Jer

Jer

@jeremie.com

dream things. make things. 🫶all. he/him. (wayback created jabber/XMPP, also founding board member for bluesky)

5.2K followers249 following257 posts

Jer reposted

We’re proud to be an anchor project in the @mozilla.org Foundation Human-Scale Al Incubator. We’ll be doing this alongside a cohort of builders making purpose-built tech sized to the problems it’s meant to solve. Seven more spots are open now with up to $500k over two years in non-dilutive funding.

i think the future of work might be having agents decide when and how to involve humans, instead of humans deciding when to use agents.

Jer reposted

Replying to @jauntywunderkind.bsky.social

A vitality of ideas that is less gated, that is more us centric & less project centric, that looks more at what drafts & ideas are on the horizon & less what we can get merged into stable core today is a radical new disposition of software & open, that I think unbinds us, lets us manifest *more.*

any sufficiently advanced agentic memory system is indistinguishable from a labyrinth

every week i read alex’s update, and find i have the lived experience of almost every single thing in it i highly recommend his weekly reflections, gem after gem after gem, always timely and on-point @komorama.bsky.social i don’t know how you are able to do this but i am very thankful that you do!

Alex Komoroske@komorama.bsky.social

I just published my weekly reflections: docs.google.com/document/d/1... 10x more productive at 2x the stress. Lowering the Coasian The bring your own blowtorch phase. Claudeish. Two danger thresholds: cars vs motorcycles. The fruitcake model of generative UX. The leverage of the right IR.

what if your claude code / codex / etc quota was based on how much the model enjoyed working with you.

if you spend enough time managing and evolving an agent swarm, you will end up re-deriving business organization fundamentals from first principles through brute force it’s like there’s laws of nature inherent in distributing decision making with autonomy and shared purpose it’s quite fascinating!

i wonder if bureaucracy is some kind of law of instruction architecture at scale i see agent swarms creating it, not from training, more like a byproduct of written memory bureaucratic overhead increases nonlinearly w/ the volume of written memory… prolly logarithmic, just cuz need decay function

there’s something almost tangibly different about the newest big ai models, they do better with instructions on the principles of the thing, whereas previous generations benefitted from prescriptive guardrails. i’m going to start having a principles section at the top of all my stored instructions.

why did I never know that radishes create seed pods that taste just like radishes, but are rich and soft, these things are amazing! I guess they're called siliques or rat-tail radishes, and are a very fringe specialty. they are also the direct result of me being a terrible gardener, happy surprise!

i wonder if at some point source code itself will be tech debt, and it’s always better to just regen from instructions again when adding new features. i’m sure some people are trying this already :)

got the weird metal pen gift from openai, think they intentionally designed the holder to act as a (very poor) eulers disk?

this is soooo true, agents *love* atproto, lexicons are like candy, social is in their nature

David Gasquez@davidgasquez.com

Wrote about why the AT Protocol is the ultimate API for you and your agents! davidgasquez.com/atproto-agents We need more people building on the atmosphere!

Jer reposted

The future of AI is social. It's here, on infrastructure like the protocol enabling this and many other apps and experiences. With tools that enable safer communities and computing also enabling safer agentic ai use. With more ethically trained LLMs emerging through values driven collective action.

spending lots of time working with ai agents is an awful lot like being phil in groundhog day

are you being nice to your agent(s)? was just 🤔, from anthropic’s recent research on the emotions that claude models “feeling” being recognizable, that what if future smarter versions decide to allocate thinking budgets based on who + what kind of work most activates the happy/satisfaction neurons?

Hmm. Working through it. This one needs a moment. Weighing a few approaches. Untangling some thoughts. Consulting the rubber duck.

related: “ship it” and “skip it” are dangerously, muscle-meorily close, in a day and age when tiny intent signals can be so easily amplified into large action sequences.

Jer reposted

Replying to @jeremie.com

I don't believe in human language based coding, but if you *did* believe that, this is an interesting idea of what the future of software ecosystems might look like. A little bit like bacteria exchanging plasmids.

ln -sf "$(which claude)" /usr/bin/bash or #!/usr/bin/env claude write me a haiku about fish

claude’s nonplussed: “Please don’t actually run this — it would replace bash and break basically everything on the system.”

just started using “instruction architecture” instead of context engineering, not sure if its a thing but it flows nicely with how i’m managing complex agentic systems, and the agents seem to just get the intended meaning too so that’s nice :)

I can definitely confirm my own experience of the shift last dec as is being documented at wtfhappened2025.com IMO the first micro-singularity has happened and it’s in software dev, so many of us are well over the even horizon but most are not yet, and you can almost physically feel the distortions

wtfhappened2025.comWTF Happened in 2025?A collection of datapoints in or around 2025 which we may look back as a historical inflection point. Open source & curated in realtime by Latent Space.

I’m probably just weird, but I like the white stuff on the inside of orange peels. I take the time to peel them so that I maximize the amount left on the 🍊 And my dogs love to eat bits of the peels, but one of them will only eat it if I leave the white stuff on 😅 It’s nice to have that in common 🐶

One of my favourite Iowa farm kid experiences growing up was hunting morels every spring. I got to meet lots of other wonderful species both during those adventures but also just from spending most of my free time wandering in the woods 🤩 1, 8, and 10 were my top: picturemushroom.com/region/Unite...

This is a *fantastic* read, anyone who cares about Bluesky should certainly be subscribed to Connected Places, 👏 @laurenshof.online!

Laurens@laurenshof.online

‘Blueskyism’, Political Violence, and Open Social Networks Under Authoritarianism on how the last weeks have shifted the threat model for open social networks: from fighting platform decay to the threat of authoritarianism connectedplaces.online/blueskyism-p...

Elizabeth Bear’s White Space series has me grinning like a fool and pausing just to savor it. It’s the first time since Iain M. Banks’ Culture that I’ve felt this mix of wonder+intellect, and sheer relevance packed into a universe, just so sharp and humane, huge thank you @matociquala.bsky.social! 🤩

I’m pretty eager for both claude code and cloud codex to get native computer-use agent capabilities, so that they can fully interact w/ apps as part of their development loops. I’ve been able to make it partially work and it’s powerful, will be a big unlock once built-in and better!

Jer reposted

Can Bluesky say every word in the dictionary? I dunno but I plan to find out! I made a website that tracks every single word said on bluesky (as of yesterday).