Won 2nd place in a crazzy hackathon by @Replit
It was super fun.
I built an virality engine by reverse engineering @AIatMeta's research paper on TRIBE v2
all in all a neuro hacking for viral reels.
anyways, thanks to @amasad for giving free credits.
and @networkschool for hosting it.
If you’re in the Singapore/Malaysia area, we have a Replit Agent 4 hackathon tomorrow at Network School.
There will be a tutorial, a little remote guest appearance by my friend @amasad, and a hackathon.
Here’s the event link:
https://t.co/s0JM5sLVSO
i build software for the most boring part of healthcare.
not the drugs. not the robots.
the admin. the phone calls. the insurance. the refills nobody remembers.
everyone wants the exciting part. almost nobody touches the part that's drowning practices.
so that's what we build at https://t.co/WarXqlRILM
AI agents that finish the work instead of dumping it back on staff.
boring problems done well change more lives than flashy ones done loudly.
you don't need another tool.
you don't need a better model, a fancier stack, or one more tutorial.
you need to ship the one idea you've been avoiding for three weeks.
the tools are ready. the only thing missing is you hitting publish.
REPO OF THE DAY
if you want a real shot at an internship or a SWE job, bookmark this.
coding-interview-university. a full, free, self-study plan to go from zero to interview-ready. data structures, algorithms, system design, all mapped out.
300K+ people starred it for a reason.
https://t.co/bXox0AKLrM
I've reached out to 200+ AI companies this year. They all have the same problem.
It's not the tech. The tech is incredible.
It's distribution.
Great product, nobody knows it exists.
The best builders are quietly losing to worse products with louder audiences.
Is that fair? No. Is it the game? Yes.
Opus 4.8 benchmarks are in. It takes the lead on almost everything:
Agentic coding (SWE-Bench Pro): 69.2% vs GPT-5.5 58.6%, Gemini 3.1 Pro 54.2%
Computer use (OSWorld): 83.4%, top of the table
Knowledge work (GDPval-AA): 1890 vs GPT-5.5's 1769, Gemini's 1314
Humanity's Last Exam (with tools): 57.9%, ahead of everyone
One honest gap: GPT-5.5 still wins terminal coding (78.2% vs 74.6%).
Everywhere else, this is the agentic frontier.
Every website and app builder quietly became an AI builder this year.
Webflow, Framer, Bubble, Wix, all of them.
The "no-code vs code" debate is dead. The new line is "describe it vs build it."
Soon the skill won't be knowing the tool. It'll be knowing exactly what you want.
@adcock_brett raised $700M for Hark to build AI that remembers you over time.
everyone's chasing smarter. the real gap is memory.
a model that forgets you after every chat is a tool.
one that remembers is a relationship. that's the next moat, and almost nobody has it yet.
the wildest thing about 2026:
one person with a laptop and an AI agent can now build what used to take a 20-person team.
the barrier to building didn't lower. it collapsed.
the only question left is what you do with that power. genuinely the best time in history to make something.
The money in AI agents isn't going to general chatbots. It's going vertical.
Legal (@harvey__ai), finance (@tryramp), support (@ada_cx), healthcare (@AbridgeHQ).
Pick one painful industry workflow, make it 10x cheaper, win the category.
General is crowded. Vertical is wide open. That's where I'd build.