How I build

What that means in practice

• I design it. I decide what the service does, which services it uses and where the data goes. The terms and privacy notice for the inbox will say so before it opens.
• AI coding tools write the code. I direct them and check what they produce. I tell you which tools I use. Today that's Claude Code and Cursor.
• Tests run on every change. The code has automated tests. They run every time the code changes.
• A second AI model reviews the work. A different AI model checks the code, separate from the one that wrote it.
• No access to your accounts. The inbox works from forwarding you control: calls you don't answer, a copy of new email inquiries, and a form on your website. I don't need your passwords or an account in your software, and I never need to be on site.
• I test your setup before it goes live. A test call, a test email and a test form entry, each checked end to end.
• AI is inside the service too. An AI model turns voicemails into text, sorts inquiries by urgency and drafts replies. AI makes mistakes, so you see every inquiry with its original message, and you send the replies yourself.
• I'm responsible. Using AI tools doesn't shift the responsibility. What I deliver is mine to answer for.

What is not in place today

These checks are not in place today.

• Automated security, dependency and secret scanning. Not in place today.
• An independent human security review for anything that touches payments or sensitive data, before it goes live. Not in place today.
• Separate test and live setups, with no AI tool able to reach your live data. Not in place today.

What I've shipped this way

I built my own apps by directing AI coding tools. Kalum, my international calling app, is live on Google Play and the App Store, with real payments, fraud controls and sanctions compliance. Play Lounge is a real-time app with 1,600+ installs. Neither has an AI phone assistant.