WEEKS 2-4
Weeks 2-3 had a conscious trade: slow down, then strike. A week of hardcore research on the world's fastest-growing SaaS startups shaped an architectural base that's solid, defensible and survivable - ready to scale, not just to demo. More time again went into sharpening every feature for the exact customers we intend to serve (no spoilers).
A framework that complies with our core values, not just raw abilities. Governance. Checks. Documentation. Review. Security gates. AI guardrails and harnesses to suit.
The payoff: now we build, at extreme pace. Week 4 has had several hundred git commits and 30+ product features live in the first 4 days of committed build. We're past MVP internally, and aiming slightly higher for public release.
Week 5+: nicer clothes. The UI has been wireframe. The next focus is all about that landing page and user journey.
- ✓Startspace induction done - the desk is real.
- ▲Application to Stone & Chalk APPROVED - residency joined. A resident desk inside Australia's largest innovation community: coworking, mentors, the lot.
- ✓AWS stood up, Activate credits applied for.
- ✓Rule of the fortnight: build the MVP. No more reading about Y Combinator.
- ✓AWS Activate founder credits approved.
- ✓Trial day at Stone & Chalk.
- ✓EOFY accounting & tax updates - the prize for incorporating two weeks before end of financial year.
- ✓A couple of days down sick, then straight back in.
- ✓Full repass on design and the MVP stack - spot-checked against Anthropic’s new Fable model.
- ✓First day at Stone & Chalk - induction done, both spaces now live.
- ✓Cloud dev box stood up on AWS Lightsail - a disposable workstation, build from anywhere.
- ✓GitHub org wired with CI on every change · one monorepo for everything.
- ✓Public site on Cloudflare, moved to proper automated deploys.
~/founder $ setup --cloud-dev --ci --monorepo --auto-deploys
✓ dev box live ✓ ci green ✓ repo unified ✓ deploys automated
- ✓Full stack & progress review, benchmarked against YC-grade buildouts.
- ✓12 AI subagents spun up to research, assess and recommend.
- ✓Verdict: on track. No course changes, no new research needed, 47 minor implementation notes.
- ✓Equivalent spend once usage-based pricing lands: ~$100
- ✓The core application flow now runs end to end - behind the wall, invite-only.
- ✓Product rails live: Vercel for the app · Supabase for data, pinned to Sydney · AWS Bedrock for the AI.
- ✓The app watches itself: monitoring, alerting and automatic self-checks are live.
- ✓Public site: terms & privacy published · search/AI-crawler groundwork shipped · Ship Log 01 posted.
- ✓Every piece of data is classified by type, and every AI task carries a sovereignty gate: a model that doesn't meet the bar for that data class is mechanically unselectable - enforced in code, not policy.
- ✓A built-in routing matrix picks the best acceptable model per task, with automatic fallbacks and manual overrides - governed end to end.
- ✓The moment a better model emerges, we switch - no rebuild, no redeploy. Best result for the customer, always.
- ▲As of today (14 Jul '26): no other competitor claims this fame. It's okay. They'll get there.
route: every task → best acceptable model ✓ sovereignty gate ✓ fallbacks armed ✓ override ready
# personal documents stay in-region - the gate makes anything else impossible
POLISH, THEN PEOPLE
The machine is built and the MVP runs end to end. Polish the flow, then open the door a crack - first invited testers. #Buildinpublic
