Andres Lockin retweeted
Introducing @ValnceCompute for Startups.
Valnce gives startups GPUs on 6 to 24 month contracts, on the terms frontier labs get.
Dedicated GPUs are sold on contracts that run for years, with money up front, because the lenders who finance them need years.
A startup that just raised can commit to one. So it pays about twice as much, or goes without.
Someone has to stand in between: sign the long contract with lenders and neoclouds, and give startups a shorter one.
That is Valnce. B200 and B300 nodes, from one node, underwritten on venture backing and runway, not years of credit history.
One startup is a risk. A portfolio is a credit. So we build each one on purpose, from a mix of startups.
Our first portfolio is oversubscribed. Startups have already asked us for more than 512 nodes. The next portfolio is forming now.
Every giant started with access. It's time to open access again.
Andres Lockin retweeted
Same chip, same data center, 35% cheaper. The only difference is who can sign for three years.
Andres Lockin retweeted
Introducing: Phone Harness MCP 🔌
Any agent that supports MCP can now control a real iPhone or Android in the cloud:
> works in Claude, ChatGPT, Cursor, Codex, and more as a custom connector
> nothing to install, just paste a URL and sign in
> sees the screen, taps, types, and swipes like you do
Available in phone-harness 0.3.3.
github.com/phone-harness/mcp
Andres Lockin retweeted
Watch an autonomous Claude Code agent loop optimize another AI agent built with the Vercel AI SDK.
Andres Lockin retweeted
Met Jensen Huang at GTC.
“Thousands of agents or your company NGMI.”
That’s it.
That’s the playbook.
Andres Lockin retweeted
Everyone's building browser use. We built phone use for the @browser_use hackathon at YC. All night. No sleep. And it's actually unhinged.
Here's the idea: AI agents that can test apps like real humans. But not generic humans, specific personas. A 21-year-old gym bro testing a fitness app. A 45-year-old accountant testing banking software. Each with different behaviors, tech comfort levels, pain points.
We built a full stack for it.
Parallel testing across multiple devices simultaneously.
Result: structured bug reports with severity ratings, user feedback loops, and actionable insights. Not just "the app broke", but "users like this demographic get stuck here for this reason."
@AndresDevX @CohenAiden99429
Andres Lockin retweeted
Everyone at YC hackathon built with @browser_use computer use agents, etc
we built 𝗽𝗵𝗼𝗻𝗲 𝘂𝘀𝗲
guess which one people actually use 8 hours a day?
we fork Browser Use and add support to just using:
'from browser_use import phone_use'
u can install these as an skill on openclaw, claude code etc
thanks to: @AmaruEscalante @CohenAiden99429
Everyone at YC hackathon built with @browser_use computer use agents, etc
we built 𝗽𝗵𝗼𝗻𝗲 𝘂𝘀𝗲
guess which one people actually use 8 hours a day?
we fork Browser Use and add support to just using:
'from browser_use import phone_use'
u can install these as an skill on openclaw, claude code etc
thanks to: @AmaruEscalante @CohenAiden99429
dev twitter used to be:
- react devs
- vue devs
- backend bros
- frontend gang
now it's:
- Setup openclaw
- Orchestrate Swarms of Coding Agents
- Rate Limits on Claude Code
- still manually debugging (lol)
the social graph reorganized around methodology, not stack
2
5
269
Andres Lockin retweeted
the stuff nobody tells u about shipping:
viral launches don't matter
that one commit at 2am when u hate everything does
motivation runs out in week 2
discipline is what separates the 6 month projects from the 6 year ones
experienced devs know this but won't say it because it's not inspiring
new builders think consistency is cringe until they realize it's literally the only thing that works
the boring daily commits > the perfect launch tweet
the stuff nobody tells u about shipping:
viral launches don't matter
that one commit at 2am when u hate everything does
motivation runs out in week 2
discipline is what separates the 6 month projects from the 6 year ones
experienced devs know this but won't say it because it's not inspiring
new builders think consistency is cringe until they realize it's literally the only thing that works
the boring daily commits > the perfect launch tweet
Orchestration layers are cool.
But they're 20x better when you coordinate swarms of coding agents.
Grateful for everyone building, sharing, and scaling these agents 🤝
If you're into Coding AI Agents, say hi 👋
3
90










