I've had the opportunity to work on several "Startups within Enterprise Companies" in the past few years. I wrote an article on the 3 types of teams you'll find. https://t.co/jelw61Qo8S
So 3 @AirCanada agents lied to my face today. Apparently there were no seats left on any flights to any airports at my final destination, but the kind @united agents repeatedly told us they had empty seats. Now we are stuck and my infant has no diapers left.
Bought HeroUI Pro. Built guardrails so agents would actually use it. Without rules, they default to generic markup despite premium components. Workflow: design guide first, system components, then code. Tools are easy. Making them stick takes discipline.
Little Ghost Coffee's site just got a full brand overhaul โ custom SVG mascot, icon set, color system, private events booking.
A coffee shop's site shouldn't look like a SaaS landing page. It should feel like walking in.
What brand detail made you take a place seriously?
Migrated a coffee shop site from Next.js to Astro 5. Zero JS shipped by default. Build went from minutes to seconds. Deployed to Hostinger via FTPS โ no Vercel bill.
The best architecture fits the problem, not the trend.
What simple tool solved your hardest problem?
Built a med spa marketing site where every treatment has a public price. Botox $12/unit. Lip filler $650. No "call for pricing."
The team was nervous. I said: if you can't price it publicly, you don't understand it well enough.
What industry needs more transparency?
Vendor integrations are lock-in risk. I run one test:
"If this service disappeared tomorrow, what would my customers lose?"
If the answer is anything critical, the architecture is wrong. Vendors should be adapters, not foundations.
What is your vendor dependency test?
Built it because I kept finding integration bugs at 11pm on deploy day. Now the agent tells me what's broken before I ship.
What's the most boring tool that saved you the most pain?
It runs against localhost after every merge, clicks every flow, checks every input binding. Nobody tests this systematically after a component swap. Not even senior engineers.
My backend returned 502 for days. Railway served a stale pre-PR build. A Vercel env var had an embedded newline from a 21-day-old copy-paste.
Zero alerts. Dashboards looked fine. Infrastructure drift doesn't crash. It limps.
How do you catch it in your stack?
The hidden cost of cheap infra: 50 signing certs in one repo. One typo means signing a client app with the wrong identity.
We moved to per-project isolation with EAS Build. Costs more, but never shipping to the wrong org is worth it.
What cheap tool almost cost you the most?
I've corrected the same shell script problem 3 times now. Every time: 80 lines of bash with JSON parsing "glue code". It's not glue. It's a time bomb.
Shell is for one-liners. Everything else is TypeScript.
(Saves production bugs from bash edge case failures.)
I've been working across a few different projects over the last year, some SaaS and some consulting with SMBs.
What ties them together? AI tools in every single one.
Nothing super trendy. Just actually useful.
I figured I should start sharing more of these experiments.
@andrejsshell Are there any plans to have better webhooks for @usekaneo? I'd love to use it with my agents, but polling the system for changes feels wasteful.
Today in Parliament, I called on the Government to impose sanctions on Israel.
Over 50 MPs from seven parties have now backed my motion calling for sanctions.
Thatโs the only way to force Israel to stop committing war crimes and to back a ceasefire.