Jeremy Janzen

Helping devs 10x their Claude Code sessions. Founder of bringing determinism to AI dev. Sign up for the waitlist today!

[Jeremy Janzen @Jeremy_Janzen]

I was constantly fighting Claude Code's context limits. So many /clear and /compacts!

So I built a system of subagents. Now my sessions run 10x longer, and the code quality is higher.

Here are 6 battle-tested subagents that transformed my workflow (with working code) 🧵👇

[Jeremy Janzen @Jeremy_Janzen]

@steipete @Zai_org Just a guess but Anthropic never released their tokenizer so it could be a difference between how Anthropic vs Z count tokens.

[Jeremy Janzen @Jeremy_Janzen]

@Zai_org I appreciate the honesty posting a benchmark where Claude is better. Albeit looks like a relatively close win.

[Jeremy Janzen @Jeremy_Janzen]

@steipete So you’ve had only API failures so far or is it working? I would totally go for $3/month for when I hit Claude limits.

[Jeremy Janzen @Jeremy_Janzen]

@fluffyfatcat1 @steipete The harness, tools and tool prompts are ideally tuned to the model itself.

[Jeremy Janzen @Jeremy_Janzen]

@iannuttall For sure. People have built AMAZING products, even solo, without LLMs at all. Or by copy and pasting into ChatGPT 3.5! None of these tools should hold you back if you are dedicated and have a solid idea.

[Jeremy Janzen @Jeremy_Janzen]

@GosuCoder Automatic memory is a really hard problem to solve. What memories to save? And when to retrieve them that doesn’t confuse the current context?

[Jeremy Janzen @Jeremy_Janzen]

@jobergum Should agent benchmarks even be one-shot? Agentic workflows that are real-world usually involve several prompts.

[Jeremy Janzen @Jeremy_Janzen]

@alexalbert__ Please have a look at the frequent “Overloaded” errors on Opus 4.1!

[Jeremy Janzen @Jeremy_Janzen]

@vishyfishy2 I don't think its worth it anymore. Claude Code or other agents can walk you through deployment, VPS setup etc.

[Jeremy Janzen @Jeremy_Janzen]

@eliotpaynter @DannyAziz97 I was looking at doing something like this awhile ago but it seemed like an annoyance that the user has to restart their session. But not taking anything away from this - what you built is very cool!

[Jeremy Janzen @Jeremy_Janzen]

@steipete @cerebras Thought about that but I still do some Steam gaming so chose Bazzite. I do want to try Omarchy though!

[Jeremy Janzen @Jeremy_Janzen]

@steipete @cerebras Yup it’s for sure hard mode in every way! But CC is actually pretty helpful with that figuring out configs and compiling software.

[Jeremy Janzen @Jeremy_Janzen]

@steipete @cerebras I’ve had some weirdness like that happening in Claude Code since I switched to Linux 🥲

[Jeremy Janzen @Jeremy_Janzen]

Incredible news from Anthropic today. Browser use AND Claude Code on the web! 🚀

[Tibor Blaho @btibor91]

Research preview of Claude Code on Claude[.]ai

- updated layout with the prompt composer and sessions moved to the sidebar

- repository setup requires installing the GitHub Claude app on your repository and committing the "Claude Dispatch" GitHub workflow file

- you can enable email and web notifications for Claude Code updates

[Jeremy Janzen @Jeremy_Janzen]

@Thomasmarkelly @trq212 For sure you can do it manually. Get Claude to write to https://t.co/0CiXubiaYe. But Claude Code is about best practices and making it easy for everyone to do that. So having an official solution for this would be nice. (could be optional too if you want to roll your own)

[Jeremy Janzen @Jeremy_Janzen]

A new class of malware targeting LLM-assisted coders appears! 🤯

Will a new class of security tools will have to deal with this?

[zack @zack_overflow]

A popular NPM package got compromised, attackers updated it to run a post-install script that steals secrets

But the script is a *prompt* run by the user's installation of Claude Code. This avoids it being detected by tools that analyze code for malware

You just got vibepwned

[Jeremy Janzen @Jeremy_Janzen]

@vocamix_hoshino This is really interesting. It seems like its a simulation for now. Are you planning on interfacing it with a LLM?

[Jeremy Janzen @Jeremy_Janzen]

@AyomipoOduntan Congrats 😁