Six months ago I couldn’t have told you what an API was.
Today I have three AI-powered features running in my product. I built all of them without writing code. My co-founder — who is technical — has reviewed all of them and told me they’re production-ready.
This is not a story about being a secret genius. It’s a story about the tools that exist right now that make this possible for any founder willing to learn the workflow.
The Stack I Use
Claude API — the intelligence layer. Every AI feature I’ve built calls Claude under the hood. It’s the most reliable, most capable model I’ve tested for business use cases.
Bubble.io — the product layer. Full web application builder, no code required. Has a native API connector that talks to Claude directly.
Make (formerly Integromat) — the automation layer. When something happens in my product, Make orchestrates what happens next — including calling the AI.
Airtable — the data layer. Stores inputs, outputs, and anything the AI needs as context.
That’s the whole stack. Four tools. No engineers.
The Three Features I Built
Feature 1: AI-generated job descriptions
A user fills in a form — role, stage, key outcomes, red flags. Bubble sends that to Claude via Make. Claude returns a formatted job description. It appears in the user’s dashboard in under 10 seconds.
The prompt I use:
“You are an expert at writing startup job descriptions that attract the right candidates and repel the wrong ones. A founder has given you the following inputs: Role: [role]. Company stage: [stage]. Key outcomes this person owns: [outcomes]. Red flags — who should NOT apply: [red flags]. Write a job description under 400 words. End with a ‘You might not be a fit if…’ section. No corporate jargon. Write like a smart founder, not an HR department.”
Feature 2: Interview question generator
Same pattern. User inputs the role and candidate background. Claude generates a structured interview guide with STAR questions, situational questions, and a scoring rubric. Delivered in the product instantly.
Feature 3: Weekly ops summary
Every Friday, Make pulls the week’s activity data from Airtable, sends it to Claude with a prompt asking for a summary, and emails the result to the user. Zero manual work.
How to Build Your First AI Feature
Start with the simplest possible version. One input, one output, one Claude call.
Step 1: Decide what the feature does in one sentence. “User pastes a job description → gets back 5 screening questions.”
Step 2: Write the prompt. Test it in Claude’s interface until the output is consistently good.
Step 3: Build the form in Bubble. One text input, one submit button, one output field.
Step 4: Connect it via Make. Bubble form submission → Make workflow → Claude API call → output back to Bubble.
Step 5: Test with real users. Give it to 5 people and watch what breaks.
The whole process takes 2-3 days the first time. The second feature takes 1 day. By the third you have a system.
The Prompt That Helps You Design Any AI Feature
“I want to add an AI feature to my product that helps [user type] with [specific task]. The user will input [what they provide]. The output should be [what they get back]. Help me: (1) write the system prompt that will produce consistently good output, (2) identify the 3 most likely failure modes, (3) suggest how to handle cases where the AI output is poor, (4) estimate how much this would cost per user per month using Claude’s API at current pricing.”
The Honest Caveat
This approach has limits. Complex features with real-time data, custom model training, or high-volume throughput eventually need an engineer. But for a first version that proves the concept and gets you to paying customers — no-code AI is more than enough.
Build the thing. Get the customers. Raise the money. Then hire the engineer to do it properly.
Want the Full Product & Growth Prompt Library?
13 prompts covering AI feature design, validation, launch, customer research, and more — free download.