Conway’s Law does not stop at software. Conway’s Law scales cleanly into corporate AI adoption, and also explains a large share of why AI initiatives stall, disappoint, or quietly die.
The original insight from Melvin Conway holds: systems reflect the communication structures of the organizations that build them.
Any organization that designs a system (defined broadly) will produce a design whose structure is a copy of the organization’s communication structure. Melvin Conway
AI systems simply make your organization reflection louder, faster, and harder to ignore.
For enterprises, AI is not a neutral tool. AI arrives inside existing teams that seek better ways of working, such as:
- Finance,
- Legal,
- Operations,
- HR,
- IT,
- Legal
Each team, despite belonging to the same organization, speaks their own language. Further, each team owns different risk and seeks to optimize different outcomes. When an organization attempts to create a enterprise AI strategy but does not change how these groups coordinate, then AI faithfully reproduces that fragmentation.
This is where adoption failure begins.
Teams may use AI to build a general-purpose model, but what happens a layer or two beneath the surface:
- Business units see something impressive, but not quite useful;
- Legal worries about exposure;
- Finance worries about cost;
- Operations sees variation; and
- IT worries about maintanence
Each groups reacts from its own vantage point, yet because they do not think systemically, their combined effect slows adoption to a crawl.
Models may exist.
Trust does not.
There is a saying: “when you are a hammer, everything looks like a nail.”
Conway’s Law then begins to act more like Murphy’s Law observation that anything that can go wrong will go wrong.
Conway’s Law is just as valid, as AI mirrors the organization chart, not the workflow. Outputs feel foreign to daily work because the people and processes within the system attempt to navigate cross-functional business to deliver customer value.
Governance feels heavy because no one owns the bounded scope of expectation.
Costs feel unpredictable because usage cuts across incentives.
Adoption stalls because the system answers questions no one asks.
AI failure needs to be blamed on data quality, model choice, and user training. Each of those matters, but all are secondary to the genuine issue: misaligned communication produces misaligned intelligence.
Primary failure mode sits upstream at leadership.
The expansion of Conway’s Law for AI looks like this: An enterprise cannot adopt AI faster than it can align decision rights, language, and accountability.
Conways’ Law explains why department-specific Small Language Models (SML) may gain traction and monolithic AI programs struggle. Each SLM reflects a their social structure:
- Clear ownership,
- Bounded domains,
- Shared vocabulary,
- Predictable incentives.
The language model may speak like the team because the team trained the model on what they know not what prompt possibilities are available. The team’s bias sets limitations and innovation is a non-starter in this cognitive morass.
Cognitive limitation, also known as bias, also explains governance outcomes as AI scope mirrors how enterprises already govern systems of record. Familiar patterns lower resistance. Trust follows familiarity. Risk shrinks because blast radius shrinks.
Instead of asking, “How do we roll out AI?”, the better question asks, “What organizational change must exist for AI to work as intended?”
Design the human system first and the technology, in this case: AI, will follow.
Corporate AI adoption fails when organizations chase intelligence while ignoring structure. AI succeeds when structure and intelligence align.
This is not a tooling problem.
This is Conway’s AI Law.
Conway’s Law is not progress.