Engineering with AI | StackLiberate

The Bottleneck Moved: When Writing Code Became the Cheap Part

AI made implementation dramatically cheaper for us. The hard part moved elsewhere: deciding what should exist, preserving intent, verifying architecture, testing the real product, and knowing when a change is actually finished.

A Roman engineering works where abundant automated construction flows toward a smaller series of inspection, review, and acceptance gates.

A Roman command structure showing human direction flowing into specialized agents that investigate, implement, test, and review software before returning results for acceptance.

By Jossue Daniel7 min read

Humans Steer, Agents Execute: Our Engineering Loop With AI

AI participates deeply in how we build StackLiberate, from repository discovery and implementation to testing and review. But scope, architecture, acceptance, and responsibility remain human decisions.

A Roman archive and engineering command center showing durable project knowledge flowing through successive agents across a long-running software project.

By Jossue Daniel7 min read

Context Is Infrastructure: Keeping Agents Aligned on Long-Running Software

Long-running software cannot depend on an AI agent remembering yesterday’s conversation. We treat repository knowledge, plans, constraints, and handoffs as infrastructure that lets fresh agents continue engineering work without rediscovering the system from scratch.

A Roman marble engineering tribunal examining a successful test result against architectural plans and structural boundaries.

By Jossue Daniel7 min read

Green Tests Aren’t Enough: How We Review AI-Generated Code

AI can produce code that compiles, passes tests, and still moves a system in the wrong architectural direction. Our review process treats green tests as evidence—not the final verdict.

Monumental Roman marble architecture representing a structured engineering plan, with ordered pillars leading toward implementation.

By Jossue Daniel7 min read

Plan First: Why Our Coding Agents Don’t Start With Code

Before an AI agent changes StackLiberate, we usually ask it to understand the system, identify the contracts involved, map the blast radius, and propose a plan. The code comes after the reasoning.