Andric

amateur connoisseur of text editors · wrangler of coding agents · building something new called `shorthand`

[Andric @astralwave]

@joshm It isn’t crazy at all.

Consumers generally don’t use business tools. Why would they?

The only time(s) consumers exhibit B2B user behavior is when they’re a student, hobbyist, or sole proprietor.

[Andric @astralwave]

This Josh Miller tweet suffers from the same category error that made the Arc to Dia pivot such a shitshow.

The Browser Company made an amazing browser for work (i.e., B2B), then decided that consumers didn’t get it (why would they?), and decided to pivot hard into consumer

Eventually ending up selling to an enterprise B2B firm (Atlassian) and still seems to confuse consumers with businesses. Again.

[Andric @astralwave]

Hot take… isn’t it kinda crazy that nobody realizes that consumers generally don’t use B2B tools?

[Josh Miller @joshm]

Hot take… isn’t it kinda crazy that nobody is really using AI Agents? I don’t mean software engineers or AI early adopters. I mean “college friends talking about it in group chat,” the feeling you got when everyone started using Instagram or TikTok.

These frontier AI models are *insane* (as are the harnesses & tool calls & the like). And every large tech co has an AI agents platform, not to mention all the YC startups doing vertical agents. Yet all of your friends and family outside of tech — who spend all day staring at their iPhones and get paid to work in browser tabs — don’t really care or find themselves using any AI agents yet.

Yes ChatGPT, Claude, etc. are extremely popular… but if you look at the engagement data the vast majority of people are still using these aI chat tools like a glorified Google + Grammarly. That’s why the AGI labs are all pushing desktop apps for Codex, Cowork, etc. so hard to non-technical ppl. And yes exceptions for lawyers and customer service but even those have some asterisks and exceptions to rule.

Look I’m not saying the ChatGPT moment for AI Agents is not coming… it most definitely is! Remember we pivoted from Arc to Dia precisely because we believe computing is going to be radically reimagined around these AI primitives. No doubt. But that’s my point: it’s just so surprising it hasn’t happened yet because all of the tech you’d need is there.

Again if you stop for a second and think about it… for all the press and money and hype and models and crazy ARR numbers… this “AI Agent” moment does not *feel* like the other breakthrough tech moments we’ve lived through (e.g. think the shift to Stories via Snapchat & Instagram, or shift to on-demand via Uber/Airbnb/Doordash).

Which is a long way of saying: if you can figure out the answer to “why” most people don’t care about AI agents yet (and have no enduring interest in using them) — especially since the models and harnesses are here and ready — the answer to that question will allow you to capture a lot of marketshare and make a lot of money in 2027.

Theoretically, the tech is ready for AI Agents to totally transform how we work and live our lives… but alas the general public dgaf… that’s the generational puzzle to solve for the next 12 months for anyone not working on the models themselves.

[Andric @astralwave]

@aka_suje Make it so you can save it as a non-temp chat if it turns out the output’s worth keeping!

[Andric @astralwave]

I wonder why the latest frontier LLMs have suddenly developed a penchant for mechanical engineering metaphors: load-bearing, gates, doors, bridges, and seams.

Both the Claude 5 and GPT-5.6 families seem to have independently converged on these.

[Andric @astralwave]

@jamwt One might even call them… “load-bearing seams”

[Andric @astralwave]

@thsottiaux Make pins sync between the desktop app and the mobile app please!

[Andric @astralwave]

@waynesutton @convex Building on Convex! Would love to join!

[Andric @astralwave]

Many details about how AI works today were anticipated by Frederik Pohl in the Gateway series written in the 1970s

I’ve only read the first 3 books but he anticipated:

- RAG / stored conversation transcripts that could be queried later

- prompt injection attacks, specifically that which could be used to exfiltrate conversations from another tenant in a multi-tenant therapy AI

- ginormous amounts of memory needed to power these AIs (he even gave an exact number)

- large hosted models and small local models, and their intelligence when it comes to playing chess against a human (the hosted model is smarter than a human, but not the local model)

- how hosted models are served to many users at once and how inference was shared across many users

[Andric @astralwave]

@gurishsharma Does constant swapping in the cache wear down your computer’s SSD?

[Andric @astralwave]

We’ll know we’ve got AGI when OpenAI sorts out its naming woes

[Tibo @thsottiaux]

Should we rename ChatGPT Work to ChatGPT Vibe?

[Andric @astralwave]

We haven’t seen good harnesses for multimedia yet (for working with images, videos, audio, music, etc).

For the most part, coding harnesses are only general purpose if what you’re working with is largely text-based.

Yes some kinds of videos can be represented in a text file, and you can generate images with Codex, but it’s not quite there yet.

[Andric @astralwave]

Make feedback agent native!

Add a special feedback chat in the ChatGPT UI itself (web, desktop, mobile apps) where you can chat with an agent that will grill you for what you’re dissatisfied about and the resulting report will be sent to your team.

Using some variant of /grill-me from @mattpocockuk

Even better if it’s a built-in skill you can trigger, like /feedback.

[Andric @astralwave]

You can develop taste independently of learning a craft.

Those two are orthogonal to each other.

It’s why “the gap” exists (you can have taste that outstrips your skill).

It’s why jobs like museum curators,musicologists, enologists, music producers, product managers, film critics, film festival programmers, etc are able to exist.

[Andric @astralwave]

It’s not common for tastemakers in many creative fields to get their hands dirty.

Directors and producers don’t often do it (unless they’re an arthouse “auteur” director, but those are almost always small-scale productions), senior product leaders in tech don’t often do it, heck, even many artists don’t do it.

And yet they’re responsible for taste, and they’re the final call on creative decisions.

Being responsible for taste calls doesn’t require you to do the actual technical craft, or even have done it before.

In many cases it even hampers your ability to make these judgement calls, because you’d be so caught up in the weeds you forget the forest for the trees.

There’s a reason that even craftsmen-type creatives, like graphic designers, illustrators, or actors, often speak of “the gap” which is where you often start out in a place where your taste far outstrips your own abilities.

But many creative fields are team-oriented and the tastemaker is not coupled to craft – software is one of these! Those who make taste calls in software work in the product management function (usually, but sometimes they are “member of technical staff” or they are C-level execs).

In such fields “the gap” is the gap between your taste and your ability to hire top talent (not your own skill).

AI closes that gap because tokens are more affordable and less scarce than top talent.

[Andric @astralwave]

@RahbanGhani @emilkowalski I don’t see how AI substantially changes the proportion of people who are lazy, have poor taste, and don’t care for making something good.

If anything, AI makes it easier for them to make better stuff, when the labs train models to output better outputs for one-shot prompts.

[Andric @astralwave]

The same way a filmmaker knows what a great film looks like, or the way a musician knows what a great song sounds like, or the way an artist knows what a great painting looks like.

Explore the classics deeply and develop your taste.

In each creative medium and each way of working there’s always been lazy people who didn’t care for doing that.

“make this dashboard beautiful and make no mistakes” could easily have been a PM working at a tech company who doesn’t know how to collaborate with designers and devs.

Or it could’ve been a shitty Hollywood exec, or a terrible Madison Ave ad man, or a CEO of a video game studio, or Rick Rubin with his poor grasp of compression.

[Andric @astralwave]

That’s nonsense.

Learning was crazy slow when you had to code by hand (around 2000s to early 2010s)

Back then “coding by hand” just meant reading browser specs and then finding out what they could achieve, the slow way, by doing tutorials

I spent countless hours on CSS Tricks, Smashing Magazine, A List Apart, MDN, Codepen, and Computer Arts magazine that I will never get back.

Later in the 2010s, when Figma/Framer/Webflow launched, there were product managers and designers who were able to work well pairing with an engineer.

Specialization between UI design & frontend dev made things go faster!

Which is not so different from working with a coding agent to get those results.

The trick is understanding what the platform allows you to do, and exploring that design space.

Working with AI makes both of these infinitely easier, NOT harder.

[Andric @astralwave]

Please make it easier for chats in ChatGPT to move between projects or change owners.

Right now you can’t move a chat in or out of a “project-only memory” folder anymore.

But projects become “project-only” when shared, and this makes it impossible to move chats afterwards!

You can’t change your mind about where a chat should be once it’s been started, which is extremely limiting.

You might want to start a shared space with someone else and realize you have a personal chat you want to move there. Or you might start a chat in a shared space and realize you’d rather move it out to your personal space.

[Andric @astralwave]

@MarkoSmilja They are both programs that take human input, which then is used to write programs.

They don’t need to do it in the same way for the comparison to make sense!

In practice they’re used in the same way.