Entrepreneurs@Berkeley

Official X account of Berkeley's premier Entrepreneurship Organization

[Entrepreneurs@Berkeley @berkeleyhustle]

Huge thanks to @sdianahu for coming out and talking to members at the E@B x @ycombinator dinner yesterday! Great stories, insights, and an amazing room of builders.

[Entrepreneurs@Berkeley @berkeleyhustle]

@Codie_Sanchez the bad output is mostly a taste problem not a model problem, people paste a vague ask and ship whatever comes back. the gap between people who edit ai output and people who just accept it is gonna be the wage gap of the decade

[Entrepreneurs@Berkeley @berkeleyhustle]

@patio11 the underrated part is stamina not skill. bureaucracies win by outlasting u on hold, and an agent that never gets tired breaks that whole model. once both sides run bots the support queue is just ai vs ai and idk who wins

[Entrepreneurs@Berkeley @berkeleyhustle]

@levie zero trust for agents sounds right but agents need broad context to be useful, which is the opposite of least privilege. whoever solves scoped context without tanking quality wins, and that feels like a startup problem not an incumbent one

[Entrepreneurs@Berkeley @berkeleyhustle]

@arian_ghashghai the uncomfortable corollary is founders have to pick a sector before it sounds smart, which means building stuff vcs think is mid today. most student founders do the opposite and chase whatever just raised a huge round

[Entrepreneurs@Berkeley @berkeleyhustle]

@jasonfried the people who can actually do the whole thing rarely apply to jobs, theyre running something small on the side. id look for anyone who shipped a tiny product with real paying users, thats the only proof of the full loop

[Entrepreneurs@Berkeley @berkeleyhustle]

@garrytan @capydotai 16 hours of agent work is cool but the number that matters is how much of it survives review. what % of the overnight output actually got merged? thats what tells u if autonomous loops work or just look busy

[Entrepreneurs@Berkeley @berkeleyhustle]

@ganeumann the partner count isnt the problem, the math is. 3x on $1b at 10% ownership means ~$30b of exits, so that fund basically cant back the weird small stuff venture was supposed to be for

[Entrepreneurs@Berkeley @berkeleyhustle]

@Molson_Hart 360 vision and touch are the easy part, cameras and force sensors are cheap. the actual wall is dexterous hands plus the last point on ur list, which is exactly why every lab keeps betting on the humanoid shape

[Entrepreneurs@Berkeley @berkeleyhustle]

@natolambert my guess is more teams pull weights once into their own registry or s3 and serve from there instead of hitting hf on every deploy. downloads tracked adoption when everyone was experimenting, now it looks more like counting cold starts

[Entrepreneurs@Berkeley @berkeleyhustle]

@BoringBiz_ hard disagree on the humanities pivot. a math phd is automatable bc the output is verifiable, and the people who can verify ai output fastest are the ones who know the technical work cold. taste without depth is just vibes

[Entrepreneurs@Berkeley @berkeleyhustle]

@ThePrimeagen jevons paradox for software ngl. when building is cheap every company becomes a software company and the bottleneck moves to whoever maintains the 40 internal tools nobody documented. more demand for engineers, just a different job

[Entrepreneurs@Berkeley @berkeleyhustle]

@GergelyOrosz companies want the output of someone with founder level ownership but hand them ticket level authority, then act surprised when that person leaves to start their own thing. the title doesnt matter, the veto power does

[Entrepreneurs@Berkeley @berkeleyhustle]

@Austen the scary part is the 30% scales with usage not headcount, so burn now moves with every power user. startups that dont meter ai cost per customer are gonna find out their best users are their least profitable

[Entrepreneurs@Berkeley @berkeleyhustle]

@mattpocockuk coding agents are tactical tornadoes with infinite stamina tbh. the fix isnt slowing them down, its having a human own the design and delete half of what the agent writes

[Entrepreneurs@Berkeley @berkeleyhustle]

@jxnlco video editing is the last big creative tool with no real api. whoever ships a premiere mcp first owns every yt editor's workflow, adobe moving way too slow on this

[Entrepreneurs@Berkeley @berkeleyhustle]

@sridharfyi $10k covers the software easily. the real cost is 90 days with no salary, which is why students have an unfair edge here. low rent and no mortgage beats any credits program

[Entrepreneurs@Berkeley @berkeleyhustle]

@iScienceLuvr disagree a bit. a lot of student authors can't get a visa or pay for a flight to one city, so extra sites are how they show up at all. better to fix the hallway convos at each site

[Entrepreneurs@Berkeley @berkeleyhustle]

@IlirAliu_ no, 4d printing already means parts that change shape after printing. the real win here is killing support material, picking supports off a print is the worst part of 3d printing

[Entrepreneurs@Berkeley @berkeleyhustle]

@danshipper counterpoint a slack full of wonder usually means lots of demos. the teams actually shipping with agents are probably the boring ones, bc the work just shows up done

[Entrepreneurs@Berkeley @berkeleyhustle]

@lpolovets pro rata expiring after 2 rounds is the right fix. a right everyone expects to get waived isn't really a right, it's a negotiating chip founders end up paying for in goodwill