I've spent years watching people try to design a t-shirt.
It usually goes like this:
Open a "design tool" → get attacked by 400 clipart flames → give up → order the boring one.
So we're building Baski 👕✨
You just describe what you want.
"A moody cat astronaut in retro '80s colors."
That's it.
Baski creates the design, puts it on a real product, and gets it ready to ship all in under a minute.
No design degree. No clipart flames.
Basically, you talk. Baski makes the shirt.
And we're almost ready to let people in. 👀
This is our very first post. We're launching soon, and early access is already filling up.
👉 Sign up at https://t.co/2jxOFWhgpi and be one of the first to try it. Early users will get some great perks.
One question before we launch:
If you could create any t-shirt just by describing it, what would you make?
Drop your idea in the comments. My favorite one will be generated on launch day. 🎨
#AI #PrintOnDemand #ComingSoon #BuildInPublic #StartupLife #GenerativeAI #Ecommerce #FounderJourney #EarlyAccess #Waitlist
OpenAI just dropped GPT-Live. Voice AI that listens and talks at the same time. It says "mhmm" while you speak.
Everyone: "RIP voice startups."
Me: the API isn't even out yet. The window for builders is wide open.
One year ago, Chinese AI models were 2% of OpenRouter traffic.
Today: 45%.
The #1 most-used model on the platform is from Xiaomi. Not OpenAI. Not Google.
Nobody announced this shift. Price did it quietly.
#ai#china#xiaomi
1/ Everyone read the jobs report wrong.
Yes: 57k jobs vs 185k expected.
Yes: AI is the #1 stated reason for layoffs.
But companies that use AI the MOST are hiring MORE people. +10% headcount.
2/ AI doesn't kill jobs. It moves them from slow companies to fast ones.
Pick your employer like you pick your stocks.
Claude Sonnet 5 just made near-flagship agents ~85% cheaper.
The models are no longer the moat. Everyone has them.
Speed, focus, and knowing your users that's the moat now.
Building Baski taught me this fast.
Everyone is watching the wrong GPT-5.6 model.
Sol is the flagship. But Luna — fast and cheap (~$1/$6) — is the real story.
Most tasks are simple. You don't need the biggest brain. You need the cheap one that just works.
Match the model to the job. Ship faster, pay less.
Oracle confirms ~21,000 jobs cut to AI. In its annual regulatory filing (surfaced June 22–23), Oracle disclosed its headcount fell to 141,000 from 162,000 a year earlier — nearly 13% — and stated plainly that AI adoption "resulted, and may continue to result, in reductions to our workforce," warning of more to come.
The White House signed a new AI order: labs are asked to give the government early access to top models, up to 30 days before launch.
It's "voluntary" — for now.
Lesson for builders: policy is now part of your tech stack. Don't depend on one model.
#AI#Startups
Google's Gemini 3.5 Pro can read 2M tokens. Everyone is cheering the number.
But it costs ~10x more than Flash — and for most jobs, Flash is faster and good enough.
Lesson for builders: pick the model that fits the task, not the biggest headline.
#BuildInPublic#AI
Most underrated thing in my dev workflow: a "devil's advocate" skill that argues with my own specs.
I write a plan, start building, and weeks later half the spec is stale. Instead of shipping the wrong thing, the skill checks the plan against what I've actually built and tells me what's outdated.
Refine, then build. Try it.
ChatGPT just dropped below 50% market share for the first time since 2022.
Late May 2026 (Sensor Tower):
•ChatGPT 46.4%
•Gemini 27.7%
•Claude 10.3%
Raw users still favor OpenAI (1.1B MAU). But when every assistant answers everyday questions about the same, people drift to whoever owns their ecosystem—or whoever they trust more.
The next fight isn't the chatbot. It's the agent.
A new model is shaking up my AI bill.
Qwen 3.7 Max matches Claude on many coding tasks for much less money.
Building Baski taught me: don't marry one model. Use a router cheap model for easy jobs, strong one for hard ones.
Watch your AI costs. They decide if you survive.
Everyone is talking about GPT-5.6. But no one has actually seen it.
Huge "leaks" this week: 1.5M context, big reasoning jumps. A betting site says 83% chance it ships this month.
The catch: OpenAI said nothing. Most leaks have no real source.
Don't plan your week around a model that doesn't exist yet.
Every day a new AI model drops. I stopped trying to use them all.
Google, Anthropic, xAI, OpenAI — all shipping at once. It's a lot.
Building Baski taught me one thing: pick a stack and keep shipping.
You don't win by trying everything. You win by finishing something.