Early in my career, I thought my job was to design the best integration architecture. The cleanest message flows. The most elegant API contracts. The most resilient error handling.
I was wrong. Or rather, I was only half right.
Architecture doesn't survive contact with a bad organization. The best-designed system will be corrupted within months if the team structure, governance, and incentives around it are wrong.
The lesson
Over 14 years across government platforms, banking systems, and enterprise transformations, I've learned that the architecture and the organization are inseparable. Conway's Law isn't a joke—it's a design constraint.
If the team structure is siloed, the system will be siloed—no matter what the architecture diagram says. If ownership is unclear, quality will degrade—no matter how good the original design was. If governance is absent, consistency will erode—no matter how many standards documents exist.
What I do differently now
When I take on an enterprise integration challenge, I spend as much time on organizational design as on technical architecture:
Team topology. How are integration teams structured? Do they own platform capabilities, or do they serve project requests? Platform teams create leverage. Service-request teams create bottlenecks.
Decision rights. Who decides which integration pattern to use? Who approves API designs? Who owns production reliability? If these decisions require escalation, the organization can't move fast enough.
Incentive alignment. Are teams incentivized to build reusable services, or just to deliver their project on time? If reuse isn't rewarded, it won't happen—no matter how many "reusability guidelines" exist.
Knowledge architecture. How does institutional knowledge persist when people leave? ADRs, runbooks, design libraries, and governance automation are not documentation overhead—they're organizational memory.
The executive integration leader's actual mandate
Design the architecture. Build the platform. And shape the organization so that both survive and improve after you step away. That's the real job.