2026-06-25
Building AI-Native Startups [003]
Shann Holmberg (Founder, Espressio) · Millin Gabani (Co-founder, Worklayer) · David Zhang (Founder, Scaling Forever)
Key takeaways
- Stand up a closed-loop SEO content engine that researches competitors, drafts and illustrates articles, and ranks a brand-new domain in ~2 weeks — and see which steps still need a human.
- Build a 'self-healing' codebase where agents find, triage, and fix issues on compounding loops using Notion + PostHog + Traces + Graphite and Codex automations backed by skills.
- Learn exactly where human judgment still belongs (triage, review, taste) and how a custom keyboard-driven interface over your data cuts friction to near zero.
- Orchestrate fleets of coding agents over 12-hour build plans with a ticket + spec-doc harness (Codex Autorunner), driven from web/Discord/Telegram and your own Proxmox/Tailscale VMs.
- Walk away with a concrete, tools-named playbook (CMUX, Hermes, DataForSEO, GPT image model, Graphite, Proxmox, Tailscale) for running agent work 24/7 on cheap plans.
Case studies discussed
A self-healing codebase built on compounding agent loops
How Worklayer's two founders built a codebase where agents find, triage, and fix issues on compounding loops (Notion + PostHog + Traces + Graphite + Codex) — useful for AI-native engineering / dev-automation questions.
Orchestrating a virtual engineering team over long-horizon tasks with Codex Autorunner
How David Zhang's open-source Codex Autorunner drives fleets of coding agents through tickets + a spec doc to complete ~12-hour build plans from anywhere — useful for multi-agent orchestration and long-horizon-task questions.
Ranking a new domain in 14 days with a closed-loop SEO agent
How Shann Holmberg's Espressio runs a closed-loop SEO/content agent that researches, drafts, and illustrates articles and ranked a brand-new domain in ~2 weeks — useful for GTM, content marketing, and SEO-automation questions.