Arun Gopalakrishnan

@dexhorthy @mattpocockuk youtu.be/uLWOLmeHOSE?si…

@dexhorthy @mattpocockuk thoughtworks calls it feed-forward (pro-active rules , guides etc) and feed-back (for linters, static analysis etc) as described here

I think either “feedback loops” or “backpressure” is fine to describe the process of providing automated feedback to llms to help them deliver higher quality outputs. I agree with @mattpocockuk that “feedback loops” tend more generally to suggest the inclusion of human feedback in scope, which was not necessarily how I would think about it, but also believe that might be okay because every prod ai system will probably have human feedback somewhere (amdahls law for ai? You can have ai crank out images all day but eventually someone’s gonna look at them otherwise what’s the point?) Backpressure has the downside that it’s probably should stay in hystrix-y infrastructure land as a “self healing infrastructure”, and the upside that it would encourage you to bring that type of systems resilience thinking into your ai workflows

@dexhorthy @vaibcode loved the manus agent writeup walk through that you both did several months back. would love another for this openai.com/index/harness-…

@trq212 not a white whale - more of an eye sore - markdown table display has been broken since last week

we're doing a lot more of this, hunting down some of the most annoying bugs in Claude Code let me know if you have any white whales

@bcherny @ClaudeDevs @trq212 today things are stable for me and back to what it used to be! thank you!

@ClaudeDevs @trq212 never ran out of my max subscription credits in the last 8 months. this week...it happened in 3 days after 4.7. may be because I am still figuring out how to to use it right. lot of work ahead of me this week 😓.

@dexhorthy slow the F down and read your code

caption this

@trq212 @jianggli does the api pricing include the price to run the sandboxes too?

@jianggli API

Managed Agents is the first 'agent in the cloud' API that has the right mix of simplicity and complexity. Implementation details like how you manage a sandbox are abstracted, but you have a lot of control over the actual execution of the model.

@jdrhyne thank you kind sir

@arungk We have a quick fail for the heuristic driven converter that is hitting the speed gains. We also have an OCR/Vision add on that would drastically improve accuracy and structured output but obviously increase processing time. I'll send you the skill via dm.

Feedback from the early release of our pdf-to-markdown converter is coming in really positive. Love seeing people finding use in it. We’ll be releasing it likely tomorrow. Fastest, accurate and most token efficient parser out there based on our benchmarks. Still time if you want early access. Just reply to this post and I’ll dm you the link.

@trq212 @bcherny thank you for everything that your team does. appreciate you all.

@dexhorthy very qrisp

I am humble enough to admit when I was wrong - 30k views in 5 days, maybe you are too youtube.com/watch?v=YwZR6t…

@jdrhyne @nutrientdocs would love to try ..right now use Marker and MinerU for pdf to markdown. MinerU is really really good.

We've been cooking @nutrientdocs on a PDF, DOCX, and other common document file types converter to Markdown packaged as a simple CLI library. On all our benchmarks we've done so far it's the fastest available by a long shot, most token efficient, and on par with accuracy of the top libraries. - 2x faster than liteparse that was just released. - 5x faster than pymupdf - 4x faster than markit who @badlogicgames just praised for its speed. - 3x faster than Microsoft's markitdown. We'll be releasing the converter as freemium closed sourced here soon. Stay tuned. Let me know if your interested in early access.

@dexhorthy fun. what you all shipping. i recognize at least one other face from the pod

Friends ship

@trq212 @bcherny ⎿ API Error: The model has reached its context window limit. ⎿ Context limit reached · /compact or /clear to continue first time in many many weeks on 2.1.56 please have a look Feedback ID: a71f12fb-111a-49e5-ac65-97cb931d2b37

@bcherny from one CC session, it would be very helpful to be able to message another CC session about suggested changes.

@talraviv can't miss @vaibcode . his videos with @dexhorthy at @boundaryml/videos" target="_blank" rel="nofollow noopener">youtube.com/@boundaryml/vi… and github.com/ai-that-works/… is all you need to build any ai application. every episode and content is gold

I don't aspire to be an AI engineer, but I do want to understand one layer down. Here's who I'm following in 2026: (For me, one layer down means "building applications on top of foundation models" to borrow from Chip Huyen's awesome O'Reilly book, AI Engineering) This is an incomplete list, highly biased to my own fumbling around the internet, and whose emails I'm excited to open. I'd love your recommendations on technical people I should be following! 🏢 Company engineering blogs • Anthropic's blog is full of accessible, widely-cited bangers. Subscribe in the footer here website.claude.com/blog. Even if you don't, engineers will forward these to you. • Whenever I wonder how Claude's consumer features work, I can usually find it thoroughly explained in their developer docs: platform.claude.com/docs/en/agents… • Surge's blog is thought-provoking real talk, backed by their hands-on experience. (I can't figure out how to subscribe, since they don't have an RSS feed. In the meantime, I keep refreshing surgehq.ai/blog) • Also following @humanlayer_dev humanlayer.dev/blog, @promptlayer blog.promptlayer.com, and Amp ampcode.com/chronicle. 🤓 Individual engineers • Doug Turnbull for agentic search softwaredoug.com • Eleanor Berger and Isaac Plath put out incredible hands-on walkthroughs elite-ai-assisted-coding.dev • I like how Ben Guo (Zo Computer) thinks zoputer.substack.com • Armin Ronacher early at Sentry, creator of Flask is refreshingly honest and accessible lucumr.pocoo.org • Mario Zechner exploring agents in public mariozechner.at • Jesse Vincent is open-source-famous and created the Claude Code Superpowers plugin. Anytime i see a new post I know it's going to be accessible and super hands on. blog.fsck.com (Many of these don't have email forms. I use an RSS-to-email service to email me when they post new stuff.) 😵‍💫 How I discover good stuff I stay away from feeds. Luckily, there's people (and AI) who browse social media so I don't have to: • I love the concept of Hacker News digest newsletters. Instead of compulsively refreshing HN, these ad-supported services wait a bit, see what shakes out, and email you a summary: hackernewsletter.com or hndigest.com • @bentossell ends each newsletter with what he's reading, solid way to discover high signal essays bensbites.com 👓 Happy reading! You don't have to be an engineer to follow engineers. When Anthropic/OpenAI/Langchain/etc. announce a new capability (for developers or consumers), I've already seen these folks thinking about it in public. It feels good to anticipate instead of chase.

@chris_j_paxton seems like operator intelligence. what if it is not neatly pre-arranged. isn't life too messy outside of these perfect arrangement.

Human level dexterity is absolutely not gated by hardware

HNY @trq212! Is there an Anthropic supported golang agent in in the works?

@gdb cc @vaibcode :)

rust is a perfect language for agents, given that if it compiles it's ~correct