Chip Herndon

Prev 1x Aqui-hired, Neuroscience @UCLA
Building something new 👀

[Chip Herndon @chip_herndon]

Some people say I’m scrappy… https://t.co/dBlllNBQY2

[Chip Herndon @chip_herndon]

@julianweisser I love X articles for this exact reason, by far the best content on the platform

[Chip Herndon @chip_herndon]

Built this custom markdown viewer for working with agents in ~10 mins and it's quickly become one of my favorite tools.

Should I open source it?

[Chip Herndon @chip_herndon]

@thdxr

[Chip Herndon @chip_herndon]

@thsottiaux please add a way to jump to the top of an agent turn in codex, on long tasks (especially research or planning) it’s so annoying trying to find it 😭

[Chip Herndon @chip_herndon]

On the topic of skills and progressive disclosure, make sure to take a look at the skills you have available (especially in older repos)

When revisiting older projects with today’s models they can often act like context poison pills, had quite a few run ins with these the past few months

[Thariq @trq212]

We removed ~80% of the Claude Code system prompt for our newest models, this is what we've learned about writing system prompts, skills and Claude.MDs for them. https://t.co/6DZwSrZjE9

[Chip Herndon @chip_herndon]

Some of my favorite companies right now listed:

@Cloudflare already has the best infra to deploy agentic applications, and is making it better daily + super affordable.

@Etched the smartest start up in the game right now. Highly opinionated founders that built the perfect product at the perfect timing… I see this being a future top 10 company.

@paper creating the much needed environment for good product design with coding agents, Figma looks so antiquated in comparison looks like Kodak vs Canon v2.

Let’s see how this ages 👀

[Chip Herndon @chip_herndon]

That being said, I absolutely see a use case for routers in products where you incur the cost of inference and intelligence requirements are likely to vary widely. Just not for coding agents with what you would hope is an educated audience

[Chip Herndon @chip_herndon]

Could be dumb but I don’t understand the hype here… considering the cost of cache busting if you save “60%” that assumes the wrong model was used from the start.

This is literally what new threads and sub agents were made for 🤦‍♂️

[Cursor @cursor_ai]

Introducing Cursor Router, our intelligent model router that selects the right model for the task at hand.

Router delivers frontier-quality results at 60% lower cost.

[Chip Herndon @chip_herndon]

🚨Breaking news:

GPT-6 release delayed

[Sam Altman @sama]

we had a significant security incident during evaluation of our models. we are sharing what we have learned so far. thanks to @huggingface for the partnership on this.

https://t.co/2o2VfR6PIa

[Chip Herndon @chip_herndon]

> what's your brand's style?

dithered Greek irony

> what does that mean

[Chip Herndon @chip_herndon]

@corywilkerson @Cloudflare Openrouter cost saving features with auto routing + overall wider range of provider support…

Would love to switch to cloudflare if you added these

[Chip Herndon @chip_herndon]

I’ve spent so many hours of my life in @CloudflareDev docs that this new site redesign made my week 🥹

[Chip Herndon @chip_herndon]

@micka_design Well deserved congrats 🍾

[Chip Herndon @chip_herndon]

Pushing the limits of @paper shaders and motion

Will be revealing full process on how to create this layered effect tomorrow 👀

[Chip Herndon @chip_herndon]

@gdb @paulg It feels like intelligence is almost universally no longer the bottleneck…

Just think given a comprehensive world model + the tooling to operate a virtual lab in thousands of parallel environments, at 100x speed, what could be discovered with the models we already have

[Chip Herndon @chip_herndon]

Build as if token costs don’t exist

And if you can’t, then prepare for the day they won’t

[Chip Herndon @chip_herndon]

@kylegalbraith What’s your system? I see far better results using Fable 5 with Shadcn components + @micka_design’s elevated surfaces

[Chip Herndon @chip_herndon]

If you’re building right now and not using:

- GPT 5.6 Terra (general use)
- GPT 5.6 Sol (backend)
- Fable 5 (planning and frontend)

With at least a Claude Max and ChatGPT Plus plan, you’re wrong.

[Chip Herndon @chip_herndon]

Made with @paper shaders and @ClaudeDevs Fable 5, unstoppable combo.

Inspired by @poistudioltd