How Theanna was built: a non-technical founder, her engineers and Claude Code

Theanna started as a platform that takes women founders from idea to $1M in annual recurring revenue: a four-phase journey with 41 milestones, AI guidance, a focused "Build Mode" workspace and Stripe revenue tracking.^[1] Its site now pitches it more broadly, as a growth tool for founders that reads their real numbers and picks and runs go-to-market experiments with them.^[13] Its founder, Nomiki Petrolla, calls herself a non-technical solo founder and a mom of four in Ohio.^[4] She has been writing up the whole journey in public, with the current ARR printed at the top of every post.^[12] That makes Theanna unusually easy to study. This piece pulls together what she has published about how it was built, using only her own posts.

One thing to know up front: Theanna is not a "built in a weekend with AI" story. The first version was built by a hired developer, and she works with engineers today. What's interesting is the way she uses Claude Code to own the whole user-facing side herself, and the business she built around it.

Before Theanna: a program that didn't scale

Theanna grew out of PDS Lab, a six-month accelerator for non-technical women that Nomiki ran at $6,500 per person using a six-step framework from 15 years of building technology for startups. Forty-three women went through it. It worked, she says, but it didn't scale: she could only support so many founders one-on-one. So she wound it down and set out to build software instead.^[1]

The first version: hired help and a long wait

She came up with the idea in October 2024 and spent that month designing the platform: wireframes, user flows and feature priorities. Development started on November 5, 2024, and Theanna launched at the end of February 2025.^[2] She deliberately kept that first version small: "No AI. No fancy gimmicks."^[2]

Waiting for it was, she says, one of the three things she hated most about her first year. She built the product before AI tools were good enough for her to code it herself with confidence, so she had to hire someone.^[3] Meanwhile she was holding three jobs (consulting, university teaching and coaching) to feed a family of six, working 14 to 16 hours a day. She was able to quit all three by May 31, three or four months after launch.^[3]

Launch: selling before the product existed

She didn't sit and wait for the build. For the four months before launch she posted on social media every day, always calling out exactly who it was for: women building software. By launch day about 950 people were on the waitlist, all matching that profile.^[2] On launch day 500 of them signed up and created a profile, helped by beta pricing and a free tier she no longer offers. Within 60 days she had 100 paying customers, with no paid ads.^[2]

The dip nobody sees on the chart

About three months after launch Theanna hit roughly $7K MRR. Then the line went down for five months, and she lost around $25,000 in ARR.^[7] Two things were behind it. A hire brought on for architecture, security and agentic AI turned out to be wrong for the company within three weeks; she says she should have acted in week one.^[7] And during a stretch of fast testing, the domain's DNS records got mismatched, so emails slowly started landing in spam or not arriving at all. Leads slipped and onboarding broke quietly, and she spent months looking at churn and content before finding the real cause.^[7] Once both were fixed, the numbers climbed again.^[7]

The build today: Claude Code for every screen

By early 2026 she had changed how product gets made. She used to design every screen in Figma, write requirements documents and hand them over, and the result never quite matched what she had in mind, after rounds of back and forth.^[4] Now she builds the frontend herself:

Her core rule is "pull down, don't create new": she works inside her engineers' existing codebase from GitHub, so the handoff is clean, and she never touches the backend.^[4] Theanna has separate repositories for the backend, the user-facing frontend, the B2B admin, registration and the marketing site, and she only works in the user-facing one.^[4] Her process, step by step:

1. 1

   Pull the repos down

   Clone the engineers' repositories to her machine and work in the development environment, which no customer ever sees.^[4][5]
2. 2

   Talk in plain English, with a screenshot

   She doesn't know what components are called, so she describes the change and attaches a screenshot of the page; Claude Code finds the right files.^[4]
3. 3

   Iterate on a local server

   She asks to "show me the UI", then tweaks layout, flows and new functionality on localhost until it feels right.^[4]
4. 4

   Commit against a Linear ticket

   She gives Claude Code the ticket number, has it push the code to GitHub and write detailed requirements into the ticket, including what the backend needs.^[4][5]
5. 5

   Hand off to an engineer

   Her lead engineer audits the code, wires up the backend and handles performance.^[4]

The first feature built this way, challenge prompts, went to production in under two days. The engineer had no questions about the frontend, and she says the same feature would previously have taken at least two weeks of back and forth.^[4] For bigger work, like a full frontend refresh, she makes a master ticket in Linear and one ticket per page.^[4]

She is clear that this is prototyping with intent, not shipping straight to users: every frontend piece goes to her engineers, who audit it, fix it and tighten it up before production.^[6]

Where Lovable fits

She uses both Lovable and Claude Code and rates them 1 and 4 out of 10 for complexity.^[5] She built Marble Jar, a chores-and-kindness app for her kids, in Lovable in three hours, and says she recreated the whole Theanna website in Lovable in a single day, wiring up 12 services including Stripe, Resend, Customer.io, Google Analytics and Microsoft Clarity.^[5]

The core product stays in Claude Code for a practical reason: she writes that Theanna's backend is built in Angular while Lovable and Replit build in React, so Lovable code wouldn't fit her engineers' codebase. Her three engineers use Claude Code too.^[5] Her rule of thumb: Lovable for quick tests, standalone side projects and fast landing pages; Claude Code for the core product, work with engineers on GitHub and anything across dev, staging and production.^[5] Her warning about Lovable is that people click "approve" on its build plan without reading it, then can't explain to an engineer six months later what they built.^[5]

The stack, with prices

In March 2026 she listed every AI tool she pays for: about $241 a month in total.^[6] Here's that list together with the rest of the stack she has described:

Claude Code for marketing, too

She built the blog herself with Claude Code on her own domain rather than Substack or Beehiiv, so every post adds to theanna.io's search presence, and she publishes two to three posts a week.^[8] Three weeks and 10 posts after launching it, she reported 89 pages indexed by Google, 53,200 impressions, 439 clicks and an average position of 6.9, and flagged her 0.8% click-through rate as the thing to fix.^[9]

For email, she had Claude Code build reusable branded HTML templates and connected Claude to her Gmail and Google Calendar through MCP, so it already knows the week's events when it drafts. She pastes the HTML into Customer.io and sends 16 to 20 segmented emails a month with no marketing team; each person only gets the four to six that apply to them.^[10]

The numbers so far

On February 19, 2026 she reported $182K ARR and 212 paying customers.^[1] A week later it was $199,584 ARR, or $16,632 MRR, growing 11.8% month over month.^[5] Her latest post, on June 1, 2026, puts Theanna at $232K ARR, after the blog showed figures between $233K and $237K through late April and May.^[11][12] All of these are her own figures.

Timeline

1. Before Theanna

   PDS Lab: a $6,500, six-month accelerator; 43 women go through it.^[1]

2. Oct 2024

   The idea, plus a month of wireframes and user flows.^[2]

3. Nov 5, 2024

   Development starts, with a hired developer.^[2][3]

4. End of Feb 2025

   Launch to a waitlist of about 950; 500 sign up on day one.^[2]

5. About 60 days later

   100 paying customers, with no paid ads.^[2]

6. About 3 months after launch

   Roughly $7K MRR, then a five-month dip that costs about $25K ARR (a bad hire and broken DNS).^[7]

7. Feb 19, 2026

   $182K ARR and 212 paying customers; she starts documenting the journey weekly.^[1]

8. Feb 25, 2026

   She publishes her Claude Code frontend workflow: features in two days instead of two weeks.^[4]

9. Jun 1, 2026

   $232K ARR.^[11]

Lessons you can steal

1. 01

   Market while it's being built

   Four months of daily posts aimed at one clear customer turned into a launch-day waitlist of buyers, not supporters.^[2]
2. 02

   Work inside the real codebase

   If you have engineers, pull their repos and build there instead of starting something new, so the handoff is just wiring.^[4]
3. 03

   Screenshots beat jargon

   You don't need to know a component's name. Show Claude Code the page and say what should change.^[4]
4. 04

   Let the AI write the ticket

   Have it write the requirements and backend needs into the ticket alongside the code, so engineers don't need to ask.^[4]
5. 05

   Pick the tool by codebase

   Lovable for quick standalone builds, Claude Code where the code must fit an existing stack.^[5]
6. 06

   Watch the boring plumbing

   A DNS mismatch quietly broke email delivery for months. Check deliverability when numbers dip for no clear reason.^[7]