Igor Rikalo, President and COO of supply chain planning firm o9 Solutions, published a Forbes Tech Council piece on Monday projecting that by the early 2030s, a senior product manager might oversee 180 AI agents and a supply chain head could coordinate 300. The scenario he sketches is specific: a senior executive starts the day reviewing 60 AI-generated briefs, gets mid-morning approval alerts from three agents requesting capital reallocation based on real-time market signals, and by afternoon signs off on a multimillion-dollar inventory move flagged by a supply agent.
What matters in Rikalo’s piece is the framing of what actually breaks — the projected numbers are secondary.
The Cognitive Load Argument
Rikalo argues the limiting factor for agentic AI adoption will not be data, compute, or capital. It will be how well humans can maintain judgment quality as agent outputs scale. He identifies three specific failure modes: attention fragmentation across hundreds of agent signals, decision fatigue from constant prioritization, and overreliance on automated recommendations that erode a manager’s ability to catch low-frequency, high-impact signals.
This is a second installment. In a June 2 Forbes piece, Rikalo made the predecessor argument: that when every knowledge worker supervises 100+ agents, the “individual contributor” role effectively disappears. Everyone becomes a manager, but managing agents rather than people. Delegation, he argued, becomes a core technical skill rather than a soft one.
The July piece sharpens the claim: scaling agent count is easy; each additional agent adds cognitive overhead to the human in the loop, and human cognitive bandwidth does not scale.
Where the Numbers Collide With Reality
Rikalo’s projections are aggressive. He cites NVIDIA CEO Jensen Huang’s prediction of 100 agents per employee as a baseline and nearly doubles it. What he does not address is how far enterprise adoption currently sits from that target.
NCT’s coverage of the VentureBeat Research survey of 573 enterprise leaders, published July 11, found that most deployed “agents” are single-prompt chatbots, 86% of GPUs run at half capacity or less, and governance controls lag deployment by six to twelve months. The gap between “200 agents per employee” and “most enterprises cannot even define what constitutes an agent” is substantial.
This gap reframes the timeline without invalidating Rikalo’s thesis. The cognitive bottleneck he describes is real, but for most organizations it sits behind a more basic question: whether they have the infrastructure, the data pipelines, and the organizational clarity to deploy actual autonomous agents at all, let alone hundreds per employee.
The Board Metrics Worth Watching
The most actionable section of Rikalo’s piece is his proposed board-level metrics. He suggests revenue per employee and agent cluster, agent density per critical role, average decision latency, human override rate, escalation load per executive, and error containment speed. These are not hypothetical. Several of them, particularly override rate and escalation load, are measurable today in any organization running agentic workflows at even modest scale.
The interesting tension is between decision velocity and decision quality. Rikalo acknowledges that fewer decisions with higher individual impact is the likely outcome: executives make fewer calls per day, but each call carries greater consequence because the agent system has already filtered, prioritized, and pre-committed resources. That is a fundamentally different management model than the one most executives were trained for.
The Governance Infrastructure Gap
What Rikalo describes as “cognitive architecture” is, in operational terms, a governance infrastructure problem. Decision rights between humans and agents. Override thresholds. Escalation logic. Bias auditing. These are the same categories showing up in enterprise procurement checklists for MCP and A2A interoperability protocols, where agent governance is shifting from a technical nice-to-have to a vendor evaluation criterion.
The pattern emerging across enterprise agentic deployment is consistent: the technology scales faster than the organizational structure around it. Rikalo’s contribution is putting numbers on the human side of that gap: 60 briefs to review, three capital reallocation requests by mid-morning, a multimillion-dollar inventory decision by afternoon. Whether those specific numbers arrive by 2030 or 2035, the structural mismatch between agent throughput and human oversight capacity is already visible in organizations running production agent workflows today.
The companies that solve cognitive architecture first will be the ones that actually reach 200 agents per employee. The rest will reach 200 agents per employee on paper while a handful of overloaded managers rubber-stamp recommendations they no longer have time to evaluate.