The Pattern Recognition Problem
In fifteen years of enterprise engineering, I've seen patterns repeat: microservices migrations, containerization, observability stacks, platform engineering. Each promised to solve coordination problems. Each created new ones.
What's different now: the bottleneck has shifted from typing code to orchestrating intent.
GenAI Is Not a Copilot
The junior framing of AI tools—" autocomplete that writes faster"—misses the structural shift. In regulated enterprises, the cost isn't development velocity. It's:
- ›Validation overhead (security reviews, compliance gates)
- ›Integration complexity (existing COBOL, vendor APIs, legacy data)
- ›Knowledge fragmentation (teams that own parts of a system they don't understand)
When a Staff Engineer spends three weeks tracing a data flow through six systems to answer a compliance audit, that's not a typing problem.
The opportunity: AI systems that model organizational context, not just generate syntax.
Three Architectural Patterns Emerging
1. Intent-Based Infrastructure
Instead of describing how to provision (Terraform, CloudFormation), describe what outcomes are needed. The system negotiates the path.
2. Contract-First Evolution
GenAI excels at schema migration, test generation, and interface mapping. This enables "continuous re-architecture"—systems that evolve without cascading breaks.
3. Documentation as Runtime
If LLMs can parse your system, they can monitor, troubleshoot, and explain it. The gap between "what's documented" and "what runs" collapses.
The Risk: Faux-Agility
Enterprises will adopt AI tooling and maintain the same committees, ticket queues, and quarterly planning cycles. They'll ship "AI-enabled" features on architectures designed for 2012.
The harder shift: changing decision velocity. Moving from "proposal deck → three reviews → six-week sprint" to "tight feedback loops with automated guardrails."
This is an organizational problem disguised as a technology problem.
What I'm Watching
- ›Event-driven reconciliation (CDC, Kafka Streams) as the backbone for AI-augmented systems
- ›Policy-as-code replacing manual review gates
- ›Teams that treat their architecture as a learning system, not a fixed blueprint
The tooling is arriving. The question is whether enterprise cultures can get out of their own way.
Built with CDC pipelines, bounded contexts, and a healthy skepticism of hype cycles.