
Leading @Precurion 4× founder (Arkive, Reflectly, Bounce)
Joined October 2011
- Tweets826
- Following740
- Followers456
- Likes1.3K
I built a Vault-Tec terminal to run my agents. 8 projects, one overseer, and the most fun I've had at work. @herdrdev 🔉
2
3
156
My co-founder got annoyed with how hard it was to stay on top of his agents in Herdr.
So he made it a game.
I’m addicted to this
51
We pushed our CI usage past our AI spend last week.
Our cracked teammate + Jog (@mattgapp) made that pace affordable.
Adapt your systems to your new capabilities. Don’t limit your capabilities to fit your systems
Talk about market timing. We were starting to see our CI bills climb rapidly. So far @mattgapp and the team are kicking ass!
We just deleted every reason to use Github’s runners for private repos. We blew them out of the water.
We’re giving away 5,000 free minutes. And when those run out, we’re 80% less. Oh, and you can get started for as little as $20 on Jog Personal.
If you want more free CI, I’ll give you an extra 5,000 minutes for every referral who starts with Jog. Just DM me with their name and I’ll credit your account when they start.
Let your Claude and Codex go wild.
Enjoy!
I’m really starting to like Rust for speeding up our infrastructure. Now I’m writing cleanup code because the build artifacts keep filling my drives.
Accidentally building a literal “Rust garbage collector”
2
99
My agents in Herdr noticed we were hitting compute limits, so the orchestrator built a queue.
Now they’re all waiting for their turn while the machines sit at 10% usage.
AI reinvented middle management
1
101
dot voice is really cool and Astra on tap is what I want in my assistant. But the rest feels buggy and underwhelming compared to Grok Bot and Muse.
Send a picture: “I can’t see it”
“Let me check that for you”: 🦗
@sama, didn’t quite stick the landing 🚀
2
88
I just told my herdr orchestrator to fire a middle manager. Not sure how I feel about the delivery
Did he get sent to Mobile Avenue?
2
81
I'm finally building intelligent systems with AI. Still nothing next to watching my daughter's little sponge of a brain learn something new 😍
1
2
73
Old ideas can become revolutionary when pushed far enough.
Google wrote the transformer paper. OpenAI showed how far it could go.
Classifiers have been around for decades. TypeSafe is a good reminder that the breakthrough isn't always a new primitive. Sometimes it's making an old one dramatically better.
At @precurion , we're already seeing how much credit work is classification. Moving more of that out of the LLM path should make underwriting faster, cheaper, and more reliable.
Excited to deploy this with partners over the next few days.
typesafe.ai/
2
81
Okay, I'm sold. Grok Bot is really well executed. Well done @jordantop and the team at Cursor.
I've been missing my always-on agent ever since my OpenClaw setup went down.
Onboarding and connecting tools was a breeze. The browser handoff is smooth. And the agent-over-chat approach makes things feel continuous in a way a ChatGPT chat just doesn't. You give it work, you come back, and it still has the thread.
I filed an insurance claim through ChatGPT a while back. Policy docs, photos, the back and forth with the broker. The kind of job that usually dies the second you close the laptop. That's when I got what this is actually for.
It's the same shape as collecting evidence for a compliance report. Or a daily digest of the business. Or the emails that have been sitting unanswered because nobody owned the follow-up.
You could always ask a chatbot to help with any of that. The difference is you can leave.
1
229
NatWest took payslip PDFs out of the mortgage path.
A PDF is a claim. Claims are cheap to fake.
Income from payroll data is a record. Different object.
That's the distinction we built around.
2
123
Opus 4.8 feels smarter.
More confident.
More personal.
Talks a lot.
I suppose that uses tokens.
Smart Anthropic.
I gave it a project.
I asked if it finished.
It said it finished.
It finished 70%.
I pointed at the rest.
It said "you're absolutely right."
It said it was fixing it now.
It was not fixing it now.
I handed it to Codex.
Codex didn't tell me I was right.
It just finished.
2
95
AI makes coding faster.
It does not make the last 20% shorter.
It just gets you up the wrong hill quicker.
Trigger happy is the new slow.
1
51
I feel like we're in the "dumb phase" of Codex 5.5's cycle right now. 5.6 must be about to launch
47







