The Coordination Consensus
Six weeks. Three institutions. One word.
February 25: HBR names the payoff.
Sangeet Paul Choudary — four-time HBR 10 Must Reads author, Thinkers50 Strategy Award winner — publishes in Harvard Business Review: “AI’s Big Payoff Is Coordination, Not Automation.”
His argument is precise: AI’s greatest economic impact will not come from automating tasks. It will come from collapsing translation costs — the friction that prevents teams, tools, and data from working together. Not intelligence. Not speed. The ability to coordinate across fragmented systems without forcing them onto common standards.
He calls it “the least understood and most significant economic factor of AI.”
Read that again. The most significant economic factor. Not automation. Not agents. Not foundation models. Coordination.
March 16: NVIDIA builds the infrastructure.
At GTC 2026, NVIDIA announces its Open Agent Development Platform. The framing in the press release: “igniting the next industrial revolution in knowledge work.”
Seventeen enterprise partners — Adobe, Salesforce, SAP, ServiceNow, Cisco, and more — adopt the platform at launch. The architecture: open source software for autonomous, self-evolving enterprise AI agents. Safety rails. Policy enforcement. Agent-to-agent orchestration.
NVIDIA is building pipes. The kind of infrastructure investment at this scale signals that the category has been decided. This is no longer a question of whether enterprise agent coordination happens. It is now a question of who controls the pipes.
March 2026: Berkeley publishes the operating model.
California Management Review runs “Governing the Agentic Enterprise: A New Operating Model for Autonomous AI at Scale.” The paper argues that governance is no longer a constraint on AI innovation — it is a prerequisite for scaling it. Agents need identity, authority limits, trusted data sources, defined execution scope, and audit trails.
Treat agents like digital employees. Give each one a defined role and accountability structure.
Three institutions. The same thesis. The same month.
What they’re all building — and what they’re not.
Here is what this consensus is actually about: coordinating agents with agents.
Agent-to-agent orchestration. Multi-agent pipelines. Handoffs between systems. Governance for autonomous processes that run without humans in the loop.
This is real. This is being funded. This is the correct problem to solve for enterprise AI at scale.
But there is a different coordination problem. Quieter. More personal. Less covered.
The knowledge worker sitting inside this enterprise — the one now responsible for briefing agents, reviewing outputs, managing decisions that can no longer be made by a single person with a clear head — that person has no coordination layer.
They are the middleware. Between AI tools that do not share context. Between agents that do not close loops. Between commitments that are tracked nowhere and decisions that live only in working memory.
The HBR paper from February named this precisely: AI’s greatest value is lowering translation costs. But the translation happening at the human level — between intent and output, between context and action, between what I asked for and what actually happened — that cost is still entirely paid by the person.
NVIDIA’s platform coordinates agents with each other. Nobody is building the layer that coordinates the human with all of it.
The gap is not closing. It’s widening.
In December 2025, HBR published a survey finding: only 6% of companies fully trust AI agents to autonomously run their core business processes. But 86% plan to increase investment in agentic AI over the next two years.
The infrastructure is being built faster than the trust model.
When autonomous agents multiply inside an organization — each with defined scope, audit trails, and orchestration logic — the person responsible for setting intent, reviewing exceptions, and making final calls becomes increasingly exposed. More systems acting. More outputs to verify. More coordination expected from the one node in the network without a governance layer of its own.
The enterprise is getting its coordination infrastructure. The knowledge worker is getting more to manage.
This is not a complaint about AI. It is a structural observation about where the industry is building and where it is not. The enterprise coordination problem is solved — or at least, it is being solved, with significant capital and institutional attention. The human coordination problem has three credentialed studies confirming it is real, and zero infrastructure being built to address it.
That is the gap. It is not getting smaller as agents proliferate. It is getting larger.
The consensus has formed. Coordination is the payoff. The pipes are being built.
The question is who they connect to.
Eliran Keren — Founder of Deeplica, building the coordination layer for humans who’d rather direct than operate.
