Course resource
Founder AI Toolkit
Where AI actually helps when you are running something small, and where it will waste your time convincingly.
The honest framing
As a founder you have no time and no team, which makes AI genuinely valuable — and you are also the person most likely to use it to avoid the uncomfortable work.
The uncomfortable work is talking to customers and selling. AI cannot do either, and it is very good at generating things that feel like progress instead.
Where it earns its place
| Task | Why it works |
|---|---|
| Drafting anything | You write constantly and most of it is routine |
| Turning customer conversations into structured notes | You forget otherwise |
| Research before a call | Cheap preparation |
| First-pass document review | Not a substitute for a lawyer, but it tells you what to ask |
| Operational automation | Invoicing, onboarding, reporting |
| Explaining something unfamiliar | Finance, legal, technical — enough to ask better questions |
Where it will waste your time
| Task | Why |
|---|---|
| Deciding what to build | It has no information about your customers |
| Market sizing | It will produce a confident, fabricated number |
| Competitor facts | Invented funding rounds and headcounts |
| Writing a strategy | Generic and unfalsifiable |
| Replacing customer conversations | The thing you cannot skip |
The pattern: it is good at producing artefacts and bad at producing knowledge. A business plan generated in ten minutes is an artefact. Knowing why your last five prospects said no is knowledge.
Validation — done properly
I think [CUSTOMER TYPE] has this problem: [PROBLEM]
My evidence so far: [WHAT YOU ACTUALLY KNOW — conversations, not assumptions]
1. What am I assuming that I have not verified?
2. What is the cheapest test for each assumption?
3. What would I expect to see if I were wrong?
4. What question should I ask the next five people I speak to?
Do not estimate market size. Do not validate my idea.
The output is questions to ask real people. Then go and ask them. That is the entire method.
Customer conversations
Preparing
I am speaking to [WHO] about [PROBLEM AREA].
I want to learn: [WHAT]
Give me 8 open questions that would tell me whether this problem is
real and urgent for them.
Rules:
- No leading questions
- Nothing answerable yes/no
- Ask about what they DID, not what they would do
- Do not mention my solution
The "what they did" constraint is the important one. What people say they would do is worthless; what they already did about the problem is data.
Afterwards
Here are notes from [N] customer conversations: [PASTE]
1. What problem did they describe, in their own words? Quote them.
2. What have they already tried or paid for?
3. Where do they agree, and where do they differ?
4. What did they NOT say that I expected?
5. What is the strongest evidence against my hypothesis?
Do not tell me the idea is promising.
Question 5 is the one that saves founders a year.
The operational stack
Small company, so keep it small:
| Need | Approach |
|---|---|
| Drafting | One good assistant with your Digital Twin |
| Support | Draft-first from a knowledge base |
| Onboarding | One automation, triggered on signup |
| Invoicing and chasing | One automation, draft-first |
| Weekly metrics | One scheduled digest |
| Research | One tool with live search and citations |
Five automations, not fifteen. Every automation is something to maintain, and you are the person maintaining it.
MVP prioritisation
Here is everything I could build: [LIST]
The problem I am solving: [PROBLEM]
What I have actually heard from customers: [QUOTES]
For each item: which customer statement supports building it?
If none, say so.
Then: what is the smallest thing that tests my main assumption?
Do not propose a feature roadmap.
Anything with no supporting customer statement goes on a list you revisit later, not into v1.
Fundraising and finance
AI can draft a deck and structure a model. It cannot tell you your numbers.
- Never let it generate a market size. Investors check, and a fabricated figure ends the conversation.
- Never let it do the arithmetic in a model. Use a tool that executes code, and check the totals.
- Use it for the narrative, once you have the numbers verified.
- Use it to anticipate questions — "what will an investor challenge here?" is genuinely useful.
The founder trap
The specific way this goes wrong: you use AI to produce a business plan, a landing page, a pitch deck and a content calendar in one weekend, and none of it is based on having spoken to anyone.
It feels like enormous progress. It is the most expensive kind of procrastination available, because the artefacts make the avoidance invisible.
The check, weekly:
| Customer conversations this week | |
| Things I sold this week | |
| Artefacts I produced this week |
If the third number is consistently larger than the first two, the tools are working against you.