Serj Goldman

AI as a working tool

I use AI every day, where it actually saves time.

See my agentAll areas
Stack
LLM agents, Telegram bots, Python
Use
Production, analytics, documents
Position
Mechanics to the machine, decisions to me
instances of the system run every day: mine and one deployed for another person

Two instances: mine and one for another person

My instance

  1. My task
  2. My agent
  3. My knowledge base
  4. The decision stays with me

The other person’s instance

  1. Their task
  2. A separate agent instance
  3. A separate knowledge base
  4. The decision stays with them

What I built

  • A personal working system on my own server: a Telegram bot as the interface, a knowledge base, agents on a schedule.
  • Automations that run every day: inbox triage, daily digests, search across conversations, documents, calendar.
  • A platform for several bots with different roles on one framework.
  • A second instance of the system, deployed separately for another person.

What it changes in production

  • A concept and a deck come together in an evening.
  • Data and documents get sorted without waiting in line for an analyst.
  • A draft spec, letter or report is already there, so the time goes into the decision.
  • Producer routine (statuses, digests, reminders) moves off people and onto a schedule.

The machine collects statuses and digests. I make the calls.

I built all of this myself and use it every day. So I talk about AI in production from practice: I know what speeds a team up and where the tool just gets in the way.

AI is good at the mechanical part of the work. Decisions and accountability stay with people.