bjconlan

Ax/agent substrate is focused on enterprise inference. This is squarely targeting the training and reinforcement learning patterns (it's not to say these technologies don't overlap, but perhaps closer to kube with knative & kata container support (and some level of session persistence across all 3). Or nomad... Or any orchestrator that supports multiple underlying isolation techniques.

I can't work out from the paper if the session can cross all 3 types of isolation (between persistence) but I'm assuming it can if it needs to evolve tool calling and security requirements.

In any case it feels like they're just composing and mangling technologies into a platform and not bringing much to the table (from what I've read so far; ie still using well established isolation tech and not something unexpected like a microkernel) but the outcomes are impressive

Yeah, this all reminds me of facebooks "enter your email password so we can invite your friends" play which seemed so wrong at the time... But every non technical person at the time seemed to do it and by doing so grew the network and valuation to a surreal point to where it still sits today. I still cant rationalize it. (Except the wave allowed meta to consume Instagram and WhatsApp)

Wow this really getting traction in the media here. No details unfortunately and given its a federal system it's likely managed by complacency to the lowest bidder I'm not surprised they don't have a clear details yet (altho I'd imagine they'll have event logs when that one guy who knows splunk come in). Most fed level projects are mandated to use a seam and follow persistence guidelines but nobody seems to actually verify/check this so we will see)

You know you're too old when you see the company name and think! Oh I wonder what Martin Odeskey , Jonas Bonér and co are up to. Wait, didn't they become lightbend... Altho this comment takes away from what these guys are doing which legitimately sounds interesting.

I have an old colleague who I've been leaning on for ai things (they're more on the pulse than I) who uses cue (cuelang.org) for distributing and formalizing a set of skills/prompts.

It's the best thing I've come across (that I don't need to mange myself) https://github.com/p3bot/start

The tool itself does more than just manage skills/prompts but I found that part of it particularly good (well new to me; not familiar with cue but the idea seems like a good fit)

I do like this; but the classic 'dll hell' issue looks like it might be ignored/overlooked (2 plugins depending on the same plugin but of different versions).

I do like that they've thought primarily about implementing an entire IoC in a runtime environment (which seems like an obvious necessity if you have agents changing code underneath you.

Anyways this looks promising for the scenarios its designed for but wholy moley (https://deepseek-harness.github.io/deepseek-harness/en/refer...); I guess spring's boot (+collection of starters) might look the same to the uninitiated or even those who remember the good ol' HKEY_CLASSES_ROOT\CLSID

Besides the actual topic. The portfolio is fantastic! I particularly liked the yin and yang of world hunger. https://www.davidrevoy.com/article43/yin-and-yang-of-world-h.... I mean not something I want to look at on a wall for too long but it draws on my enjoyment of dark comic book illustrations while also pointing to something topical

Thanks, I want to add https://www.arraycast.com/2025/01/04/Decker.html which introduced me to it. I wish more people had exposure to q/k/apl and think this is a great way to do it.

I think this is the first time I've seen this hit #1 which is nice

Wow thanks for pointing this out! This is actually what I was hoping would happen when the deepspec stuff dropped! And having zlab create these confirms my bias that I think open models are the way.

Yeah I'm with you, this feels like a case for assertions & not a new core feature. Perhaps I'm missing something because it was promoted by Michael Fogus and has been ratified by those who would know (Alex I'm assuming). To me it doesn't pass the test of necessity as something needed at the core but at least it feels somewhat ideomatic.

This is a pretty harsh critique, while I see the agent related markdown the application seems measured and well applied. Hardly seems like slop; it's might be a one man show+ai agent but the direction and project composition feel like it's something that's logical and manageable.

I personally like the idea even as a p.o.c (inline with bellards work). I'd like to see how nostr & cloudflare workers providing out of band ops on what is effectively a container per tab.

If you're looking to use this with the community edition I would recommend reading through https://dataintellect.com/blog/running-torq-with-kdb-x-commu... as they outline some of the factors that need consideration.

I really do love what kdb is doing but I think maybe they lost too much momentum (and confidence) when back flipping on previous agreements for the community editions. (Which I don't think they will again with the kdb-x releases as they have a good balance this time) It would have been nice to see torqx just work (tm)

Redacting "I noticed that a kernel level impl is only 9000 loc (shallow dive https://lwn.net/Articles/1029851/). As it doesn't fully implement the feature set.

Also I noticed that Plan 9's IL protocol (avoids head of line because the type of messaging it's used for) another point pointed out in the x comments.

I wonder if he's focusing on consumer distributed AI inference (due to quic being browser supported) or maybe just a comment on the protocol in general.

This is great! I feel the same way about the deepseek v4 architecture for commodity hardware.

Also have enjoyed playing with https://huggingface.co/HuggingFaceTB/nanowhale-100m-base (but early days for me understanding this space)

This is the kind of clojure port that I always was looking for. Mostly because I thought go's core library and channels abstractions hits a simpler/nicer base API which would with the core & async apis (not to mention scratches my big beautiful binary itch)

Thanks for your work will definitely check it out again once I get over renewed love for cpp (26)

Edit how did glojure go under my radar also a great project from the looks

Ah this reminds me of my favorite stack which I never got to actually work with; luvit.io (libuv with luajit; both hot items at the time).

I hope this becomes a thing, typing in lua might make it easier to scale for teams (which I think is always an issue with dynamic language)

I really should check up on what Mr Caswell is up to. He was quite inspiring with both his diy home improvements as well as activity in the development space.

While I haven't read all the posts here I was wondering if anyone also noticed a 10% usage before their most recent weeks usage even started? (Specifically over 2026-03-27/28) I was seeing weird service outages over this time too. I suspect they're not being 100% truthful with how they are recording usage (feels like they had an agent run a backfill approximation). So they blur it with weekends rates etc.

Anyways I don't have the knowledge as to how to audit this (claud pro) to confirm what feels like an onboard at any cost business behavior.

Is anyone currently auditing through openrouter/litellm and seeing any poor correlation to the session/weekly limit?

Sorry, I did compose it from my phone and wasn't clear. I realize the article talks to the Maven tooling I was talking to it's data aggregation and modeling suite more generally. (And obviously would have been used as part of the Iranian engagement. The "if at all" statement was meant to be "tongue in cheek" given the current mess being reported)

I'm curious as to how Palantir has been used during the war or Iran (if at all or does it suffer from subjective bias). I know there were larger movements at play on a political level here but I'm becoming concerned about how much one "thought group" (in private corps) is having on the world's largest war machine. might be dulling critical thinking.

Token usage and agent usage optimisation?

It seems like a real problem for me. Probably because I'm not overly inspired to pay for a Claude x5 subscription and really hate the session restrictions (esp when weekly expend at the end of the week can't be utilized due to session restrictions) on a standard pro model. Most of my tasks are basically using superpowers and I find I get about 30-90m of usage per session before I run out of tokens (resets about every 4 hours after which I generally don't get back to until the next day (my weekly usage is about 50% so lots of wastage due to bad scheduling). A tool like this could add better afk like agent interoperability through batching etc as a one tool fits all like scenario.

If this gets its foot in the door/market-share there is plenty of runway here for adding more optimized agent utilization and adding value for users.

I used to work for a human that did this (sits mostly on the classical therapeutics side). He actually started a business where he was reviewing and auditing the submission processes outlining approvals but he had been around the game enough to know where the next submission would put them in the approvals process for a number of agencies.

https://maestrodatabase.com/

Looks like he's still on top of everything given the most recent blog post is from 6/2/2026.

I believe the insights here could be useful given he has sense of when the penultimate submission has occured (but I'm not entirely sure what that is on a % basis nor as a basis for if the stock of the company reacts)

Yeah, but it's specifically testing things that implement against a posix API (because generally that's what "native" apis do (omiting libc and other os specific foundation libraries that are pulled in at runtime or otherwise) I would suspect that if the applications that linked against some wasi like runtime it might be a better metric (native wasi as a lib/vs a was runtime that also links) mind you that still wouldn't help the browser runtime... But would be a better metric for wasm (to native) performance comaparison.

But as already mentioned we have gone through this all before. Maybe we'll see wasm bytecodes pushed through silicon like we did the Jvm... Although perhaps this time it might stick or move up into server hardware (which might have happened, but I only recall embedded devices supporting hardware level Jvm bytecodes).

In short the web browser bit is omitted from the title.

Or toit, which unsurprisingly has Lars Bak involved. A man with history touching all self, Jvm and v8 codebases.

I wouldn't be surprised if toit primaries, Kasper or Florian also have experience in these technological intersections.

Perhaps if a supply chain attack is your largest concern then using some well vetted system like wolfi is more up your alley. (See some of their related repos on GitHub https://github.com/projectbluefin - I've been following the development of it and currently it still under development.)

Again "vetting" is a source of contention here as I'm not sure how the quality of official rpm sources compare to those outlined in an sbom

Yeah, but there is something else here too... I used cachy for a heartbeat and it advertises the same benefits; it just felt slower (notably on boot) Maybe it was just all the graphical load screens.

There's something clear had that made it feel modern, familiar and boring (which might not be for everyone) 90% of my tasks were in vscode devcontainers so kept things simple and out of the system for the most part.

Fingers crossed. I probably just did my last fresh install of this a couple of days ago and my last swupd update now. You will be missed...

I do love the warnings here... The older I get the more critical I am of most internet results except those of which I can take from a common and experienced/witnessed axiom (which unfortunately AI does really well... At least entrusting me to said point). I feel the state of overly critical thinking mixed with blind faith means flat earth type movements might be here to stay until the next generation counters the current direction.

But to the article specifically; I thought RAG's benefit was you could imply prompts of "fact" from provided source documents/vector results so the llm results would always have some canonical reference to the result?

I must say this is amazing. The psychology and manipulation makes me realize how poor I am regarding trust even when the other side is pushing for some unconfirmed equilibrium.

In the game I acknowledge that I was aligned to the "Simpleton" strategy (before it was outlined). Looks like simpleton might actually be applicable in a more general sense too which is a little disheartening.

Yeah I feel yah; But to bolster the post:

I now want to hear more as to why Defold now has a clojure repl! I noticed some musings around some native bindings in gh issues which is "interesting" but I'm not quite getting it. I guess off to the forums I go!

This is where I need to shout out to everyone's favorite developer Justine for keeping Linux cool:

https://justine.lol/pledge/

Which also points to landlock-make[0] or vice-versa (the original project that made me aware of the kernel functionality (although didn't realize it also isolated network which is great).

[0]https://justine.lol/make/