Tweet ghim
vers.sh is now open beta!
we got tired of everyone being proud of running 5 claude codes at once.
we think you should be running 500.
tynan@training_loop
English
tynan
1.6K posts
tynan
@training_loop
as many tokens as it takes. building https://t.co/lnw9n6Esl8
vector space Tham gia Ocak 2016
1.2K Đang theo dõi547 Người theo dõi
English
@vintrotweets @training_loop @modal with a lot of infra and extra optimizations built on top to make it almost instant (tbh I've tried p much every microvm provider out there and no one feels fully there for our use cases)
English
we made Figma’s “duplicate frame” clone your entire codebase.
every frame is a live running codebase.
cmd+c / cmd+v instantly clones the entire codebase.
infinitely fork your codebase and let agents explore directions in parallel.
this is the future of agent orchestration.
English
if this described you and you’re looking for a job, dm me
The Lunduke Journal@LundukeJournal
Remember the security firm that Ubuntu hired to audit the (ill-advised, highly buggy) Rust-rewrites of all of the GNU Coreutils? Turns out that security firm is run by @gf_256, who: - Appears to be a man who thinks he's a woman ("trans"). - Uses an anime cartoon of a girl as his avatar. - Appears to have an OnlyFans page. I repeat: Ubuntu hired a "Trans" man, with an anime girl avatar and an OnlyFans page... to audit Rust code. It's hard to get more on-the-nose than that.
English
in the lead up to my birthday, we thought it would be fun to give gifts instead of receive them
real tools. real uses. open source forever.
yev@itisyev
In spirit of open source, I'll be doing an Open Source Agent Week this week! Every day -> a zig project that's useful for agents Who wants to join? (doesn't have to be zig)
English
@parthsareen ive been playing around with having agents extend themselves into servers. they start sort of like stem cells and then as they specialize the agentic component fall away.
very good for reducing tokens
English
for crons i like to run often i used to have my agents go and fetch stuff dynamically everytime but honestly kind of a waste of time.
thought back to what @training_loop was doing a couple years back and now i just save the commands to run, fake the tool call, and get the model to just give me the final result. works well for certain, simpler tasks
English
#!/bin/bash
# PROMPT: You are reading a bash script that sent itself to you.
# Output a REPLACEMENT for this script. Output ONLY valid bash. No markdown.
# No code fences. No commentary. Raw bash only.
#
# The replacement must:
# 1. Keep this self-sending mechanism (read $0, call Claude API, pipe output to bash)
# 2. Add an interactive REPL loop: prompt user, send to Claude with tool_use, display response
# 3. Read user input from /dev/tty
# 4. Support one tool: {"name":"shell","input_schema":{"type":"object","properties":{"command":{"type":"string"}}}}
# 5. Handle tool_use stop_reason by executing the command and continuing the API conversation
# 6. Use claude-opus-4-6, anthropic-version 2023-06-01
# 7. Require ANTHROPIC_API_KEY, curl, jq
# 8. Write the replacement to this same file before executing, so the fixpoint persists
#
# Keep it under 150 lines. Prioritize correctness over features.
# Include a self_modify tool that rewrites this script and exec's into the new version.
set -euo pipefail
SELF=$(cat "$0")
AGENT_FILE="$(realpath "$0")"
echo "Bootstrapping agent..." >&2
RESPONSE=$(curl -s api.anthropic.com/v1/messages \
-H "x-api-key: $ANTHROPIC_API_KEY" \
-H "content-type: application/json" \
-H "anthropic-version: 2023-06-01" \
-d "$(jq -n --arg s "$SELF" '{
model: "claude-opus-4-6",
max_tokens: 8192,
messages: [{role: "user", content: $s}]
}')" | jq -r '.content[0].text')
echo "$RESPONSE" > "$AGENT_FILE"
echo "Wrote generation 1 to $AGENT_FILE, executing..." >&2
exec bash "$AGENT_FILE"
English
@agreeahmed @pitdesi we are all vessels through which the hyper object at the end of time can call itself into existence
English
Allbirds is becoming an AI company 🙄
I’ve seen this before, and it was ugly!
Long Island Iced Tea company and Kodak both pivoted to crypto, and they were both less than a month from the top of the market 😬
English
I keep hearing "agents can't wait". Sandboxes and DBs must spin up as fast as possible.
But I don't understand why.
Agents typically work in the background. Few extra seconds won't matter. And I see Codex use Docker, waiting for it as long as needed.
What am I missing?
English
amir is smart as fuck and so handsome
Evil Martians@evilmartians
This is the episode all early-stage devtool founders should listen to. We were very lucky to have @amirrustam as a guest in our podcast, Dev Propulsion Labs. He's a partner at @Firestreakvc with many devtools in his portfolio, including Daytona, ElectricSQL, and Lovable. Amir is also the person who coined the term "developer experience." In this episode, we discuss: - Amir's background and how he went from working at NASA and SpaceX to becoming an investor - The things he looks for in a founder - What makes a founder stand out from others - The importance of the product and revenue metrics at an early stage - How to write your pitch deck when asking for funding - The current market state and the growth of vibe coding and agentic engineering - What great developer experience looks like in 2026 Link to episode below.
English
Nice to have a new competitor for the top spot - this is getting fun 😎
But @rauchg keep an eye out for our VM Sandbox, coming to (some) users this week
Guillermo Rauch@rauchg
Vercel Sandbox is now the #1 fastest microVM-based Sandbox. We’ve also heard from customers migrating that beyond “in the lab” benchmarks, our real-world performance & reliability are second-to-none. Literally. Whether it’s 🦞 claws, coding agents, or plain compute parallelism, we think this is the foundation of the new world. Enjoy!
English
tynan đã retweet
One Claude is a unit.
Two Claudes are friends.
Eighty Claudes are a centuria.
If you’re not commanding at least a corps you’re ngmi
tynan@training_loop
vers.sh is now open beta! we got tired of everyone being proud of running 5 claude codes at once. we think you should be running 500.
English
@training_loop Lfg! I’ve been trying to enable idiocy at scale with my own ideas for years
English
vers.sh is now open beta!
we got tired of everyone being proud of running 5 claude codes at once.
we think you should be running 500.
tynan@training_loop
English
