Gabriel Martinez

Gabriel “Gabo” Martinez worked as an engineering manager at G2i in 2026. His central concern is how teams retain judgment and ownership when AI makes code easier to generate than to understand. His critique of AI slop focuses on work that looks finished before the thinking is complete: ambiguity gets resolved through guesses, and consequential choices pass unnoticed.

Generation moves the bottleneck to review

Martinez treats software review as a constraint on engineering productivity. A thousand-line pull request or a twenty-page generated document can be quick to produce while leaving someone else with the expensive task of understanding it and deciding whether it makes sense. Pressure to merge faster compounds that burden, allowing generation to outrun the team’s ability to evaluate the work.

A convincing prototype creates a related problem: stakeholders can mistake visible progress for completion. Martinez argues that code is a liability because teams remain responsible for what they ship. Adding agents to a messy codebase can make the mess grow faster; producing more code does not resolve the decisions or maintenance obligations behind it. His position supports AI use while insisting that engineering judgment keep pace with its output.

Keeping engineering judgment in the workflow

• Small reviewable changes: Divide work into pieces people can understand and evaluate. Clear ownership keeps responsibility attached to the software after generation.

• Diagrams as a review tool: Make the system’s shape visible so reviewers can examine design decisions alongside the implementation. Review needs a way to assess how the pieces fit together, as well as the code itself.

• Conventions that reduce decisions: Rails provides a model for shared patterns that reduce repeated choices and the cognitive overhead of understanding unfamiliar code. Conventions give teams a common structure for assessing changes.

Martinez connects these priorities to G2i’s ORC, which orchestrates specialized agents to plan, implement, audit, test, and review software. Adversarial reviewers challenge the agents’ outputs before acceptance, making review a gate in the workflow rather than leaving it as a task after generation. That mechanism gives concrete form to his central concern: faster production still needs deliberate evaluation and clear accountability.