Robb Wilson
I co-founded OneReach.ai and run it as CEO. We build the software that keeps AI agents working inside large organizations, and to me it is a memory system with a runtime built around it.
Getting an agent to answer a question is the easy part now. Keeping it the same agent through a long project is harder, because by then it has to remember what it has done, what it is trying to do, and which facts it can trust. Take that away and what’s left is a prompt running in a loop. The same gap stands between a company and what Josh Tyson and I call organizational AGI, a system that knows the organization well enough to run large parts of it. Memory is the part I work on, and the subject of a book I’m writing with Roger Forsgren, former Chief Knowledge Officer at NASA.
What I argue
Memory is more than a database. When people say an agent needs memory, they usually mean a place to keep things, like a vector store or a graph. You need one. But a store alone doesn’t keep an agent consistent from one task to the next, stop it reading what it shouldn’t, or let it learn from last week’s work without repeating last week’s mistakes. Those are the jobs I mean by agent memory.
What it looks like running. When I demonstrated a working agent on Invisible Machines, its memory was on screen in parts: reference memory holding canonical knowledge, the organization’s agreed source of truth; what it was taking into account for the current step; and a record of what it had done and meant to do next. I also showed how the second version of its project, an agent to help schedule our podcast, would begin by evaluating the first, with points along the way where a person could step in or let it run.
Where it leads. In Age of Invisible Machines, Josh Tyson and I wrote that the “organization” part of organizational AGI is “the context that a generally intelligent machine needs to do things the way a team or company wants them done.” Most of that context is memory: the facts the company has agreed on, what its agents have already done, and what each of them may see. Without that context, an agent doesn’t know how your company wants things done.
Related reading:
- What an Agentic Runtime Environment Is—and Isn’t, on what the software under an agent has to do, memory included
- LangGraph vs. CrewAI vs. OneReach GSX, my comparison of agent platforms, including when you don’t need ours
Bio
Short: Robb Wilson is co-founder and CEO of OneReach.ai and co-author of three books, including Age of Invisible Machines (Wiley) with Josh Tyson. He coined the term organizational AGI (OAGI).
Longer: Robb Wilson is co-founder and CEO of OneReach.ai, which makes a platform for running AI agents in large organizations. Earlier, he founded EffectiveUI, a user experience agency that WPP acquired in 2012. He is co-author of three books: Age of Invisible Machines: A Guide to Orchestrating AI Agents and Making Organizations More Self-driving (Wiley, second edition 2025) with Josh Tyson; WISER Method Master Playbook (First Strategy, 2026) with Anthony Franco, describing a method drawn from Wilson’s work at OneReach.ai; and Effective UI (O’Reilly, 2010) with Jonathan Anderson, John McRee and the EffectiveUI team. He coined the term organizational AGI (OAGI) for “the general intelligence it takes to run vast portions of a business.” He is writing a fourth book, on agent memory, with Roger Forsgren, former Chief Knowledge Officer at NASA. He founded UX Magazine and donated it to EffectiveSC, the nonprofit that now publishes it. He co-hosts the Invisible Machines podcast with Josh Tyson and has written for Harvard Business Review and Fast Company.