Sign in

Jake Gold

@jacob.gold
30K followers 4K following 4.8K posts

Building agent multiplexer @ Clor.com SF Bay Area. Helped launch and scale Vibecode and Bluesky. Prev: Nuro, Docker, Google, founder. Addicted to history, computers, and open systems. Unconditional love for all conscious life. jacob.gold

PostsRepliesMedia
Jake Gold @jacob.gold · 48m
Really enjoyed The Odyssey But surprised and a little disappointed to say that the Ralph Fiennes movie The Return did the final scene much better. www.imdb.com/title/tt1986...
imdb.com
010
Jake Gold @jacob.gold · 1h
It's weird that Slack hasn't added support for ACP + diff viewer etc. So that using agents via Slack wasn't such a huge downgrade vs using them with a purpose-built interface. Slack is already fine for non-devs but I don't see how devs live with generating code, it would slow me down so much.
agentclientprotocol.com
Introduction - Agent Client Protocol
Get started with the Agent Client Protocol.
000
Jake Gold @jacob.gold · 4h
What? Agile did not speed up software development by 10x. With respect, that’s just a silly claim to make and there’s no data that supports it. People actually following Agile (which almost no one did) probably moved slower. Open source libraries sped things up but probably not even by 10x.
3290
Jake Gold @jacob.gold · 20h
San Francisco Startup Rule #5823: you're not allowed to judge a person for ditching out on an important meeting if they're going to see The Odyssey in IMAX 70mm the day before it stops playing at The Metreon.
1140
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.
1210
Jake Gold @jacob.gold · 29/09/2026
I knew there was a reason I had an extra cup today, I just didn’t know why. Until now. Turns out I was celebrating.
1170
Jake Gold @jacob.gold · 29/09/2026
If AI is as big as the Industrial Revolution that’s pretty scary The Industrial Revolution took hundreds of years to unfold, giving civilization time to adapt. AI will absolutely not take hundreds of years. We’re all just hoping it takes at least decades.
3110
Jake Gold @jacob.gold · 29/09/2026
trust or no trust?
3110
Jake Gold @jacob.gold · 28/09/2026
This is great. PLC being centralized is a legit weakness in the Atmosphere's story and this goes a long way to mitigating it. I still want @dholms.at to be the PLC CEO, regardless of whether it makes any sense or not. We had a deal, kinda.
1330
Jake Gold @jacob.gold · 28/09/2026
Jensen got into GPUs for the love of the game. Literally.
0110
Jake Gold @jacob.gold · 28/09/2026
FYI, ChatGPT Voice mode is finally good enough to tutor you in Japanese! (and probably other languages) I've tried every version since it was released and it didn't work well enough. It's still not perfect, but the endless patience and lack of embarrassment make it incredibly productive.
2171
Jake Gold @jacob.gold · 28/09/2026
I feel exactly the same way about the kind of LLM-based AI we have today. But the AI x-risk argument is about what happens when we have "real" AI that can actually think like humans but a million times faster. There are no real experts on this and yet it does seem worth worrying about in advance.
170
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
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
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 · 24/09/2026
This is what giving a shit looks like in a startup btw. Same energy I brought when I was on the team and pre launch I was told we weren’t planning to translate the app, and then I pushed back hard and helped improve the Japanese translation so we could have million Japanese users sign up on launch.
1440
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
Every repo I've had access to in the last 6 months looks something like this. Productivity has increased ~10x for people who are taking advantage without becoming reckless. This is the new norm for software engineering.
2311
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
We should expect them to have something similar in months.
190
Jake Gold @jacob.gold · 18/09/2026
When processing complex image/video formats/codecs, your choices are: 1. Use a memory safe language and don't bypass it with unsafe code 2. Do the work in disposable unpriviledged sandboxes 3. Get hacked
0233
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 · 18/09/2026
Software dev is all the 2nd 80% now.
1150
Jake Gold @jacob.gold · 17/09/2026
x-risk vs xxx-risk
092
Jake Gold @jacob.gold · 17/09/2026
It is difficult to get a man to understand something when his sex life depends on his not understanding it.
29511
Jake Gold @jacob.gold · 16/09/2026
Most likely near-term AI security apocalypse is a zero day exploit in Supabase, or another service that exposes database access over the public internet.
0241
Jake Gold @jacob.gold · 16/09/2026
There's a very strong comparison between falsely yelling "fire" in a crowded theater and broadcasting that there's a ">10% chance AI could kill all humans within the next decade" to a billion people. It was cute 10 years ago (when it was just us nerds) but now it's time to stop.
1112
Jake Gold @jacob.gold · 15/09/2026
Funnily enough, this looks like most companies that were started pre-AI coding. New companies don't have as many cooks in the kitchen or legacy concerns to deal with. So they can just plan, generate, review, plan, generate, review, commit, push to staging, push to prod.
070
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 · 14/09/2026
Warn people about AI risks, explain what could go wrong, and make the strongest argument you can, but please stop making up numbers.
jacob.gold
Please stop quantifying your AI existential risk vibes
Last week Evan Hubinger, who leads alignment science at Anthropic, said he thinks there’s a greater than 10% chance that AI kills all humans in the next decade. Bryan Cantrill wrote a good response ab...
1011310
Jake Gold @jacob.gold · 14/09/2026
The Bay Trail is how I ward off full blown AI psychosis.
1460
Jake Gold @jacob.gold · 14/09/2026
If William Gibson says you're living in the future, then you're living in the future.
062
Jake Gold @jacob.gold · 14/09/2026
Great post, but I do think it went slightly too far with the ending: "...you had nothing to fear then — and (at least with respect to extinction risk!) you have nothing to fear now." These are both extraordinary claims that require evidence: a) 10% existential AI risk b) 0% existential AI risk
2110
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
In case you weren't aware how the anonymous Hacker News moderators control the narrative in Silicon Valley. (Bluesky Communities can't come soon enough)
47112
Jake Gold @jacob.gold · 12/09/2026
I guess we really are doing this.
1029815
Jake Gold @jacob.gold · 12/09/2026
Okay I blagged it. Imagine if this actually fucking happened‽
jacob.gold
An open letter to Dario: if you mean it, open the weights
Dear Dario, You published We Must Pace the Frontier today, which committed Anthropic to embedded third-party evaluators and asked governments to require every other frontier company to match. Sam Altm...
1017043
Jake Gold @jacob.gold · 12/09/2026
There's no slowing down frontier models by regulation that wouldn't lead to regulatory capture. But we could simply force all AI model companies to release their model weights openly. It's the world's data after all and nothing would slow their funding/progress more effectively and universally.
1125442
Jake Gold @jacob.gold · 11/09/2026
If you're still running coding agent TUIs in terminal tabs, moving to an agent multiplexer GUI with isolated workspaces is a major upgrade. With each task in its own workspace, you can manage more work in parallel and start being more ambitious about what you build.
0201
Jake Gold @jacob.gold · 11/09/2026
1. Generate and push artifacts to a git repository (ideal) and/or cloud storage. 2. Use a memory system that stores/searches all agent conversation history (transcripts) for 7-30 days, across all agents/computers.
1172
Jake Gold @jacob.gold · 11/09/2026
Yeah, be safe, download your brain from the computer.
1100
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 · 10/09/2026
As is often the case, Sean Carroll has the informed and sober take that I agree with. I'm 1000x more worried about three bitter, very old men with their fingers on their nuclear buttons.
1161
Jake Gold @jacob.gold · 10/09/2026
What works for me is giving myself permission to start on the easiest/most fun part, as long as it's in related to the goal. Once I build some momentum, it's easy to keep it going and grind away at the hardest stuff.
0121
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
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