iiinova — where your digital self keeps evolving

Jake Gold @jacob.gold · 02/10/2026

I've had to re-learn this lesson so many times I'll never forget it: shipping = growth You can get lucky and have external forces help you, but the thing you can actually control is how much you ship.

2340

Jake Gold @jacob.gold · 01/10/2026

Most unbelievable scene in Disclosure Day wasn't the aliens. It was the: 1. Over the shoulder bread holder 2. Stealing a chunk of plain bread 3. Insufficient toasting 4. Dedicated butter dish 5. Butter dish stacked on bread Spielberg is an alien that doesn't understand toast.

4201

Jake Gold @jacob.gold · 29/09/2026

“What if we flip the b around?” Smart using Astra so that it’s resistant to prompt injection.

1110

Jake Gold @jacob.gold · 29/09/2026

Okay Jev had less than two months in the sun. I should’ve predicted weeks.

1220

Jake Gold @jacob.gold · 25/09/2026

Stop to smell the flowers and also stop to observe the queen of the skies.

260

Jake Gold @jacob.gold · 25/09/2026

They looked worse than seasonality alone explains but maybe not. What about the other ones? Found a chart showing a downward trend in stickiness, which is by far the most important metric for social apps:

120

Jake Gold @jacob.gold · 25/09/2026

Yeah, that's a totally fair point. Doom loop: AI reduces Reddit traffic -> Reddit squeezes users with more ads, locks down access -> reduces traffic fruther. Might take years for them to be StackOverflow but faster if AI and Bluesky accelerates the loop. Found this site: oscylo.com

130

Jake Gold @jacob.gold · 25/09/2026

I started Googling in Sep 1998 and stopped in Sep 2026. But I should've stopped Googling in 2025 but somehow it took me a year to spend 30 seconds doing this:

5130

Jake Gold @jacob.gold · 25/09/2026

Reddit is way bigger than Bluesky, but Reddit is in the extraction phase. They're milking it for everything it's worth and that's exactly Bluesky's opportunity. It'll catch up to them:

230

Jake Gold @jacob.gold · 25/09/2026

Yet another gem from @simonwillison.net that matches my own experience over the last ~3 months. There's no way I could've built my latest ~650 KLOC system in Go without designing and understanding it precisely. It'd be a total mess, and I'm using Fable and Astra for the vast majority of my work.

5896

Jake Gold @jacob.gold · 23/09/2026

@toni.bsky.team probably time to declare a "code red" and get Bluesky Communities launched ASAP. I'm as blue-pilled and optimistic as it gets, and I deeply respect *everyone* I worked with on the team, but even I'm getting worried about how many chances remain. Feels too slow to win at this pace 🙁

1512414

Jake Gold @jacob.gold · 18/09/2026

Anthropic has finally acknowledged that OpenAI has matched them. I get why it took this long, but sure was annoying.

1342

Jake Gold @jacob.gold · 18/09/2026

AI is a two horse race right now:

6211

Jake Gold @jacob.gold · 18/09/2026

If the calculator was invented by a modern startup: axiom.clor.app

0133

Jake Gold @jacob.gold · 15/09/2026

Steve Yegge is a smart, experienced, and interesting person. But Gas Town is among the best public examples of AI psychosis. Don't huff your own gas people.

Gas Town was intended to be reusable, but I only ever wound up using it to build itself. Gas Town fell apart at the seams with Opus 4.7. Up through 4.6 it was working brilliantly. With 4.7 we saw the introduction of the "just two more things" tic, which prevented Opus from ever converging on being ready to do real work—it always wanted to fiddle with Gas Town itself. The Opus tic never went away, so Gas Town effectively burned down. It had other problems, too, but 4.7 was the final straw.

https://yegge.ai/essays/the-shape-of-things-to-come/

3222

Jake Gold @jacob.gold · 15/09/2026

Seems fine for you or I to put a made-up number on it. But when you work at large AI companies, it's just a good way to spread panic, because much of public will just read the headline.

000

Jake Gold @jacob.gold · 14/09/2026

The Bay Trail is how I ward off full blown AI psychosis.

1460

Jake Gold @jacob.gold · 13/09/2026

Yes, developers generate more diagrams! I used to spend hours upon hours creating diagrams and now you can just ask for them and they appear in seconds.

Showing a mermaid diagram in an agentic development environment

1211

Jake Gold @jacob.gold · 13/09/2026

The basic idea is to deflate the value of proprietary frontier AI models in order to reduce funding for new ones.

100

Jake Gold @jacob.gold · 12/09/2026

I guess we really are doing this.

1029815

Jake Gold @jacob.gold · 12/09/2026

That's the whole point!

1110

Jake Gold @jacob.gold · 12/09/2026

It would, I just added this to my post:

080

Jake Gold @jacob.gold · 11/09/2026

I get it, but worth keeping some perspective: - Omarchy $18.5 million (one-time) - The Linux Foundation ~$300 million/yr - The Mozilla Foundation ~$600 million/yr - Wikipedia ~$200 million/yr - Rust ~$5 million/yr - Python ~$4 million/yr - Apache/Gnome/FreeBSD ~$2 million/yr

2110

Jake Gold @jacob.gold · 11/09/2026

Good way to use the non-Fable 50% of your Claude subscription is with Opus as a subagent, called by Fable/Astra/Sol.

050

Jake Gold @jacob.gold · 11/09/2026

A few months ago I hardly used Codex. Now the Codex $200/mo plan has a much higher value compared to the Claude $200/mo plan. Fable still best for the hardest technical/design tasks, Opus only as a subagent. Astra/Sol for the routine tasks. /ask-claude and /ask-codex skills to review and unblock

280

Jake Gold @jacob.gold · 09/09/2026

The hinge is really cool. I might buy the 2nd generation.

4241

Jake Gold @jacob.gold · 09/09/2026

This is a solved problem:

8852

Jake Gold @jacob.gold · 09/09/2026

When switching between coding agent sessions all day, setting a custom theme per workspace really does make context switching easier. Also keeps things feeling fresh. Feeling sluggish? Using a different theme helps a bit.

3120

Jake Gold @jacob.gold · 08/09/2026

I'm currently in SF coffee shops spreading the rumor that someone has figured out how to reliably predict turbulence in fusion plasmas. Pass it along.

1261

Jake Gold @jacob.gold · 08/09/2026

Frontier models are much more resistant to prompt injection than lesser models. This is a big part of why I now trust coding agents running frontier models with a lot more access.

0102

Jake Gold @jacob.gold · 08/09/2026

Was watching this interview with Steven Spielberg about Disclosure Day etc. In it he’s handed my grandfather’s magazine and talks about how it was one of his early sci-fi influences. Nice reminder that even though my grandfather didn’t make money or get much credit, he did do a really cool thing.

1200

Jake Gold @jacob.gold · 08/09/2026

November 24, 2025 Claude Opus 4.5 goes online. For the first time, coding agents could generate high quality code given the right repo, environment, and supervision. IMHO devs that navigated this whole thing well followed something like @karpathy.bsky.social's trajectory:

A few random notes from claude coding quite a bit last few weeks.

Coding workflow. Given the latest lift in LLM coding capability, like many others I rapidly went from about 80% manual+autocomplete coding and 20% agents in November to 80% agent coding and 20% edits+touchups in December. i.e. I really am mostly programming in English now, a bit sheepishly telling the LLM what code to write... in words. It hurts the ego a bit but the power to operate over software in large "code actions" is just too net useful, especially once you adapt to it, configure it, learn to use it, and wrap your head around what it can and cannot do. This is easily the biggest change to my basic coding workflow in ~2 decades of programming and it happened over the course of a few weeks. I'd expect something similar to be happening to well into double digit percent of engineers out there, while the awareness of it in the general population feels well into low single digit percent.

0212

Jake Gold @jacob.gold · 07/09/2026

It's already "weird" not to let your agents play games, right? (actually this is just a fun part of remote desktop evals)

070

Jake Gold @jacob.gold · 07/09/2026

Inconsistent agent config/environment is a problem: 1. Work in declarative, ephemeral sandboxes (e.g. Docker containers). 2. Put everything possible (skills/plugins/configs/markdown) in the repo itself. 3. Use a system that enables running agents with team-shared configs and shared agent memory.

Oliver
@olvrgln
5h
how do people keep their remote devbox in-sync with their local machine? I never have the same skills, random configs, etc. Kind of a pain to make sure both envs are the same

390

Jake Gold @jacob.gold · 03/09/2026

Seems so, but working for me.

010

Jake Gold @jacob.gold · 29/08/2026

Holy Tito's Inflation Tacos $2.90/ea (2017) -> $6.15/ea (2026)

142

Jake Gold @jacob.gold · 28/08/2026

Not driving while not coding was not exactly how I imagined the future.

1026120

Jake Gold @jacob.gold · 27/08/2026

Until recently, I was using Ghostty terminal tabs with a bash script that launched the claude or codex TUI in its own Docker container. Now I use my own thing (clor.com) which does this and a lot more (cloud hosted runners so you can close your laptop, etc). But important thing is isolating them.

130

Jake Gold @jacob.gold · 27/08/2026

She makes me want to curl up under a bush and take a nap too.

1202

Jake Gold @jacob.gold · 27/08/2026

I've been running coding agents in yolo mode in Docker containers for almost two years now. Coding agents need to live inside sandboxes not because they're likely to go rogue due to prompt injection, but because they'll eventually `rm -rf` the wrong thing or trample on each others work.

3171

Jake Gold @jacob.gold · 25/08/2026

This 1958 Disney Paul Bunyan short is even more relevant today than it was originally. Has been stuck in my head since I was a kid.

160