Integration is the unglamorous centre of enterprise software. Nobody pitches it in sales meetings. Nobody puts it on the homepage. But it's 40-60% of every enterprise project we've ever delivered, and it's the thing that kills projects when it goes wrong. These six posts track what we've learned about integration across eight years, from the original "integration tax" observation through to AI-era architecture patterns.
The Tax Nobody Budgets For
- Every enterprise project we've ever run has underestimated integration costs. Connecting systems is the silent budget-killer in bespoke builds. The data doesn't match. The APIs aren't documented. The legacy system does something undocumented that three departments rely on. This post named the problem.
Chapter 1 - Aug 2017. Naming the problem.
The Integration Tax
Take·5 min read
Patterns That Work
Four years later, we'd developed practical patterns for making enterprise systems talk to each other. APIs, middleware, event-driven architecture, and when to use each. Not theoretical architecture diagrams. Patterns drawn from connecting systems that were never designed to be connected.
Chapter 2 - Mar 2021. The pattern book.
Integration Architecture for Growing Organisations
Guide·6 min read
The Complexity Ceiling
Every system you add makes the next integration harder. It's not linear. It's exponential. The compounding cost of connections that nobody budgets for, and the architectural decisions that either manage that complexity or surrender to it.
Chapter 3 - Mar 2022. The exponential problem.
The Integration Complexity Ceiling
Article·6 min read
AI Makes It Worse
When we started building AI systems, we assumed the integration patterns from traditional enterprise would transfer. They didn't. AI needs unstructured data, real-time feedback loops, confidence thresholds that affect routing, and fallback paths for uncertainty. Every existing integration point got more complex when AI entered the chain.
Chapter 4 - Oct 2023. The complexity multiplier.
When Integrations Meet AI: The Complexity Multiplier
Article·9 min read
The AI Integration Tax
Two years into AI delivery, the integration tax has a new dimension. AI adds latency, non-determinism, and model versioning to every integration point. The hidden cost of connecting AI to legacy systems, data pipelines, and existing workflows is consistently the most underestimated line item in AI project budgets.
Chapter 5 - Jan 2025. The new tax.
The AI Integration Tax
Article·9 min read
Making It Manageable
Eight years of integration pain condensed into a practical approach. Event-driven architecture as the foundation. API gateway patterns that handle both traditional and AI integrations. The integration layer as part of the AI foundation, not an afterthought. There's no silver bullet. But there are patterns that compound instead of collapse.
Chapter 6 - Nov 2025. The current answer.
Solving the Integration Problem (Or At Least Making It Manageable)
Guide·13 min read
Why This Series Exists
We wrote these posts because integration is the most consistently underestimated cost in enterprise software, and AI has made it worse. If you're planning a project budget, read Chapter 1 first. If you're mid-project and over budget, there's a decent chance the answer is in Chapter 3. If you're building AI on top of legacy systems, Chapters 4 and 5 will save you months.
I've been saying "the integration will take longer than you think" for eight years and I've never been wrong. AI didn't simplify the integration problem - it promoted it from an engineering challenge to an architectural one.
John Li
Chief Technology Officer