About this session
Over the past three years, we’ve seen three distinct waves of AI adoption: access through chat interfaces, reasoning through advanced models, and most recently, agentic coding through tools like Claude Code & Codex.
But something changes when you move beyond demos and actually run agents continuously in the real world.
In this talk, Ashish shares what he learned from running OpenClaw, an open-source personal AI system that automates tasks, manages context, and operates across tools on behalf of its user. Not as a prototype, but as a system that runs daily, makes decisions, and occasionally fails.
Three patterns emerge.
First, the harness matters more than the model. Models are increasingly interchangeable, but the orchestration layer that connects memory, tools, and workflows becomes the real source of reliability and control.
Second, memory and instructions act as force multipliers. Systems that preserve context and evolve instructions over time become more aligned, more efficient, and significantly more capable than stateless interactions.
Third, sustained use reveals what demos hide. Reliability, failure modes, and real-world usefulness only emerge when agents operate over days and weeks, not minutes.
This talk introduces a new mental model: AI is shifting from tools you use to systems that work for you. We’ll explore what this means for developers, product builders, and organizations designing the next generation of AI systems.
