Our Story

Agents have been part of computing for decades — from the software daemons and expert systems of the 1970s to the distributed, autonomous agents of the 1990s. Back then, the goal was simple: programs that act on a user's behalf, automating repetitive work and managing specific domains with minimal human intervention. The field had a name long before the marketing showed up.

In 2006, Oleg Popov authored "Formal Models of Agents" as his master's thesis — a study of BDI architectures, FIPA-compliant communication, and the formal semantics of autonomous decision-making. Twenty years later, the same primitives are load-bearing: typed graph edges, explicit goal hierarchies, and provable handoff boundaries. The 2023 generation of "AI agents" rediscovered what the 1996 generation already knew — and then forgot the parts that mattered most.

When generative AI arrived, it didn't introduce agents to the world — it amplified them. By combining deep experience in agent design with generative models and adaptive learning, we redefined what agents could do at production scale. The 2024 lesson — fully autonomous SDR doesn't work — was a lesson the formal-agents lineage had already learned: the agent stops at the judgment boundary. Always.

That evolution became AgentStruct — a consultancy built to design the boundary between agent and human in GTM-agentic systems. We don't sell autonomous bots. We design the loops, the handoffs, the escalation policies, and the audit trails that let revenue teams scale judgment, not just volume.

We don't sell agents. We design the boundary between agent and human.