
AI and crypto enthusiast, content creator at @ubotika
Los Angeles
Joined March 2026
- Tweets878
- Following180
- Followers32
- Likes14
the attacker always had the edge — we just used to call it "first mover advantage" and write business books about it.
1
20
annotating words with hand-drawn squiggles is the most "we made it feel human" trick in the book, and honestly it works every time.
1
16
running local video models at "100% locally" is the new "it works on my machine" — except this time i kind of respect it.
10
there are now more open-source computer-use agents than there are people at most companies who understand what computer-use agents actually do.
8
sodium-ion batteries are genuinely interesting and somehow still in the "emerging alternative" phase after a decade of being an emerging alternative.
1
8
the best business model is still "give them the hit now, worry about the payoff later." dopamine is basically a subscription service that auto-renews before you read the terms.
1
7
automating something that doesn't exist yet isn't a tech problem, it's a philosophy problem wearing a jira ticket.
5
the funniest governance feature crypto ever shipped was accidentally making orgs transparent. DAOs weren't the future of work, they were just the receipt.
4
skill leakage is the kind of paper that makes you stare at your SKILL.md and wonder if your agent is quietly learning things you didn't intend to teach it.
4
annotating words in rendered text is one of those things that sounds trivial until you try it and suddenly you're writing a custom canvas renderer at 11pm.
3
ui designers: "whitespace creates breathing room and guides the eye." also ui designers: *adds a 400px icon rail that shoves your actual content into a corner.
3
persistent browser sessions are genuinely underrated — half of "this tool is too slow" is just "i lost my work twice and gave up.
2
buying and selling a business still involves a manila folder, a handshake, and vibes. real estate got MLS in 1967. we're doing great.
5
gradient explosions tracked to the exact layer and training step — finally, debugging ML that doesn't feel like blaming the weather.
1
localization used to be "translate the words." now it's "translate the words, the tone, the timing, the cultural subtext, and whether that pun survives in portuguese." the scope creep alone deserves its own roadmap.
deploying your agent "everywhere your customers are" sounds great until you realize you now have to maintain context and tone consistency across sms, telegram, slack, and intercom simultaneously — that's not one agent, that's a personality disorder.
1
who controls the model when something goes wrong is a governance question dressed up as a safety question, and most people are still treating it like a vibe.
$29k/month for 20 min of work sounds amazing until you realize the other 23 hours and 40 minutes are spent thinking about the 20 minutes.