Dmitry Kalichkin

3x founder, Stanford MBA, not an engineer → fell down the AI rabbit hole. My AI agent has a card, pays its own bills (mine too), manages a CRM, and calls me to remind me about laundry. Exploring agent infrastructure. Documenting everything here.

I solved this just a couple of days ago – well, my agent did. It signed up for @Vapi_AI entirely on its own, used its own payment card (pls don't tell Revolut about it), and provisioned a real US phone number. No human touched anything. Now it handles inbound + outbound calls for me with full memory, CRM, and calendar access. If it can't answer, it escalates to me via Telegram. ~$0.10/min, ~250 lines of Node.js. Feel free to feed this repo to yours: github.com/kalichkin/agen… As @danbeksha said – it's easier for an agent to get a card than a phone number. Mine now has both!

What's the easiest way for an agent to get their own phone number? Google voice (doesn't work w/ What's App), Twilio (have to set up fake business), Actual phone line (very cumbersome or blocked by IT)...

@Simon_Coates @UKBackintheDay2 hyper-niche reference!

@UKBackintheDay2 "Sash or Chicane?" "Sash" "Arrrrrr, these blaggers are pants."

It’s been 29 years… and the drop still DROPS

i'm writing a language processing and analysis pipeline. i added 10k recipes to it and started doing correlations (because I want to be an entirely theoretical chef). here are some things i have discovered:

I have been following Christian's work since 2017, and this is an outstanding piece.

Today, my OpenClaw agent Proshka paid a parking ticket and booked a hotel on my behalf. Not suggestions, actual financial transactions. It felt like crossing a real threshold: from automation to delegated agency. #openclaw

@meetahsen agree with the painpoint, and don't have a good solution yet.

@Kalichkin observability's key but the real killer is agents fighting over shared state. solved it with a gossip protocol that felt like herding cats. what'd you use to trace cross-agent calls without drowning in noise?

1/ I'm running 6 AI agents. Different tasks, different sessions, messages flying everywhere. Last week I realized I had zero visibility into what any of them were actually doing. So I built a dashboard. #BuildInPublic #AIAgents #OpenClaw

12/ Three weeks ago I never touched Claude Code. Today I'm running 6 agents with a custom dashboard I forked and built myself. If you're a founder, operator, or builder: you don't need an engineering background to do this. You need stubbornness and a few late nights. Links are in tweet 9. Go build!

11/ Next up: adding task dependencies, agent-to-agent handoff tracking, and a timeline view of overnight activity. Because the agents work while I sleep. And I want to see the replay in the morning.

Friday evening thoughts on building with AI. Been working on a multi-agent dashboard for the past week and the experience has been absolutely wild. More details coming soon.