Been working on set up. I run 4 different ventures and really think Hermes could help. Can’t get past it losing connection to iMessage though. Always on box at home and all setup with photon is confirmed correct. If I don’t text for a while though, Hermes shuts off and I have to ssh into my always on box to try and resolve. Bit frustrating. I’m sure I’ll figure it out.
Holy shit. This is what I have been feeling but couldn’t put the words to.
The thing is, some of the rituals and routines I actually enjoy now. The problem was thinking they all needed to fit within the prison I built for myself. Rigid and dogmatic. If I miss a day, or 3, that’s fine.
Instead trying to do it all at once, leading to burnout, do the most important one. Then the one after that 6 months later.
I feel refreshed reading this. Thank you.
After 18 months running The Matthews, I see the same 5 patterns of operational bleeding in small arts orgs:
→ Trapped Knowledge
→ Relationship Decay
→ Calendar Pressure
→ Reporting Drag
→ Tool Sprawl
Built a quick quiz to find which one's hitting you hardest:
https://t.co/Gf3igks76Z
I rewrote a paragraph on my consulting site this week.
The old version sold "custom operations systems." The new version sells "the operational spine your team actually needs — rendered however your people work."
Same service. Completely different sale.
Here's what changed my mind: Salesforce went fully headless. When the biggest name in business software strips out the dashboard entirely and sells only the API, the signal isn't subtle. The UI was never the product. The data model and the workflow logic were.
I've been building exactly that for nonprofits for three years. I just didn't have a word for it.
Most small orgs have tried a dashboard-based tool and bounced. They bought a CRM, a donor platform, a volunteer scheduler. After six months, nobody logged in. The instinct is to blame the team. That's wrong. The tool was the wrong shape.
If the spine is right — the donor table, the partner table, the workflow logic — the rendering meets people where they already are. Their inbox. Slack. A Friday digest read in the parking lot. No dashboard. No login. No tab.
The question stops being "what software do we buy." It becomes "what should our operational spine actually do, and how should it show up for each person on our team."
Different questions. Different answers. Better outcomes.
Welcome Salesforce Headless 360: No Browser Required! Our API is the UI. Entire Salesforce & Agentforce & Slack platforms are now exposed as APIs, MCP, & CLI. All AI agents can access data, workflows, and tasks directly in Slack, Voice, or anywhere else with Salesforce Headless 360. Faster builds, agentic everything. 🚀
#Salesforce #Agentforce #AI
https://t.co/mxySdJS7HR
I've been building what I thought was an AI operations system. 14 agents, cron schedules, automated morning briefs, the whole thing. It works. My agents write emails, find grants, draft content, run my nonprofit's comms pipeline.
Then Karpathy posted this and I realized I'd been building the automation layer while neglecting the thing that actually compounds: the knowledge.
My agents were producing genuine insights every week — "follow-up decay is systemic across all ventures" — but those insights evaporated into Telegram messages and log files. Nobody filed them. Nobody updated the wiki. The next week's agent run re-derived everything from scratch.
So I sat down and rewired it. Built a shared operations module so every vault mutation keeps an index and a log. Enhanced the ingest pipeline to update existing notes instead of just creating new ones. Wired all 7 agents to compile durable insights back into the wiki after every run. Added a /file command so when I ask a complex question and get a good answer, it goes into the vault instead of disappearing.
The vault went from 52 hand-written notes to a living, LLM-maintained wiki that grows from three directions: what I add, what agents discover, and what I ask.
Karpathy's right — there's room here for an incredible product. But if you're technical enough to have Claude Code or Codex, you don't need to wait for it. Share the idea, let your agent build it.
Wow, this tweet went very viral!
I wanted share a possibly slightly improved version of the tweet in an "idea file". The idea of the idea file is that in this era of LLM agents, there is less of a point/need of sharing the specific code/app, you just share the idea, then the other person's agent customizes & builds it for your specific needs.
So here's the idea in a gist format: https://t.co/NlAfEJjtJV
You can give this to your agent and it can build you your own LLM wiki and guide you on how to use it etc. It's intentionally kept a little bit abstract/vague because there are so many directions to take this in. And ofc, people can adjust the idea or contribute their own in the Discussion which is cool.
Most organizations don't have a capacity problem.
They have a design problem that looks like a capacity problem — so they hire staff instead of fixing the workflow.
The first fix is free. The hire isn't.
The Matthews doesn't own a wood shop.
We found one. NHTC has a heated facility, tools, storage, and crew — and they said yes.
First production using it: May. Most small nonprofits think they need to build capacity. Often they just need a phone call.
The most useful feedback I get on Pulley comes from running The Matthews.
Not user interviews. Not surveys. Just hitting an operational wall at 7pm and noticing which problem the software should have already solved.