I love an axiom, something easy to remember, fast to say, and punchy enough to stick. With my older boys, I’ve often said, “If in doubt, don’t.” With my younger son, who is autistic, I say, “Stay close; stay safe.” These are short phrases with big truths, the kind that helps in the moment when time is short and the stakes are high.
That’s probably why ideas like Moore’s law, Amara’s law, and Parkinson’s law continue to resonate with technology leaders. They help us hold on to simple ideas as we make sense of tech adoption, value, and implementation at scale. They’re part of how we stay the course and keep our people anchored when the media shrill rises, the consultants’ decks get thicker, and LinkedIn pundits fill our feeds with certainty. And that brings me to Conway’s law.
Platform Choice Isn’t The Starting Point
I’ve been presenting multiple times per week to public-sector clients a workshop session entitled “Overcoming The Hurdles Of AI Adoption At Scale.” Recently, a client said, “Conway’s law plays out every single time. We want to implement systems before addressing our business … and every time we end up with the same results, in that the systems end up just as messed up as our organizations.” In other words: Start with the operating model and organizational structure, then orient platforms to the right domains.
One of the great mistakes in AI right now is the belief that the answer lies primarily in choosing the right platform, model, or vendor stack. It doesn’t. If the operating model is unclear, fragmented, or built for an earlier era of work, the AI system will inherit those flaws and reproduce them at machine speed. That’s why Conway’s law feels so relevant again: Systems don’t transcend organizations — they mirror them. And in the age of agentic AI, they amplify the worst of them: the silos, the politics, and more.
Start With Your Organization And Your People
This point sits at the heart of what we’re doing with our research into the cognitive operating model, intelligence enterprise, and skills-oriented architecture. And the core premise of this research is the AI productivity paradox: Gains dissipate inside operating models designed for human-only, task-based work. Bolting agents onto yesterday’s roles, workflows, and decision rights is technology deployment with better marketing from companies that need to maximize IPO valuation to get the capital needed to feed the AI cash furnace.
That’s also why the shift from generative AI to agentic AI matters so much. GenAI was the warm-up: Agentic AI changes the game because we move from prompts to plans. These systems now retrieve, decide, trigger, notify, and act. That shifts the conversation from output quality to governance, accountability, orchestration, and legitimacy — especially in government, where explainability, fairness, and public trust are nonnegotiable.
The Operating Model Shift Matters
If your operating model is siloed, fragmented, overloaded with handoffs, and built around a human-only conception of work, your AI estate will reflect that complexity. Agents will be selected, deployed, and governed according to those same fault lines. The result? Duplicated capabilities, fragmented context, inconsistent controls, and point solutions masquerading as transformation.
What Conway’s law explains is why the operating model shift is so central. At its core, agentic AI is a work architecture problem and an operating model shock. If agents increasingly become the default executors of routine cognitive work, then the organization must be redesigned around that reality. Roles, workflows, escalation paths, management assumptions, and accountability models all change. Otherwise, the technology will simply automate the archaeology of today’s enterprises.
The Skills And Context Matter
This is why our work encourages our clients to move away from use-case thinking and toward skills as the atomic unit of design. A use case describes a problem to solve. A skill describes a bounded cognitive capability that can be reused, governed, and composed across roles and workflows. Organize agentic portfolios around isolated use cases, and you get fragile, siloed deployments that resist scale. Organize around skills, and you create the conditions for composition, governance, and durable operating-model change: dynamic, agile, and flexible.
The other half of this is context. Capability on its own isn’t enough. Real competence depends on the surrounding semantic layer of policy, vocabulary, memory, decision traces, tacit knowledge, and organizational logic. Without a coherent way to surface and govern context, agentic systems will mirror the enterprise’s missing knowledge, fragmented policy interpretation, weak accountability, and rising costs.
Conway’s Law Matters
If I had to turn Conway’s law into a practical checklist for leaders in the age of agentic AI, it would be this:
Start with the operating model. Let the platform follow the work, the problem domains, and the outcomes the organization needs to achieve.
Build reusable organizational capabilities. Design skills, roles, workflows, and governance structures that compound across use cases.
Treat context as organizational intelligence. Make policy, knowledge, memory, and decision logic machine-readable, governable, and available at the point of work.
Design agents around the organization you want to become. Agents amplify the system they operate within, including its strengths, gaps, and accountability model.
For me, that’s the modern value of Conway’s law. In the dizzying storm of change we’re in, if we want agentic AI to create compound value, we must first redesign the operating model that surrounds it. That’s the work, that’s the hurdle, and that’s why our current research is so focused on structure, context, and the redesign of work itself. Otherwise, we aren’t building the future of work — we’re automating the past. So remember, kids: “Operating models deliver outcomes.”
