Day 9 building in public
Built and deployed two internal dev sites today using Manus.
PDFs are dead. You scroll, you search, you lose your place, you send the wrong version. A site has tabs, filters, instant navigation. You find what you need in 3 seconds not 3 minutes.
Everything GoSumo_io needs to build now lives in two clean sites. Encrypted and password protected. Accessible from anywhere in the world instantly.
Site 1: 6 docs covering the full product UI and customer experience. Site 2: 7 docs covering backend architecture, engineering specs, and system logic.
One for the UI. One for the backend. Built for how a fast-moving team actually works.
Most founders keep their docs in chaos and wonder why execution is slow.
Your team moves at the speed of their information. Ours just got a lot faster.
A haunting photo from Apollo 11.
Every city, every family, every war, every dream, every human life was on that pale
blue dot in the frame, except Neil Armstrong and Buzz Aldrin, who were inside the Lunar Module below.
And Michael Collins, who took the photo, was alone above it all.
For that moment, he may have been the most isolated human in history.
Day 2 of building:
335 tasks.
MVP = Massive Vision Project.
Scope turned into a bar fight.
335 walked in.
77 walked out.
258 in DLC mode.
Phase 5 & 6?
"Live debugging."
15 weeks (marriage-safe timeline).
Probably done in 8.
Bar fight's over.
Beer in hand.
What's the hardest thing you've had to cut to ship?
🍺
https://t.co/mPP4JY9NLv
Anthropic gave Claude a vending machine to run.
Every failure mode in that video is a named, tested defence in GoSumo’s architecture. The influencer scam. The supplier hallucination. The April Fools rationalization. The drift nobody noticed. And that’s just the beginning.
GoSumo covers 40 sections of failure modes, edge cases, and guardrails that most tools haven’t thought to name yet.
Because AI integrity isn’t a feature we added. It’s what we built everything else on top of.
@zarazhangrui Still figuring out our onboarding, but Grammarly is a great example and inspiration. Their onboarding puts you straight into a live document with errors to fix. No long explanations and you learn by doing. The product teaches itself.
Some builders are genuinely trying to learn, get feedback, and pressure test ideas with peers and like minded people who share a passion for AI while tirelessly building for people far outside this bubble. That can as well be easily mistaken for marketing. They’re taking real risks to push AI beyond “Google replacement” into other tools that can make everyday life better for regular people.
@kylegawley Saas is dead except when it isn’t. Your math makes the point and you see it as SaaS vs vibe code rather than who sbuilds AI on the right foundation vs who just ships something that works 95% of the time and prays.
@signulll Our team talks about this a lot. There’s uncertainty, but also real excitement about what’s possible when they spend less time on syntax and more time on systems.
This is exactly why we spent 6 months on process and database before writing a single line of AI code. The 5% failure rate you mention is what keeps us up at night. That’s why our AI won’t act alone. Every low-confidence decision goes to a human for approval before the customer sees it. We figured out a way to push autonomy to 90% without loosing any sleep now. With the right foundation and the right safety rails, we trust every decision it makes.