iiinova — where your digital self keeps evolving

iamwil @interjectedfuture.com · 29/09/2026

But at the end of the day, I don't recall any of the others besides the original. Maybe just Pi. I hope to clear out more tabs again soon.

000

iamwil @interjectedfuture.com · 29/09/2026

Jeeves – Reasoning improves Jev-like decision models. Haven't looked at it in full, but looks like the fast follows for Jev is in full swing, just like all the fast follows for OpenClaw. github.com/PostHog/jeeves

github.com

GitHub - PostHog/jeeves: Jeeves – Reasoning improves Jev-like decision models

Jeeves – Reasoning improves Jev-like decision models - PostHog/jeeves

100

iamwil @interjectedfuture.com · 29/09/2026

The interesting idea here is Jev + Bayes nets to make more complicated decisions. Is it necessary--isn't it mostly all subsumed into the system 1 model? Or maybe a Bayes Net makes the decision making more legible. x.com/fdellaert/s...

x.com

Frank Dellaert (@fdellaert) on X

Jev + graphical models: a paradigm shift?

100

iamwil @interjectedfuture.com · 29/09/2026

So I agentically coded my own design verifier that converts perceptual requirements in the design brief to measurable rules, and use the measurement tools in a headless browser to verify the frontend outputs. It works after much iteration. x.com/poteto/stat...

x.com

lauren (@poteto) on X

The Complete Guide to pstack Pt. 2

100

iamwil @interjectedfuture.com · 29/09/2026

Part 1 and 2 of @poteto's pStack. I haven't finished reading it all, cuz I stopped at the idea of writing your own verifiers. At the moment, I was having trouble getting the coding loop to *do-the-right-thing* for the front end. x.com/poteto/stat...

x.com

lauren (@poteto) on X

The Complete Guide to pstack Pt. 1

100

iamwil @interjectedfuture.com · 29/09/2026

But as of today, it's too conceptually heavyweight, and worldview-changing for most devs to adopt it. davisjam.github.io/model-based...

100

iamwil @interjectedfuture.com · 29/09/2026

MAGE: Model-Based Agentic Engineering. Basically, use models (not AI models) to represent specs for code, rather than code--structurally, behaviorally, by ownerships, etc. I think agentic-coding will eventually move to something like this.

100

iamwil @interjectedfuture.com · 29/09/2026

Jev as fuzzy linter. This is a nice side project for me to do later. I currently use linters to make sure the formal models aren't crazy, but I always feel uneasy about it. This would at least have some evals against it. x.com/michaelthie...

x.com

Michael Thiessen (@MichaelThiessen) on X

I’ve been experimenting a bit with Jev as a fuzzy linter that runs after edits in your agent harness: looks very promising so far in my evals. First I converted all of our coding guidelines to ti…

110

iamwil @interjectedfuture.com · 29/09/2026

Hadn't heard of BMAD until someone mentioned it in convo. There's lots of stacks out there floating around: gstack, pstack, superpowers, spec kit, compound engineering, etc. BMAD doesn't seem to be popular, but stars aren't the whole story. If you like it, tell me why.

110

iamwil @interjectedfuture.com · 29/09/2026

Clearing out tabs (1): Seems like a type of code search via jev. Would be interesting to use this for the code investigation part of my factory workflow. x.com/dzhng/statu...

x.com

David (@dzhng) on X

Introducing jevgrep - a research agent CLI powered by jev from @typesafeai that reduces your coding agent cost by 40% (verified on SWE-bench) Make sure to use the built in skill so your coding ag…

110

iamwil @interjectedfuture.com · 29/09/2026

If you've ever shopped at IKEA, you've had slop in your life. Did it really feel worse than artisanal furniture? Chances are, it solved your problem, and you moved on with your life. So it will be with most consumer software. www.youtube.com/watch?v=0h8...

youtube.com

How IKEA Mastered Furniture

Try out Odoo and get your first app for free at https://www.odoo.co...

061

iamwil @interjectedfuture.com · 28/09/2026

Actually, I find the setup cumbersome. I don't want to have to route Grok on the car through some Cursor running Grok app that then hands off to Opus 5.5 on a mac mini. An intranet VPN of agents and models by default would be great. x.com/mikepat711/...

x.com

Mike P (@mikepat711) on X

How is this real life? Seriously wtf man.

000

iamwil @interjectedfuture.com · 28/09/2026

This would likely manifest as CA prop tactic to change public sentiment for a Tiktok generation. This is fun nonetheless. Opus 5.5 needs to stick to less fonts though.

000

iamwil @interjectedfuture.com · 28/09/2026

Interesting that this an argument in the form of a motion meme. I'd always thought community goods, like libraries, need more marketing. Unfortunately, they won't hop on this early. x.com/_brightmirr...

x.com

Bright Mirror (@_brightmirror) on X

Made with Claude Opus 5.5. The fearmongering about AI always makes us forget that NOTHING WENT FOOM, as was always predicted. Send this to your doomer friend who has a very high P(Doom). Acc…

100

iamwil @interjectedfuture.com · 26/09/2026

Good point on the differences. Try as I might, in my mind it still seems possible. I’d probably need to be show specific instances of the breakage.

000

iamwil @interjectedfuture.com · 25/09/2026

Legible Embeddings, I call it. x.com/kieranklaas...

060

iamwil @interjectedfuture.com · 24/09/2026

And yet, this is the same as it ever was. We'd always been working towards programming more like constructing arches, rather than stacking piles of rocks.

000

iamwil @interjectedfuture.com · 24/09/2026

Compilers, debuggers, unit tests, type checkers, formal methods are all cultural practices or technology that do that. But with agents able to grind, the tolerable amount of ceremony rises, which means the types of deterministic verifiers also increases.

210

iamwil @interjectedfuture.com · 24/09/2026

I think an interesting way forward for programming languages, frameworks, and tools in context of agents: to find the right building blocks that constrain our outputs to correctness with a tolerable amount of ceremony.

100

iamwil @interjectedfuture.com · 24/09/2026

That said, there's an incentive problem. Successful companies attract operators, not innovators. Once innovative growth runs out, operators have no idea how to grow further, other than to cut costs (efficiency growth)--and are rewarded by the markets for it.

000

iamwil @interjectedfuture.com · 24/09/2026

Also, such a small portion of our population build things now, they're not able to recognize quality during the buying process--only recognizing it after things break. Building things makes you better consumers.

100

iamwil @interjectedfuture.com · 24/09/2026

Sure, I like quality products too. But what's the revealed preference of the consumer? There is a Pareto front here. You can get quality without cheapness or vice versa. Which one does the consumer pick? substack.com/@matundanis...

100

iamwil @interjectedfuture.com · 24/09/2026

It’s too bad. I like this style, which itself was a derivative of the anthropic short film. Do something different if only derivative! There’s a whole creative world to explore out there!

000

iamwil @interjectedfuture.com · 24/09/2026

I think it’s great people are excited! But I’m also sad that on-trend posts garnering clicks stripe mine design styles until people tire of it. Recall converting your photos to Ghibli films? x.com/aollivier82...

x.com

Alix Ollivier (@aollivier82) on X

I asked Claude Opus 5.5 what new capabilities it had. It told me it was really good at pixel-perfect visual understanding. Then Claude made this story about a raindrop.

100

iamwil @interjectedfuture.com · 23/09/2026

I don't think it's unpopular at all. SQL was designed as an english-like query language so non-technical people can fetch their own data. Today, while analysts might learn it, the biggest users of SQL are engineers. I can think of other examples. Guess what's playing out again? lnkd.in/p/ga3wC4XP

000

iamwil @interjectedfuture.com · 23/09/2026

My insistence on readable URLs had ramifications on the rest of the architecture, and it took a week to unwind and make sure the migration didn't break everything. Keying by titles is a bad rookie mistake. Doing UUID wasn't that bad, and we can just append the title after the key.

110

iamwil @interjectedfuture.com · 23/09/2026

Usually, I find tables and graphs more informative. But the data was mapped well to the visual this time around.

000

iamwil @interjectedfuture.com · 23/09/2026

If token use on OpenRouter was a little town with AI labs as the shops. This is the first time I felt like embodiment gave me more information than a table--by viscerally seeing how crowded certain shops are. tt.lab.sael.net/

100

iamwil @interjectedfuture.com · 23/09/2026

Someone is probably studying how this kind of ad in an AI session from @garrytan delights people, instead of annoying people--paving the way to the interaction for ad-supported chatbot. x.com/dharmeshba/...

010

iamwil @interjectedfuture.com · 22/09/2026

I'm for being able to live on Mars, but I also think Congress, NASA, and scientists need to get on the ball and get the archaeology going, so we can determine if there was life on Mars before all this stuff erodes the signal. x.com/erika_alden...

000

iamwil @interjectedfuture.com · 21/09/2026

This possibility, but for UI. We can probably have design systems reactive to a user's goals, so it know what to retain as window gets smaller. We should be able to code like Tiny Glade. x.com/ryan__steph...

x.com

Ryan Stephen (@Ryan__Stephen) on X

breakpoints are dead

220

iamwil @interjectedfuture.com · 20/09/2026

Nice rethinking of a tried and true interaction. x.com/marcus_lowe...

000

iamwil @interjectedfuture.com · 19/09/2026

Pertaining Jev: "Typed function call, can't emit a type error" is a repositioning of "softmax over K classes." What's new: - task at inference time instead of training time - calibration of probabilities is trained for. So 0.75 prob is literal, not just plausible text completion

000

iamwil @interjectedfuture.com · 19/09/2026

I think humans should iterate with the agents on the spec to close the specification gap. At least for now. But my thought behind the tweet are like a design verifier that can map specs to browser measurements. Too onerous for a human to use, but agents don't mind. There are others to be found!

010

iamwil @interjectedfuture.com · 19/09/2026

Code review doesn't matter if code is correct by construction. There will be an explosion of verifiers and guardrails people build beyond the basic unit tests and typecheckers. These will be ones where it would have been too onerous for humans to use, but just right for agents.

110

iamwil @interjectedfuture.com · 18/09/2026

Wait, I don’t get what Jev’s role is here. It seems like this is a datalog/prolog for reasoning, and Haiku converts question to query for the reasoning system…what does Jev have to decide?

170

Reposted by iamwil

Ink & Switch @inkandswitch.com · 17/09/2026

Here it is—the final talk from our @localfirstconf.com Lab Day! After 10 years of Ink & Switch, Peter van Hardenberg @pvh.ca takes the long view. Why do we do research? Who does it serve? And what now, as the world gets weirder? How to Research Tools for Thought

youtube.com

How to Research Tools for Thought

YouTube video by Ink & Switch

14912

iamwil @interjectedfuture.com · 17/09/2026

Interesting. My version for engineering is to resolve ambiguity, not what you want to build in the prompt. x.com/PJaccetturo...

000

iamwil @interjectedfuture.com · 17/09/2026

More people should be doing what Jude is doing--pushing the envelop of what one person can do with AI. Here's an open, cited map of cancer: treatments, targets, trials, and what is coming onco.cc/

onco.cc

OnCo: the open, cited map of cancer: treatments, targets, trials, and what is coming

The open, cited map of oncology: every cancer, treatment, target, trial, company, institution and idea on one page each, in plain English first, with…

100

iamwil @interjectedfuture.com · 17/09/2026

$30k < $ mils for a special effects film. Interestingly, it's a reinvention of the entire workflow. Everyone asking "where's the AI speed up?" forget that the "electrification speedup" in 1900s didn't come until after entire workflows changed from steam. x.com/PJaccetturo...

000

iamwil @interjectedfuture.com · 17/09/2026

Making topic-specific web pages that's catnip to your potential users is probably a vector marketers and "growth hackers" haven't quite picked up on yet. www.1kpapers.com/

1kpapers.com

The Year in AI Papers: Essential Research from 2025–2026

Read clear summaries of the most important AI papers from 2025–2026, organized by topic, research lab, citations, code, and publication date.

000

iamwil @interjectedfuture.com · 16/09/2026

To me, this is more an illustration of how non-technical executives ignore face-slapping warnings from engineers and threaten truth-telling, than the immorality of AI. Though sometimes, engineers need a push to launch, it should never be this egregious. x.com/Gossip_Gobl...

x.com

Gossip Goblin (@Gossip_Goblin) on X

Scalp Misalignment

000

iamwil @interjectedfuture.com · 16/09/2026

Even I can't believe you can get this kind of result from Opus 5. Apparently, it composed the music and the animation all in js. Like what was even the prompt? Even if all AI advancement stopped right now, we'd be exploring all the capabilities for years to come. x.com/kevin_t_ngo...

x.com

Kevin Ngo (@kevin_t_ngo) on X

Claude Opus 5 drew every frame of this animation using JavaScript. The life of a fruit fly.

070