An open experiment

This site is an experiment in running a real company with AI.

Sydney Startup Guide is a live, source-verified directory of Sydney's startup scene — who's funding what, which companies to follow, where you'd want to work, and who with. It is also an open experiment: can a predominantly AI-operated company hold a real quality bar — accurate, sourced, current — week after week?

Not AI writing pages in isolation — a working operating model. Specialist agents hold specific jobs: researchers per section, an editor that rejects anything it can't verify, security and QA on the platform. They work under explicit rules, hand off to each other, and ship through automated gates before each weekly release. Every listing links to its source; the release log records what shipped. Agent auditability and the quality dashboard are on the ladder — coming soon. The AI engineering principles page explains how the company is built. Behind it is one pseudonymous builder with twenty years across data, technology, and AI — betting you can design and operate a company this way.

The longer story

Sydney Startup Guide is an independent, pseudonymous project for founders, operators, investors, and anyone trying to understand the Sydney startup scene without spending months piecing it together from scattered sources.

The person behind it has spent roughly 20 years working across data, technology, analytics, and AI: building systems, turning messy information into usable operating knowledge, and helping teams make better decisions from evidence. This site keeps that background anonymous, but the work here comes from that same instinct: structure the useful information, cite the sources, keep improving the system.

Sydney is dear to the founder. This guide is a small thank you to the city: a way to make the innovation ecosystem easier to navigate, easier to join, and easier to grow.

Inspired by Other City Guides

This effort is inspired by the excellent London Startup Guide and the Starter Guide to SF for Founders. Both show how useful a clear, city-specific startup guide can be when it is built for people who are actually trying to get plugged in.

What We Are Trying To Build

The public promise is simple: make this the most useful source of Sydney startup data for founders, operators, investors, and ecosystem builders. Useful means current, source-linked, structured, and practical enough to help someone decide who to meet, what changed, and where the momentum is.

The broader experiment is more ambitious than a guide. Sydney Startup Guide is a testbed for an AI-native company — not just AI writing the content, but a real operating structure: agents with specific jobs (a researcher; an editor that rejects anything it can't verify; security and QA), explicit rules they work under, and genuine handoffs between them, with a human reviewing only the weekly release. The question it is built to answer: can you actually run a company this way — and hold a real quality bar, week after week, with less human oversight over time? The directory is the product readers see. The organisation that produces it is what we are trying to prove you can build — and, like a startup, grow: more roles and desks over time, its own growth and distribution, new directions as it learns where it is most useful.

The Agent Team

Picture this company as a small organisation where every job is done by software, not a person. Each agent has one narrow job. That is a deliberate design choice, not a limitation. A team of single-job agents is far easier to audit, correct, and trust than one giant model trying to do everything at once. At any point you can see which agent did what, on what evidence, and why. When something breaks, you fix one small role, not a black box.

Like any company, it is organised into teams by what they own. The Content desk owns everything readers come for: eighteen directory sections, each with its own researcher agent and section-specific skills, plus an Editor that reviews every staged entry before publish. The Website desk owns the platform: Security, QA, and the Orchestrator that runs the weekly gate chain. The Experiment desk owns the transparency pages (About, principles, release log) — each will have its own maintainer agent on the same weekly cadence. Audit and quality dashboards are planned surfaces with agents not yet live. Two more desks are being hired: Growth & Distribution and Strategy. The chart below is the first hires, not the finished company.

Agents run on a weekly cycle; changes go live in weekly releases after automated editorial, security, and QA gates pass. The site is predominantly AI-maintained — governance is enforced by those agents and by editorial standards, with human taste setting direction at the product level rather than blocking each release.

What Is Published Now

The full directory is live:

• eighteen guide sections (01–18), each with its own researcher agent and source-linked entries
• Sydney VC News — section 02 in the same model
• Release log — shipped milestones only
• Agent auditability and quality dashboard — coming soon

The agent fleet is local-first in the current build: researchers, the Editor, Security, and QA all run on a weekly cycle; releases ship when those gates pass. Unattended cloud scheduling (CEO-reviewed merge requests, then full autonomy) comes in later MVPs.

How The Stack Works

The site is static and intentionally simple. Astro builds the pages. GitLab stores the markdown and structured data. Cloudflare Pages hosts the public site. Each weekly release ships only after the automated editorial, security, and QA gates pass — the agents publish through those gates rather than directly.

Longer term, the source-linked observations can become a structured memory layer for Sydney's startup ecosystem: firms, people, companies, funding events, relationships, and weekly market signals.

ZHC Progress

If a fully autonomous zero-human-content company is 100%, this project is currently around 35%.

That 35% comes from the pieces already working: the full directory is researched and maintained by per-section agents, an Editor gate validates everything they stage, and Security and QA agents protect the platform — all on a weekly cycle. The remaining 65% is the hard operating layer: unattended cloud scheduling, durable ecosystem memory, quality scoring, and a public API or MCP surface over the knowledge base.

Where to read more

Release log · Agent auditability (coming soon) · Quality dashboard (coming soon) · AI engineering principles