hrishi

Hi HN! Hrishi from Southbridge here.

Long story short: we needed something to actually do brownfield agentic engineering with: something that made it easier to reason about what an agent will see in context at any point in time, remove and flush context often, clear handoffs, better primitives.

Something we noticed is that LLMs have started co-evolving with their harnesses - anthropic gets more training data from you using claude code, claude code gets new features from claude first, etc etc - so we decided to change the base primitive from the llm call to the agent loop. almost any harness - codex, opencode, etc - is supported, you just need to generate a shim using one of the hanks like we do.

So we ended up building a runtime that executes sequenced AI programs - blocks of prompts and code that get loaded into context, run, monitored and then torn down.

We’ve been using it for a while, open-sourcing it in case its useful. Maybe it’s our visibility set but there’s not much out there focused on reliable executions and doing less so you can think more.

It’s single agent threaded (no parallel agent things), designed for headless (we’re a data company), no MCP or skills things - our general policy has been ‘build what you NEED’ and we just haven’t EVER needed those things.

Have fun!

Read more about the idea behind it here: https://southbridge.ai/hankweave

Love this!

Thanks for all the feedback! I've added a section at the end addressing as many as I can.

I like buttons too. Quite a fair bit. But when you get to too many buttons, they become icons.

Add that to the fact that all user tests we've done have fared better with search and more verbose interfaces, and I wonder if we are both of us far from the target user profile.

Good catch - fixed!

That's a good point at the end, and where we'd like to be. Unfortunately dev time is always a finite resource, especially for an early-stage product. You need to pick one over the other - at least a little.

Some of our inspiration definitely comes from CAD, NX CAD and Fusion. IMO the work they've put into making a mountain of functionality useful and intuitive to a new user has paid off.

Thanks for this! We're definitely working it into our system. It's always good to have multiple paths when possible, so a menu that connects to search and backwards might be quite useful to introduce users to features, but tie in muscle memory once they are used to it.

We're guilty of this, despite caring about the product and the users. It's hard to focus on maintenance and deep iteration when there's more critical (and cool) functionality to chase. Not to mention that iterative UX improvements have the same problem as negative results in academia - it's hard to get people excited about it.

I've used the UK verify service, and it was pretty horrible then. Only country I've seen do it somewhat well is Singapore, with Singpass.

Simple solution: Write, but don't publish. Most of the blocks of feeling like you have nothing to add get better if you write more.

Write for your colleagues, your partner, or just yourself, so you can clarify the streams of thought in your head and bang them against something.

Even if you never publish, you'll benefit. Chances are, you'll find a few you want to publish along the way.

Because of how titles and attention spans work.

I've wondered this myself for a long time, before realising that titles like these get most of the attention, enough to rise to your feed.

If this was indeed 'Consider using /events', chances are you'll never see it.

Depends on what functionality you'd like to add as you go. I recently built my site from scratch (with no prior experience) using NextJS, MDX and mdx-bundler [1]. Having the ability to add custom components, or statically generate a related posts matrix are all super useful.

1: https://github.com/kentcdodds/mdx-bundler

https://olickel.com

I'll add my few experiences. The primitives of UI that we use in English are often untranslatable, or need really verbose translations simply because they've been inducted into English as part of the tech UI lexicon.

I speak four languages fluently, and 'OK', 'Abort', 'Retry' and 'Load' are some of the hardest things to translate.

I do! I hadn't come across it yet, and it's always great to see a creator introduce his work well, great work all around!

Datasette looks really useful, the introduction [1] is well done!

1 - https://www.youtube.com/watch?v=7kDFBnXaw-c

This has exactly been my experience. Automation hasn't helped much - we're still not at the point where (as someone else here pointed out) you have a system that can mimic your mind and add the metadata for you.

Clearing out the lists make a big difference. I'm reminded of a video I watched [1] where deleting photos more often made it easier for the author to remember his life better. I've been trying it more, and he's right - it works!

1 - https://www.youtube.com/watch?v=GLy4VKeYxD4

I'm motivated by the same experience. My memory is fading a little, perhaps due to the sheer volume of things I'm trying to remember, or age.

The part about anecdotes hit home, where I'll remember the description for something but not the name, or the name and how it's connected but not enough to really explain it. "I remember it made sense when I was reading it, you'll need to trust me" is not that great.

Agree with the subcomments, the work involved becomes smaller and smaller as your brain gets better at it - somewhat like journaling. If you're willing to put in the extra work at the beginning, it's also a great memory booster. Your brain learns to understand that you intend to remember something, so the connections feel stronger. I definitely recommend it. Even something like a short journal at the end of the day makes a big difference.

The biggest thing is being okay with letting some things go - which has been hard for me, yet has made a huge difference. If you don't have the time to properly put it down to be remembered, accept that it's likely to be lost and move on. If it's important or interesting enough, your mind will remember without the need to save it.

It's a tiny, tiny slice of things that I can ever pull out of the knapsack. Either I've forgotten enough that I don't have enough metadata (or even that they exist), or I remember enough that I can just find it again.

I do miss Ecco Pro, Notion has been a decent replacement. If you want offline and caching, Dendron might be better.

The apps are improving, but for now the web interface is pretty good for a cross-platform experience.

Definitely trying out logseq, thanks!

Some thoughts: the schema is only partially there, at least for me. I can read and understand the content, but if I'm removed enough from when I first read or stored it, I've lost the context in which I was seeing it and the relationships within.

Having something that models the mind is the absolute goal - that would be amazing. It's definitely possible, and a lot of this labour might equate to needing to have a good sense of direction and map-reading skills before we had GPS.

Great book! Graph databases are useful, but I'm not sure if they're still a great interface for a direct human interface. Problem is listing links and finding connections is tiring for us to do.

That said, with enough automation and context-aware software, graph dbs start looking like a great idea.

SQL vs NoSQL is a perfect example of schema-on-read vs schema-on-write, even though with jsonb in postgres and more templates in Mongo you can use both either way.

Agreed. I've also found myself discussing how landlords - or at least professional ones - seem to be in a different social class than tenants, and the ones that aren't have been the most pleasant in my experience.

I don't entirely mean in terms of wealth. It does something to a person to be a landlord in ways I cannot fully explain. In almost every other situation, to be "rent-seeking" is considered a bad thing. It may or may not apply to real-estate, but the landlords I've met just seem to think differently about people and homes in a way I cannot understand.

"Peace broke out" is one of my favorite openings to the Monty Python Funniest Joke in the world sketch. This is well done!

I have taken git apart and learned it three times now, and it all makes sense to me. The commands however, never clicked. The terminology never felt intuitive, nor predictably applied.

So I can explain how git works in great detail, but ask me how to perform an action I haven't in a month, and it's a lot like figuring our a tar command.

The biggest issue IMO is that most software begins as a startup, and despite becoming big companies later, the startup mindset and habits don't really change.

Before product-market fit, you don't really know what data you need and what data you'll need. You can't go back in time to collect data, you need to do it now if there's a chance it will be useful later. In a pre-seed company (or even at seed), you don't have the resources to audit every package much less get SLAs in place. Most companies I know do the bare minimum for GDPR, and it's not a lack of care for the user, but that often it's not the best place to deploy resources when considering the survival of a company.

Most software starts out being assembled in flight with multiple changes in destination - and those habits stick.

Couldn't finish but I'm definitely planning to. Beautifully written, amazing to track the entire journey (almost) in a single page. Did not know what I was getting into, but the writing and concise explanations behind the thought process kept me hooked.

I hope this gets past New!

Well written. I didn't know this by name, but it's something I've noticed in my life - often the circumstances of an action, or even the circumstances of an individual explain an event far better than the individual themselves.

There was a recent post that went into a modification of Hanlon's Razor to "Never attribute to stupidity what can be explained by opportunity cost". Another example, I think of the same.

Yeah you're right, but it's still the biggest information resource we have. Some sort of effort towards making Wikipedia more machine readable - maybe through parsing or automated synonyms - would really go a long way there.

This is super awesome - it's amazing all the things we can do with our equivalent of the library of Alexandria. Don't see enough projects using Wikipedia as a base layer, hopefully this continues!

Would also love for this to be adapted to fandoms and other types of wikis - might not need much modification.