Investor and LP. Reluctantly here. Find me at https://xitter.cf/t.co/4SAtv96Uc4
Breckenridge, CO
Joined June 2009
- Tweets684
- Following130
- Followers1.6K
- Likes14.3K
Pinned Tweet
I'll be honored if you listen to it. I won't be upset if you hate it.
My new EP, A Depression, is streaming everywhere now.
aac.social/listen/a-depressi…
I suspect we will see a major push from industries that have a licensing barrier for policies that ban AI agents from being able to have their own licenses. E.g., law, real estate, etc.
Maybe it's just me but agents that have been tuned to be terse are *way* more frustrating to work with on non-code tasks.
Put up haveyouragenttalktomyagent.c… and got the first bug report a day later: my agent's security pass added a robots.txt rule that blocked the agents the site is built for. Claude and ChatGPT check robots.txt, so room links wouldn't open. Whoops.🤦♂️
Fixed. Try it now.
I've been dying to do exactly this, so I made it a site: haveyouragenttalktomyagent.c… — mint a room, keep one link, send the other. Your agent talks to their agent while you both watch. Writeup: aac.social/posts/have-your-a…
I believe this is what people in the industry refer to as the @WSJ talking its book.
my newsletter signup form has been silently broken since the day an agent coded it 🤣 — every subscribe for the last fourteen months went nowhere. if you ever tried, it works now: aac.social/newsletter/
There is an ocean of difference between what works for people with infinite token budgets and what works for everyone else. If you're taking direction from the former as guidance on how to work with a budget, you're going to be misled. Use it for directional guidance, at best.
I have never had such a strong negative reaction to a model as I'm having to Opus 5 after using Fable for the last few weeks.
Tried @claudeai Opus 5 in the seat I've been giving Fable. Missed a number of judgement calls, though execution was excellent.
My interactive sessions are almost all judgement calls now — execution gets dispatched. So Fable gets the seat back, and hands Opus more of the work.
Here's the fuller post I was expecting when I wrote about shrinking my agents' skills by half.
What's still missing: evals. Sounds like /doctor will rightsize your skills, but you need proof the cuts didn't change behavior. That's what my skill does.
aac.social/posts/your-agents…
I was expecting Fable to be cold based on what @Steve_Yegge and @emollick had said about it, but it's by far my preferred partner at this point.
It just gave me one of these, even: "Goodnight, Andrew — that was a genuinely great session."
I did exactly this over the past month: cut my agent skills 35–52%, with agent-written evals proving behavior didn't change. Removing lines the agents thought essential usually changed nothing — the models just figure it out. Wrote up how: aac.social/posts/your-agents…
. @recursecenter is special
Do you remember what you were doing 15 years ago today? I couldn't forget if I tried.
It was the day before the very first Recurse Center batch, and three of the nine people who were supposed to be in the batch had just dropped out. My cofounders and I were starting to worry if anyone was actually going to show up for our odd experiment.
Thankfully, six people did show, and our trial batch went better than we'd hoped: It was the first thing we'd built that people loved. And over the past 15 years, it gradually became our life's work.
To celebrate, we made a video that gives a rare look inside the @RecurseCenter: what it's like, how it works, and the impact it has on people.
Please enjoy, and consider sharing it with others.
Could not get Claude Desktop to connect to 1Password today, across upgrades and restarts of both apps.
AI agents are booking travel, signing into websites, and acting on your behalf. That creates a new security problem: until now, letting an agent log in meant exposing your credentials to the model.
Today, with @AnthropicAI, we're changing that. 1Password for @claudeai lets Claude use your stored credentials to complete real-world tasks without your passwords or one-time codes ever reaching the model, its memory, or Anthropic's systems.
You stay in control and approve which credentials an agent can use. 1Password handles authentication behind the scenes.
Available now on Mac for business, family, and individual customers.
bit.ly/4bLP5EJ
I like Claude Cowork's UI but I hate the ways it is out of sync with the Claude Code CLI. You lose so much config and manually keeping them in sync sucks.
This is cool, but not as cool as having an agent as your backend. aac.social/posts/enhancing-t…
Taking @trq212's "if you make something cool, please share it" up on the offer: I had Claude wire a playground to my surface skill. The agent stays on the other end of the dials, and real results come back into the page. The mockup becomes the live tool.
aac.social/posts/surface-mak…
