پن کیا گیا ٹویٹ
Yesterday I demo-ed Secureflow at @TheBarnWaterloo . To my surprise, it was well received and I received a lot of positive feedback!
Check out Secureflow 👇
codepathfinder.dev/blog/introduci…
Waterloo, Ontario 🇨🇦 English
Shivasurya
4.2K posts
Shivasurya
@sshivasurya
senior software engineer | security + AI | @UWaterloo @Dropbox @Zoho alum | building https://t.co/bMnGeuZ1tX | 🍁 🇨🇦
Waterloo, Ontario شامل ہوئے Haziran 2013
425 فالونگ696 فالوورز
me in the morning: "NVIDIA DGX Spark is literally the only thing standing between me and greatness. $8,000? Pocket change."
me in the afternoon: "Ah yes, renting GPUs. A wise, rational decision. Very mature of me."
me in the evening: "Stick to Claude Code Max Plan. YOLO."
me in the night: "keep reading baseten Inference engineering book..."
next day, 9am: "So anyway, about this DGX Spark…" 🔁
English
Good signs of Code-Pathfinder MCP adoption in last 30 days! 🚀
It's now published MCP registry: registry.modelcontextprotocol.io/?q=dev.codepat…
More updates dropping soon 😉!
Shivasurya@sshivasurya
LSPs are built for humans typing in IDEs. Pathfinder MCP is built for agents querying codebases. Demoed it tonight at @TheBarnWaterloo despite the storm. Check it out here: codepathfinder.dev/mcp Thanks @shriji for the pic 📸
English
@allgarbled And that gradient colours, and every colorful div too 😂
English
I don’t know what you call them, but these little side tabs are like the emdash of vibe coded UIs
Juri Strumpflohner@juristr
What's your AI adoption level? (according to Steve Yegge)
English
Opus 4.6 1M on Claude code was exciting at first but beyond 300k every request lags and slows down.
English
Anthropic just pulled the oldest trick in SaaS pricing.
I pay $200/mo for Claude Max. My limits have been noticeably worse this past week.
Now they announce 2x off-peak usage for two weeks. Sounds generous.
But here’s what actually happens: limits quietly drop, a temporary 2x makes the reduced limit feel normal, the promo ends, and you’re left at a baseline lower than where you started. You just didn’t notice the downgrade because the 2x absorbed the transition.
These AI plans are massively subsidized. The raw compute behind a heavy user costs multiples of the subscription price. Every move like this is the subsidy quietly correcting.
Very sneaky, Anthropic.
Claude@claudeai
A small thank you to everyone using Claude: We’re doubling usage outside our peak hours for the next two weeks.
English
@ctbbpodcast Or you can use codepathfinder.dev as a tool to drive with LLM to trace the code path 😉 more deterministically.
Having tried with Gemini beyond context window size the results get poorer and misses code path.
English
If you're running AI agents for hacking/research (you should), one of the coolest tips we got from dawgyg is to make them log what failed too. Brain dump it every few minutes to a file, everything it tried, everything that didn't work.
Without it, agents loop back into the same dead ends and you won't notice until you've wasted a ton of time. If the agent cracks something and you only see the result, you've got nothing for next time.
Also: never let them delete files.
For Chrome specifically, dawgyg only uses Gemini because it's the only model with enough context window to trace a code path across hundreds of files without losing state. RCA that used to take 1-2 hours manually now takes about 2 minutes with AI help.
English
@CtrlAltDwayne This falls under: anything that is verifiable is a good signal for AI training/RL loop
youtu.be/b6Doq2fz81U?si…
YouTube
English
The best argument for Rust in 2026 is not memory safety or performance. It is that AI writes better Rust than it writes C++. The compiler feedback loop is so tight that models self-correct in real time. Every error message is a free training signal. Rust was accidentally designed for AI-assisted development 10 years before anyone knew that mattered.
English
@49agents Exactly! Writing techspec and being more specific is way to go!
English
@sshivasurya claude code overnight is the new alpha. specs before sleep, wake up to implemented and tested code. the workflow shift is real - writing the spec well matters more than writing the code
English
Spent the evening reviewing a tech spec, scrutinizing every detail, and splitting it into stacked PRs.
Woke up to all of them implemented, tested. 🤯
Claude Code overnight is the new **alpha** ?
English
Shivasurya ری ٹویٹ کیا
@ycocerious Curating all edgecases, working on another techspec or reading book!
English
Spent an hour with pencil.dev and I'm sold. If you're an engineer who has strong design opinions but zero design skills (ie. me), this is your tool.
Free, runs on all platforms, uses your own claude sub with no setup, and the built-in prompts actually work great.
English
@claudeai If it costs $25 per PR review, RIP code review startups 😆.
English
Shivasurya ری ٹویٹ کیا
I packaged up the "autoresearch" project into a new self-contained minimal repo if people would like to play over the weekend. It's basically nanochat LLM training core stripped down to a single-GPU, one file version of ~630 lines of code, then:
- the human iterates on the prompt (.md)
- the AI agent iterates on the training code (.py)
The goal is to engineer your agents to make the fastest research progress indefinitely and without any of your own involvement. In the image, every dot is a complete LLM training run that lasts exactly 5 minutes. The agent works in an autonomous loop on a git feature branch and accumulates git commits to the training script as it finds better settings (of lower validation loss by the end) of the neural network architecture, the optimizer, all the hyperparameters, etc. You can imagine comparing the research progress of different prompts, different agents, etc.
github.com/karpathy/autor…
Part code, part sci-fi, and a pinch of psychosis :)
English
By popular request, next month at NYSRG is on program analysis. Maybe you use fuzzers or sanitizers, but how do they work? How about symbolic execution and formal verification?
Some cool and useful tools that help people build the most complex systems :)
notes.ekzhang.com/events/nysrg
English
@jordanreviewsit Same with Apple to use finder search and spotlight search for an app 😉
English
Make the Microsoft CEO search for an email on Outlook
English
English
@SebAaltonen Totally agree on tooling leverage. I wrote the original deep-dive on this (that diagram is from my blog) and the token efficiency gains are just as significant as the speed: ~75% fewer tokens on semantic queries.
Full breakdown with benchmarks: karanbansal.in/blog/claude-co…
English
One of the many examples why LLM tooling matters a lot. There's several 10x+ gains to be had just by improving the LLM tool interfaces and LLM runner history management. We could achieve so much more with the LLMs we currently have if we tooled them better.
Om Patel@om_patel5
claude code has a hidden setting that makes it 600x faster and almost nobody knows about it by default it uses text grep to find functions. it doesn't understand your code at all. that's why it takes 30-60 seconds and sometimes returns the wrong file there's a flag called ENABLE_LSP_TOOL that connects it to language servers. same tech that powers vscode's ctrl+click to jump straight to the definition after enabling it: > "add a stripe webhook to my payments page" - claude finds your existing payment logic in 50ms instead of grepping through hundreds of files > "fix the auth bug on my dashboard" - traces the actual call hierarchy instead of guessing which file handles auth > after every edit it auto-catches type errors immediately instead of you finding them 10 prompts later also saves tokens because claude stops wasting context searching for the wrong files 2 minute setup and it works for 11 languages
English