Test Custom Infrastructure: Avoiding the House on Stilts

James Phoenix
James Phoenix

Summary

Custom tooling that doesn’t work creates cascading failures downstream. Like building a house on stilts, this article shows how to treat custom infrastructure (test utilities, CLIs, parsers, build scripts) as first-class code with comprehensive tests, preventing compound failures and increasing confidence in automation.

The Problem

Custom tooling (test utilities, CLIs, parsers, build scripts) often lacks tests. When these foundational tools break, everything built on top of them fails. LLMs generate code using broken infrastructure, creating a cascade of confusing failures. Teams spend hours debugging application code when the real issue is in the untested tooling layer. This is the “house on stilts” problem: unreliable foundations doom everything above them.

The Solution

Treat custom infrastructure as first-class code requiring comprehensive tests. Before writing application code that depends on custom tooling, write tests that verify the tooling works correctly. Use integration tests to validate end-to-end tool behavior. Apply the same quality standards to infrastructure code as to application code. This creates a reliable foundation that LLMs can confidently build upon.

Leanpub Book

Read The Meta-Engineer

A practical book on building autonomous AI systems with Claude Code, context engineering, verification loops, and production harnesses.

Continuously updated
Claude Code + agentic systems
View Book

Related

References

Explore it visually

Topics
Agent ReliabilityCi CdDeveloper ExperienceQuality GatesTesting

Newsletter

Become a better AI engineer

Weekly deep dives on production AI systems, context engineering, and the patterns that compound. No fluff, no tutorials. Just what works.

Join 306K+ developers. No spam. Unsubscribe anytime.


More Insights

Cover Image for Why Unrealistic /goal Targets Keep Coding Agents Working Longer

Why Unrealistic /goal Targets Keep Coding Agents Working Longer

I used an unrealistic tutor-discovery target to keep Codex working beyond a finished-sounding handover. Here is the run, the results and the reusable prompt.

James Phoenix
James Phoenix
Cover Image for Why I Threw Away My Video Harness When Opus 5.5 Arrived

Why I Threw Away My Video Harness When Opus 5.5 Arrived

I spent £700 on video analysis. A week later, Opus 5.5 outperformed my custom harness and changed how I think about constraints, consistency and model progress.

James Phoenix
James Phoenix