In this episode:
- Peter bought memo.sh (a $1500 domain) about fifteen minutes before recording, mostly so he could announce it on the pod
- What he's building: a shared memory layer for AI tools, so Claude, ChatGPT, Codex and the rest stop making him repeat himself
- His latest solo retreat, the Suhm ApS "board letter", and why doubling down on network beat doubling down on software
- Niklas is back from San Francisco energized, hiring another engineer, and pulling his ops hire's start date forward to Monday
- Why terms and privacy policies actually matter when you're venture-funded (it's the due diligence, not just the text)
- AEO/GEO: how comparison pages and FAQs are quietly how AI decides what to recommend
---
Niklas makes a new podcast rule on the spot: if someone just bought a domain, that's where the episode starts. Handy, because Peter had spent $1500 on memo.sh maybe fifteen minutes before recording, partly so he could talk about it here. The idea came out of a recurring annoyance. He uses a bunch of different AI tools (ChatGPT on forest walks because the transcription's better, Cursor or Codex or Claude Code back at the desk) and they never share context, so he keeps explaining the same thing twice. On his last retreat he vibe-coded a fix he calls Memo: a shared space where his agents write stuff down and read it back later. Codex one-shotted a Laravel app off a spec he'd built with Fable, wired up to the MCP and Passport packages he worked on at Laravel, and it just worked. He used it nonstop for two days and it stuck. The waitlist is up now at memo.sh. Niklas thinks fixing context across agents is going to be huge no matter who wins, and got a Turbopuffer plug in while he was at it.
The other big thread is the retreat. Peter does these solo (a hotel, ten thousand words of notes, Claude helping him spot the patterns) and writes himself a board letter to Suhm ApS, where the board is just him. The takeaway, again: every time he thought the opportunity was software, the thing that actually paid off was network and personal brand (the conference circuit, the speaking, working with Niklas, this podcast). OG Kit stays the perfect side project, not the main bet. Niklas suggests publishing a public version of the letters, Berkshire-style.
Then it's Niklas's turn. He came home from SF fired up, mostly from seeing companies further along and wanting to pick up speed. They're wrapping a big telephony infrastructure project (important, but invisible to customers), hiring another engineer, and his new ops person now starts Monday instead of after summer. He also gets into why legal isn't optional when you're venture-funded: terms and privacy policies are the contracts that get picked apart in due diligence, and AI-generated ones come back to bite you. Plus a rant about the FCC robocall program sending him more spam than it stops.
They close on AEO (answer engine optimization), following up from last time. The short version: AI recommends products off bite-sized comparison pages, FAQs, and testimonials with a clear source of authority. Tactics that are dead for SEO but work great for LLMs. Probably something both OG Kit and Phone.inc should build.
---
Still Early is sponsored by
Phone.inc: a business phone number, welcome greeting, call routing, and opening hours, all from an app on your personal phone.
