open questions.'

Read the PRD for one thing only: the open questions. These are the assumptions your idea rests on that have not been tested. Pick the three most important ones.

Design the smallest possible test for each open question. A Typeform survey, a landing page, a five-minute customer call. Run all three before writing any code.

After the tests, return to the PRD and update it with what you learned. A PRD updated by real evidence is worth ten times one built purely from assumptions.

Pull your last 30 days of ad spend and revenue. Calculate your ROAS. Now subtract COGS, fulfillment, returns, and customer service costs from that revenue. What is your actual margin per order?

Calculate your true CAC: total marketing spend divided by new customers acquired, not total orders. Include your own time at a realistic hourly rate if you are running ads yourself.

Calculate LTV using your actual repurchase rate, not your hoped-for one. If you do not have repurchase data yet, assume it is 20% lower than your gut says.

Divide LTV by CAC. If the ratio is below 3:1, you are acquiring customers you cannot afford. Write down the one lever — price, COGS, retention, or CAC — that, if improved by 20%, gets you to 3:1.

Build a simple model in Google Sheets or ask Claude to build one: input your current numbers, and the model shows you how LTV:CAC changes as you move each lever. This model is more valuable than any dashboard.

Export your last 12 months of business expenses from your bank or accounting software as a CSV.

Ask Claude to categorize each transaction by IRS Schedule C line item. Prompt: 'Categorize each expense as one of the following IRS Schedule C categories: advertising, car and truck, commissions, contract labor, insurance, interest, legal and professional, meals (50%), office expense, rent or lease, repairs, supplies, taxes and licenses, travel, utilities, wages, or other expenses. Flag any that are unclear.'

Review the flagged items with your accountant. The goal is not to replace the accountant — it is to walk into that meeting with 80% of the categorization already done.

Identify the three largest expense categories and ask: am I capturing all deductible expenses in this category? Most founders underreport home office, vehicle use, and professional development.

Set a monthly calendar block to run this same categorization pass. Batch processing monthly takes 15 minutes. Doing it for a full year at tax time takes two days.

Build a simple spreadsheet with four sections: (1) Emergency fund — how many months of expenses can you cover? (2) Liquid investments — what percentage of your savings is immediately accessible? (3) Fixed expenses — what is your minimum monthly burn? (4) Income predictability — how stable is your income over the next 24 months?

Apply CJ Follini's two-barrier test: do you have 2-3 years of budgeting history? If not, start tracking every dollar in and out today before considering any illiquid investments.

Score your readiness to allocate to alternatives: 0-40% liquid savings = not yet, 40-60% = could consider liquid alternatives only, 60-80% = ready for limited illiquid allocation, 80%+ = follow the Yale Model.

Pick one alternative asset class from the 14 CJ identifies that matches your existing knowledge. Real estate if you understand property, art if you have domain knowledge, venture if you understand tech. Start with the one where your due diligence advantage is highest.

Read the offering documents for one deal in that category — even if you do not invest. Learning to read an operating agreement or term sheet takes one afternoon and pays dividends for life.

For one week, record a 3-minute voice note at the end of each day answering the same three questions: What surprised me today? What did I decide, and why? What do I want to think about tomorrow?

At the end of the week, transcribe all five recordings using Whisper (free API) or any voice-to-text tool. Paste them into a single document.

Ask Claude to identify: recurring themes across the five entries, decisions you made with low confidence, and questions you raised but never answered.

Build a simple folder structure: /voice-notes/YYYY-MM-DD. Set a phone reminder for 9pm daily to record. This is your raw thinking, not performed for anyone.

After 30 days, run the same Claude analysis on the full month. The patterns it surfaces will surprise you — they represent your actual priorities, not the ones on your task list.

Ask every team member to fill out a 5-minute anonymous survey: (1) Which AI tools do you use daily? (2) Name one task you completed faster because of AI this week. (3) Name one task you tried to use AI for and it failed. (4) What is the most important thing your role requires that AI cannot do?

Score the results by role. Anyone who answers question 2 with 'none' is a passenger. That is not a firing offense — it is a training opportunity.

For every passenger, identify one specific task in their role that AI can demonstrably accelerate. Design a 30-minute demo using real data from their actual job. Show, do not tell.

Greg Shove's threshold: you want 30-50% AI adoption within a team, not 10%. Map your current adoption rate by function. The functions at 10% are your highest-leverage improvement areas.

Create a monthly AI wins share-out: a 10-minute team meeting where one person demonstrates something that changed how they work. Rotate presenters. Recognition drives adoption faster than mandates.

Pick one decision your team makes weekly that requires judgment: a prioritization call, a customer escalation response, a content approval. This is your target workflow.

Map the current workflow: inputs, decision maker, time spent, error rate if you know it. Write it down in under 10 minutes.

Design an AI-assisted version: which part does the AI do, which part does the human do, and where does the human override the AI? Be specific about the handoff point.

Run the new workflow for two weeks with one person. Track: time saved, decision quality (did the outcome match the goal?), and any moments where the human overrode the AI and was right to do so.

After two weeks, document what the AI got wrong and why. Use that failure analysis to improve the prompt, the constraints, or the human review step. This is Geoff Gibbins' CRI (Collaboration Readiness Index) in practice.

List every member of your team. For each one, score them 1-5 on the four skills Uma identifies as AI-resistant: (1) strategic influence — can they move people without authority? (2) visibility — are they known for their work beyond their immediate team? (3) communication — can they explain complex things simply under pressure? (4) emotional intelligence — do they read rooms and respond to what is actually happening?

Any team member scored below 3 in two or more areas is at risk of being sidelined as AI raises the floor on technical execution. That is your coaching pipeline.

For your highest-potential team member with the lowest leadership scores, design one visible assignment: a cross-functional project, a customer presentation, or an internal talk. Something that exercises the weak skill.

Ask Claude to generate a 30-day development plan for one skill area. Review it with your team member. The best development plans are built with the person, not for them.

Set a 90-day review. Uma's six-month Sought After Leader Experience exists because real behavior change takes time. Your experiment is the on-ramp, not the destination.

Choose one year of your life — preferably the most recent full year — that you want to preserve meaningfully. Pull all photos from that year into a single folder.

Use Claude to generate a structured reflection prompt for that year: major decisions made, people who influenced you, beliefs that changed, moments you want to remember in 20 years.

Record your answers to each prompt as voice notes. Do not type them — voice captures the texture that text edits out.

Build a private Notion or Obsidian vault with one page per month. Embed or link the photos. Paste the voice note transcripts. Add one paragraph of context for anything that would need explanation in 10 years.

Share nothing publicly. This archive is for you and whoever you choose to give access to. Its value is in its privacy.

For every piece of content you have published in the last 90 days, write down the honest reason you made it: genuine creative impulse, or because you thought it would perform?

Identify the three metrics you use to judge whether content succeeded. Now ask: if those metrics did not exist, what would you make instead?

Create one piece of content this week entirely divorced from performance metrics. No tracking, no A/B test, no engagement monitoring for 72 hours. Publish it and walk away.

After 72 hours, note how the content performed without the usual optimization. Then note how it felt to make it without the metrics pressure.

Build a ratio for your content calendar: for every 4 pieces optimized for distribution, make 1 piece optimized for truth. Track whether your audience responds differently to each over 90 days.

Pick 10 podcast episodes you have listened to in the last 90 days that were genuinely valuable. For each, write down the single most important insight — not a summary, but the specific thing that changed how you think.

Run those 10 insights through Claude with this prompt: 'Identify patterns across these insights. What recurring themes appear? What tensions exist between them? What question do they collectively suggest I should be asking?'

Build a personal insight tracker: a simple Airtable, Notion database, or even a Google Sheet. Columns: source, date, insight, category, and a link to where you can find it again.

Set up a Whisper API script (or use MacWhisper for a no-code version) to auto-transcribe any audio you flag as worth reviewing. Pipe the transcript to Claude for a structured insight extraction.

After 30 days, run the same pattern analysis on your full insight tracker. The recurring themes are your actual learning priorities — not the ones on your reading list.

Identify one relationship in your life — professional or personal — that has unresolved tension or distance. Write a one-paragraph description of the emotional state of the relationship as it is today.

Use an AI music generation tool (Suno, Udio, or similar) to generate a piece of music that reflects that emotional state. Do not think about what it should sound like — just describe the feeling in plain language.

Listen to the generated piece privately. Note whether anything shifts in how you think about the relationship after hearing it. This is Ziah Orion's core thesis: the emotional distance between thought and sound creates space for something new to emerge.

Generate a second piece that reflects where you want the relationship to be. Note the difference between the two. That difference is your work.

Share the experiment result — not necessarily the music — with one person you trust. Ask them to try the same exercise with something unresolved in their own experience.

Pick your most important current product idea or the next feature on your roadmap. Write it in one paragraph: what it does, who uses it, and what makes it different from what exists today.

Open SenseIP and run your idea through Leo. Do not filter or polish the description — paste your raw thinking. The goal is to see what the patent database already knows about your space.

Review the prior art Leo surfaces. For each result, write one sentence: how does your approach differ from this patent? If you cannot answer, that is a gap to address before building.

Ask Leo to help you identify what is most novel about your idea. Use the feedback to sharpen your description — not to file a patent necessarily, but to articulate your defensible differentiation before your next investor conversation or customer pitch.

If anything in the results looks like a freedom-to-operate concern — a granted patent that covers something you were about to build — flag it before any engineering work begins. A 10-minute check now costs nothing. Discovering infringement after launch costs everything.

For the next 30 days, log every non-trivial prediction you make: a hiring decision, a product bet, a sales forecast, a market call. For each one, write the prediction in one sentence and your confidence as a percentage (e.g. '70% — this feature ships by Friday').

At resolution, log whether you were right or wrong. After 30 entries, tally your results by confidence band: how often were your '70%' predictions correct? Your '90%'? This gap between stated and actual confidence is your calibration error.

Prompt Claude: 'I made these predictions over 30 days [paste log]. Identify patterns in where I am systematically overconfident or underconfident. What categories of decision consistently have the largest calibration gap?'

Pick your worst-calibrated category — the type of decision where your confidence most consistently doesn't match outcomes. Design one process change: a second opinion, a checklist, a waiting period. Apply it for the next 30 days and measure whether calibration improves.

Ben Turtle's Polymarket insight: real-money bets accelerate calibration because the cost of overconfidence is immediate. Choose one upcoming prediction and put something meaningful on the line — a public commitment, a small bet with a peer — to create the same forcing function.

List every tool in your current production stack that has an API: transcription services, show notes generators, CRMs, enrichment tools, data sources. Write each one on a single line with its API documentation URL.

Install the Lava MCP in your Claude Code instance and load your wallet. Then prompt Claude: 'Check which of these services are accessible through the Lava gateway. For any that are, list the endpoint name and pricing model.'

For each available service, write a single chained system prompt that takes one input (a recording file path or timestamp) and chains all downstream tasks: transcription, show notes generation, title options, and formatted asset output.

Run the workflow end-to-end on a real piece of content. Time the full run. Compare it against your current manual process time and note how many browser tabs and copy-paste steps it eliminated.

Log the actual cost in Lava wallet credits for the full workflow. Divide by the number of production steps completed. This is your cost-per-task number — the baseline you optimize against as you scale.

List every role on your team and classify each one: (A) benefits most from in-person presence (sales, onboarding, design collaboration, mentorship-dependent roles), (B) works best with protected remote focus time (deep coding, writing, analysis), or (C) works equally well either way. This is your baseline for designing the cadence.

For each A-category role, identify the three in-person interactions that create the most value: a standing team standup, a weekly pipeline review, a design critique session. These become your anchored in-person days.

Design a 4-week schedule that creates at least 2 days per week where your A-category team members are in the same space - and protects at least 2 days per week for heads-down remote work. Share the draft with the team before implementing.

Run the cadence for 30 days without changing it. At the end of week 2 and week 4, ask three questions: (1) Did you have the in-person interactions that mattered? (2) Did you have enough protected focus time? (3) Did commute days feel worth it? Score each out of 10.

Compare the week 4 scores to week 2. If specific days or team groupings score consistently low, adjust for month 2. Track one proxy metric throughout: either time-to-decision on cross-functional items or number of async threads that required follow-up versus resolved cleanly.

List every data source your company uses for AI-related tasks: CRM, customer support tickets, sales call transcripts, product analytics, contracts, billing records. For each one, write a one-sentence description of what it contains and how it currently feeds into any AI workflow.

For each data source, answer three questions: (1) Is it deduplicated - or do records exist in multiple places? (2) Is it consistently formatted - or do field values vary arbitrarily? (3) Is it accessible - can an AI system reach it programmatically, or is it locked in a legacy format or manual export?

Score each data source: green (AI-ready), yellow (needs moderate cleaning), red (blocked from AI use without significant work). Tally your results. This is your data readiness map.

For your most important red-category data source, estimate the cost of the workaround you are currently using: manual data entry hours per week times hourly cost. Annualize it. This is the baseline value of fixing it.

Pick one yellow-category source and run a 5-10 record sample through an AI extraction task - ask Claude to pull specific fields from a sample of records and compare accuracy against the actual data. The gap between expected and actual accuracy is your data quality problem made concrete.

List every type of document your company receives, produces, or stores where humans currently extract information by hand: vendor contracts, customer intake forms, support emails, invoices, proposals, meeting notes. For each category, estimate: how many per week, and how many minutes a human spends processing each one.

Pick the highest-volume, highest-time-cost category from your list. Collect 10 representative sample documents from that category - make sure they include edge cases (unusual formats, incomplete fields, scanned documents, tables).

Prompt Claude or another LLM: 'Here is a document [paste content or describe structure]. Extract the following fields: [list 5-8 specific data points you actually need]. Return as a structured JSON object.' Run all 10 samples through the same prompt.

For each output, verify accuracy against the source document manually. Tally: how many fields were correctly extracted? What error patterns appeared? Were edge cases handled or did they fail completely? This is your baseline accuracy score.

Calculate the value of getting to 95%+ accuracy at scale: annualized time savings (hours per document x documents per year x hourly cost) minus the cost of the AI tool or service. If the multiple is greater than 3x, you have a use case worth building or buying a solution for.

List every AI agent or automation currently running in your company: outbound sequences, content pipelines, data enrichment workflows, customer support bots, internal Slack agents. For each one, write one sentence describing what it does and one sentence describing what 'broken' looks like.

For each agent, list the three things that go wrong most often: authentication failures, hallucinated outputs, broken integrations, stale data, or incorrect routing. These are the monitoring requirements for the role you are designing.

Draft a one-page job description for an AI Agent Monitor. Required sections: daily monitoring checklist, error escalation protocol (when to flag vs. when to fix independently), tools used (Slack, your automation platform, any dashboards), and success metrics (uptime percentage, mean time to resolution, weekly incident log).

Prompt Claude: 'Here is my AI agent stack and the errors each agent produces [paste your list]. Write a daily monitoring SOP for a non-engineer who will review outputs, test key workflows, and log anomalies. Format it as a repeatable checklist.' Use the output as the core of your JD's responsibilities section.

Post the JD to one offshore staffing platform or share it with Penbrothers. The goal is not necessarily to hire immediately - it is to see whether the role as written would attract qualified candidates, which tells you whether your process documentation is clear enough for someone outside your company to execute.

Define your two success metrics before you start: one for virality (e.g. 500 upvotes, 1,000 shares, press pickup) and one for revenue (e.g. 5 paying customers, $500 in transactions). Write both down. These are binary - you either hit them or you do not.

Pick the single most stripped-down version of your idea that could prove or disprove the core mechanic. Not an MVP - a proof-of-concept. The question is: does anyone care? Not: is the product complete?

Set a launch date inside the 100-hour window and commit to it publicly - a Product Hunt pre-launch page, a waitlist with a countdown, or a post announcing the date. External commitment eliminates the temptation to delay for polish.

Build only what is necessary to deliver the core experience to the first user. Log every hour. When you find yourself building something outside the critical path, write it on a 'post-sprint' list and keep moving.

On launch day, do not watch metrics passively. Actively post to three communities where your target user lives: a subreddit, a Slack group, a newsletter audience. Measure engagement in the first 24 hours only - the signal is in the initial reaction, not the long tail.

Write down what you currently sell in one sentence: 'I sell [X] and customers pay [price] to [access/use] it.' Now write what your best customers actually buy: the outcome they achieve after using your product. 'After using [X], our best customers [achieve Y outcome in Z timeframe].' If you cannot complete the second sentence cleanly, that gap is your first insight.

Rewrite your value proposition as an outcome contract. Draft a one-paragraph internal memo: 'We will deliver [specific outcome] for [customer type] in [timeframe] for [price].' The outcome should be something the customer can verify - a placed hire, a filed permit, a shipped design system - not a capability or a feature.

Identify the three AI workflows you would need to build or meaningfully improve to actually deliver that outcome reliably at scale. For each, rate your current capability: (A) already doing it well, (B) could build it in 30 days, (C) not yet possible with available tools. A-category workflows are your moat. C-category workflows are your roadmap.

Price the outcome: what would a customer pay for this result if it came with a delivery guarantee? Anchor to what they currently pay for the analog human service - recruiter fee, agency retainer, consulting engagement. That is your ceiling. Calculate your delivery cost with AI leverage. If the margin is positive and the gap is large, you have an outcome-as-a-service opportunity.

Find one design partner willing to pay for the outcome before you build the full system. Pitch the outcome, not the technology. If three people say yes in principle, proceed. If three people say 'interesting but not now,' the outcome is a vitamin - return to step one and find the painkiller version.

List every agent, script, automation, and integration in your company that has access to production systems or external APIs. Include n8n workflows, Zapier zaps, CI/CD pipelines, cron jobs, AI coding assistants, and any tool that acts on your behalf programmatically. For each, write one sentence describing what it can do.

For each item on your list, document where its credentials live: (A) hardcoded in source code, (B) in a .env file on a developer machine, (C) in a secrets manager or vault, (D) in the CI/CD platform's secrets store, (E) device-bound or time-limited. Rate each A-E. Most teams discover more A and B than they expect.

For each A or B credential, calculate the blast radius if it were compromised today: what systems could an attacker access, what data could they read or exfiltrate, what actions could they take? Write one sentence per credential. This exercise makes abstract risk concrete.

For each agent, document the human authorization behind it: who approved this agent running with these permissions? When was that decision made? Is it documented anywhere? If you cannot answer all three questions, the agent is operating without a traceable authorization chain.

Write a one-page Permission Minimization Policy for your team: (1) no production credentials in .env files on developer machines - use a secrets manager, (2) all agent credentials are scoped to minimum required permissions, (3) all agent credentials have an expiry date or rotation schedule, (4) every agent has a named human owner responsible for reviewing its permissions quarterly.

List every equity position your founding team holds: company shares, vested options, any secondary positions. For each, estimate the current value based on the most recent 409A or funding round valuation. Then calculate what percentage of each person's total estimated net worth (including savings, real estate, public market investments) is represented by their primary company equity.

Apply the 38% rule: of every US Series C company from 2010-2015, only 38% produced value for employee shareholders a decade later. For each founding team member whose company equity exceeds 50% of their net worth, model the outcome if your company is in the 62%. What is their financial position? Can they still afford their life? Do they have enough diversified assets to weather a decade of effort that produces no equity payout?

Map your liquidity options by category: (A) secondary marketplace sale (taxable, commission applies, immediate cash), (B) single-stock loan (expensive, low loan-to-value ratio), (C) exchange fund contribution (tax-free diversification, no immediate cash but LP interest), (D) nothing yet (company not eligible for any of the above). Document which options are realistically available to each team member today.

Identify the vesting cliff: for each equity holder, calculate how much is vested today vs. unvested. Liquidity and exchange fund options require vested shares. Build a timeline showing when meaningful vested equity will be available for each team member - this is the earliest practical window for any diversification action.

Schedule a 60-minute team conversation about equity concentration and financial planning. The goal is not to pressure anyone to sell or exchange shares, but to ensure every founding team member understands the 62% scenario and has considered whether they have enough diversified assets to be financially secure if the company takes a decade to exit or does not exit at all.

List every scheduled touchpoint in your current customer journey — onboarding calls, check-ins, automated messages, renewal conversations, support tickets. Map them on a timeline from sign-up to churn or renewal. For each touchpoint, note whether it is initiated by you, the customer, or triggered automatically by a data event.

Identify every gap between touchpoints. These are the windows where your customer is using your product (or not) and you have no visibility. For each gap, write down: what is the customer likely doing, experiencing, or deciding during this window? What could go wrong that you would not know about until it was too late?

For each invisible gap, ask: what single data point, if you had it, would change your ability to intervene early? Think in terms of usage signals (frequency, depth, feature adoption), sentiment signals (support contact volume, NPS shifts, response rates), and outcome signals (whether the customer is achieving the goal they bought for). Prioritize gaps where the cost of missed intervention is highest.

Rank the gaps by risk: which ones, if a problem developed and went undetected, would lead to churn, a bad outcome for your customer, or a reputational hit? These are your remote monitoring product priorities — the places where proactive alerting would have the most leverage.

For your highest-risk gap, design one lightweight monitoring mechanism: a usage health score, an automated check-in message at day 14, a flag when a customer has not logged in for five days. Build and deploy it. Measure whether early intervention changes the downstream outcome (retention, expansion, satisfaction score) for customers who triggered the alert vs. those who did not.

For any service business that takes inbound calls, log every incoming call for one full week. For each call, record: (A) answered immediately, (B) answered after hold, (C) missed entirely. If you have a phone system with missed call logs, pull the data. If not, ask your front desk or ops team to track it manually. The goal is a real number - not an estimate.

For every missed call, apply your average order or ticket value. Multiply missed calls times average ticket to calculate your weekly phone revenue leak. Annualize it. For most service businesses, this number is larger than expected - Christian Wiens cites 40-50% of restaurant calls going unanswered; the dollar figure per location is typically thousands of dollars per month.

For answered calls, track how often your team attempted an upsell or cross-sell. Calculate your actual upsell rate as a percentage of answered calls. Compare this to what 100% AI upselling would look like: if AI upsells on every call and 30% of customers accept, how many additional dollars per call does that represent over your current baseline?

Add the two numbers together: recovered revenue from missed calls plus incremental revenue from higher upsell rate. This is your gross AI opportunity for phone channel alone. Divide by a realistic monthly AI phone agent cost (typically $200-500/month for a tool like Loman) to calculate your payback period.

If the payback period is under three months, run a 30-day pilot with an AI voice agent on one line or location. Use the same call volume data as your control baseline. After 30 days, compare: average ticket size, answered call percentage, staff time freed from phone handling, and any measurable change in customer satisfaction or complaints.