Manoj Bajaj

I'm exploring two areas. Agent-human interfaces: how people and agents actually collaborate. And harness engineering: the scaffolding that turns a capable model into a system you can trust.

I write about that here.

Previously: researcher at Lossfunk, co-founded Ruzo.ai, built authsome.ai (stable, open-source credential broker for agents). Before that: proprietary trading infrastructure at Squarepoint Capital, early-stage at FalconX. IIT Delhi, mechanical engineering, 2017.

Writing

September 10, 2026

Agent Learning Bench: score the learning, not the job

August 19, 2026

Hierarchical memory for coding agents

August 18, 2026

Research Goal

August 10, 2026

Continual learning is the missing half of harness RSI

August 5, 2026

Reason, act, then reflect

March 15, 2026

Agent Identity: Core agent primitive

March 10, 2026

The Agent Economy Needs a New Stack

March 1, 2026

MCP vs CLI is wrong question

February 5, 2026

You’re Still Building for the Wrong User

January 15, 2026

Agent Tooling Should Be Composable, Not Convenient

January 11, 2026

Minimal Agent Primitives

December 20, 2025

Don't Build Agents, Build Primitives.