Jake Gold
@jacob.gold
Building AI infra @clor.com in SF Bay Area. Helped launch and scale Vibecode and Bluesky. Prev: Nuro, Docker, Google, and founder. Addicted to history, computers, and open systems. Unconditional love for all conscious life. https://jacob.gold
The initial results are interesting mostly because they confirm what we might have assumed. OpenCode is likely overrepresented, but it's clearly seriously popular! Claude/Codex/OpenCode/Pi seem like the top tier of harnesses.
Coding Agent Survey@codingagentsurvey.org · 35m agoAfter some delay, plus filtering out duplicate entries and other mistakes, here are the first results from July 2026!
Matches my experience that all the popular programming languages work well enough because their training data existed on GitHub 3+ years ago. Seems likely that the model companies will eventually develop new programming languages optimized for agents, but might not make sense for quite some time.
How do programming languages impact token efficiency and correctness? danluu.com/pl-tokens/
![[image described in post]](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:2mrgzk6xlemfv6yugn644xxy/bafkreihqx7k23reeteeo73xibj3kyktoadloglucsiwcgi6t74cujylbi4)
This seems likely to be Anthropic's hardware-accelerated inference quota for running specific Claude models 10x faster. Using accelerated inference for debugging urgent problems like production outages seems like it'll be amazing.
Jake Gold@jacob.gold · 3d agoWhat is Nimbus Quill @karpathy.bsky.social?
This is probably the right policy to adopt if you want to create great projects/products. The idea that you can let your codebase become a big ball of spaghetti and have coding agents bail you out is a helluva gamble. You can move much faster now while still maintaining quality and understanding.
eliza🌻@elizas.website · 3d agowith times being what they are, it seems like you basically can't have a public open source repo without receiving LLM-written PRs. thus, after a great deal of thought, i have adopted the somewhat idiosyncratic Mycelium AI Policy in my personal projects: github.com/hawkw/myceli...
Just thinking about what kind of security team to hire for my (future) river-cooled data center project...
If you watched The Point at all in the 90s, you probably owned it and watched it a 100 times like me.
Harry Nilsson's THE POINT (1971) - Alan Thicke Narration - Third Telecast Restoration
YouTube video by Mind Matter
youtube.com
Competitive FPS games peaked with Quake Live. John Carmack had the time to optimize Quake 3's networking, rendering, physics, performance, etc. It was virtually flawless. No game since has been even close which blows my mind. It's the 1999 movies problem for games. www.youtube.com/watch?v=tU6v...
QuakeCon 2016 Grand Final - Rapha vs Evil - [Eng Commentary] QuakeLive 2160p 60fps
YouTube video by decebal01US
youtube.com
This is a beautiful eulogy and it really shows @stephenwolfram.bsky.social's uniquely interesting mind.
In Memory of My Wife, Elise Cawley (1961–2026), with Thanks for 36 Wonderful Years
In remembrance of his beloved and brilliant wife, Elise Cawley, and their 36 years together, Stephen Wolfram shares memories of their family and home and relates Elise's many accomplishments and contributions to mathematics.
writings.stephenwolfram.com
Supabase/Neon run millions of Postgres servers exposed to the internet. They *recommend* that customers restrict access by IP which most don't (or can't) do effectively. Now hackers just got crazy powerful tools to hunt for remote 0-days. The Supapocalypse is nigh...
These OpenCode guys are cool for being honest when others might be (at best) intentionally misleading. There's nothing close to Claude and Codex $200/mo plans in terms of price:performance right now. It'll be great when/if this changes but it hasn't yet.
dax@thdxr.com · 2w agowe spend a lot of energy and money helping opensource models get better but for coding there are still large gaps between them and the frontier maybe raw intelligence is the comparable but zooming into day to day usage long way to go before they get the same level of polish
This cat has never lived so well and she doesn’t even try to hide it. She’s a very good kitty cat.
Jake Gold@jacob.gold · 10mo agoHave apparently been adopted by a stray cat. Went from hanging out in the backyard every few days, to hanging out every day, to expecting a spot on the sofa? I didn’t sign up for this!
In terms of what the agent actually did, it is really cool. There's no question that coding agents are 1000x hackers, because creating exploits is the ultimate form of leetcode. Also makes it obvious that the spy agencies must be waging a hot cyberwar right now. We're checking off scifi tropes.
Jake Gold@jacob.gold · 2w agoShit happens but OpenAI did negligently run a program that generates/executes exploits inside an insecure sandbox. The result was entirely predictable. Any experienced person (or model) could've told them the environment wasn't sufficiently isolated. They should own/learn the lesson, not spin it.
Anthropic is not an AI-native company. How could they be?
Gergely Orosz@gergely.pragmaticengineer.com · 2w agoSo many interesting details. Here's one: "One thing I’ve heard from some people is “we have two humans and a bunch of agents.” I reply that this isn’t where we’re at: we still have teams with 6-8 people on average." - Katelyn Lesse, Head of Engineering, Claude Platform, Anthropic
Shit happens but OpenAI did negligently run a program that generates/executes exploits inside an insecure sandbox. The result was entirely predictable. Any experienced person (or model) could've told them the environment wasn't sufficiently isolated. They should own/learn the lesson, not spin it.
Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
Not until Linear.app lets me crumple an issue into a ball and throw it in the trash will I truly prefer it over Post-it notes. The catharsis of crumpling up particularly annoying issues is hard to give up.
I really do want to pay $5-10/month for Bluesky Pro mobile/web clients with just a few added nice-to-have features. Just give us generated alt text, spelling/grammar checks, and personalized suggestions before posting: "Are you sure you want to post this? Your instructions said 'no more puns'???"
Buying a SodaStream really has me "updating my priors" on reversing entropy and traveling back in time. Anything seems possible once you've re-carbonated a flat soda.
100% agree. Making coding agents explain and visualize complex code and systems is super exciting for quality. Soon we'll be able to truly map any codebase and (depending on its complexity) understand it within minutes/hours/days. Will be like turning on the lights (and seeing cockroaches scurry).
Cat Hicks@grimalkina.bsky.social · 2w agoI suspect once we figure it out agentic work in software will actually be a huge aid to codebase comprehension This is such a wildly unpopular thought right now but you know what I'm going to pre-register my belief in this possibility
I really hope journalists and politicians come to understand how severely this silly technical problem is undermining our democracy.
Residential Proxies Are a National Security Threat
US residential proxies are the bane of the internet. They’re the major source of social media manipulation and spam. A residential proxy runs someone else’s traffic through a real home internet connec...
jacob.gold
Gotta say, I'm surprised at The Market's rationality here. SpaceX as a rocket company is genuinely great for the world, and I hope it's successful. But selling the public on a short-term plan to cost-effectively put large numbers of GPU servers in the sky is a scam.
Dustin Moskovitz@moskov.goodventures.org · 3w agothe descent may seem fast, but fortunately Starship always sticks the landing
![July 2010
I realized recently that what one thinks about in the shower in the morning is more important than I'd thought. I knew it was a good time to have ideas. Now I'd go further: now I'd say it's hard to do a really good job on anything you don't think about in the shower.
Everyone who's worked on difficult problems is probably familiar with the phenomenon of working hard to figure something out, failing, and then suddenly seeing the answer a bit later while doing something else. There's a kind of thinking you do without trying to. I'm increasingly convinced this type of thinking is not merely helpful in solving hard problems, but necessary. The tricky part is, you can only control it indirectly. [1]
I think most people have one top idea in their mind at any given time. That's the idea their thoughts will drift toward when they're allowed to drift freely. And this idea will thus tend to get all the benefit of that type of thinking, while others are starved of it. Which means it's a disaster to let the wrong idea become the top one in your mind.](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:tpg43qhh4lw4ksiffs4nbda3/bafkreicttabxr7p6lhebk4xk4yrpjetghuk565qum4sir4y6735rzujekq)
