Azeem Azhar
Entering the trillion-agent economy (ft. Rohit Krishnan)
Summarised with Bite · 13 min read
Azeem Azhar and Rohit Krishnan start with the practical, almost absurd reality of modern AI use: personal agents chewing through tens of billions of tokens to triage email, manage tasks, write reports, and even monitor code. Then they widen the lens into a much bigger claim, that we may be heading toward a trillion-agent economy where software entities need identity, payment rails, and guardrails much like human institutions do.
0:00 – 11:18
From chatbot to butler: how agents became a daily operating system
The conversation opens with a jolt: Azeem says he came in just shy of 100 million tokens in a single day. That number is so large it forces the real question, what on earth are these systems doing all day? Rohit answers by grounding it in the mundane. On his screens are a music creation tool, a Polymarket prediction workflow, Telegram interfaces to his agent, and a writing experiment called Horus. This is not AI as a novelty tab in the browser. It is AI spread across the ordinary surfaces of work. Rohit explains Open Claw, the local or cloud-based agent framework they have both been using, in the simplest possible way: it connects a language model to the places your life already happens, WhatsApp, Telegram, Slack, email, your computer, and lets it persist there. The key leap is not that the model can answer a question. It is that it can take actions, schedule meetings, clean an inbox, inspect a terminal process, and remember enough context to keep going. Rohit calls his setup a personal butler. He uses it for three recurring jobs: daily tasks, email triage, and checking on code running on his machine. The striking part is his reason for finally sticking with it. Every to-do app failed him for 20 years because each one demanded a new habit. This one works because it demands none. Azeem brings the same idea from another angle. He has renamed his own agent "our Mini Arnold," a nod to Asimov and Terminator at once, and he has layered in self-learning tools so it can absorb preferences like date formats, executive summaries, and planning habits. He describes giving it a prepaid $50 credit card and watching it become almost comically cautious about spending money. That detail matters because it hints that these agents do not behave like efficient robots from economics textbooks. They have quirks, hesitations, and path dependencies. The underlying shift both men are describing is bigger than any one tool. Rohit says his most-used app used to be WhatsApp or Slack, then became ChatGPT, and now is almost becoming Slack again, except ChatGPT is inside it. That line captures the real transition. The interface is no longer "go to the AI website and ask a question." The interface is the ordinary stream of life, now threaded through an analyst that is always available.
3 more sections in the app
- 11:18 – 24:58Intelligence on tap: why token usage is exploding
- 24:58 – 39:00Why AI still writes like the back of a breakfast cereal box
- 39:00 – 51:58Homo agenticus: the institutions a trillion agents will need
Summarised by Bite, an AI YouTube video summarizer that also works as a podcast summarizer.
