How AI is used in builds here

Where AI genuinely helps (and is used)

Scaffolding and boilerplate: configuration files, repetitive component shells, test scaffolds, migration scripts, the 40 lines of correct-but-tedious code around the 10 lines that matter. First-pass drafts: documentation, alt-text candidates, test-case lists, edge-case brainstorming, always edited by a human before shipping. Research acceleration: summarizing documentation, comparing library behaviors, drafting regex and data transforms for review. Explaining legacy code: reading an inherited codebase with an AI second pair of eyes before touching it, the rescue work runs faster for it. The honest measure: AI compresses the boring 60% of some tasks. It does not compress the 40% that is the actual engineering, and pretending otherwise is how AI-built sites end up in the rescue queue.

The eval gates, every AI-assisted artifact passes them

• Gate 1, human review of the diff. Nothing AI-generated merges unread. Every line gets read the way a senior reads a junior's pull request: does this do what it claims, what does it break, what does it quietly assume?
• Gate 2, tests before merge. Functionality proves itself in tests or manual verification on staging, the staging discipline applies double to AI-assisted changes, because the failure modes are creative.
• Gate 3, security pass. AI-generated code is checked for the classic leaks: injection-prone queries, secrets in code, over-permissive defaults, dependency surprises. The security practices apply to human and machine output alike.
• Gate 4, accuracy verification for content. AI-drafted copy gets fact-checked claim by claim, this site's published rate bands and technical statements are human-verified numbers, never model output.

What never touches an AI tool

Client credentials, database contents, customer data, unreleased business information, and anything covered by an NDA. Prompts are constructed to work without them, and where a task genuinely needs context, it uses sanitized snippets. This is the AI services drawn in the workflow itself, not just in policy.

The productivity claim, quantified honestly

AI-assisted work here runs measurably faster on scaffolding, research and documentation, and at human speed everywhere judgment lives. What that means for your quote: the estimate reflects the accelerated process where it applies, and the eval gates are priced in rather than skipped. You are not buying "AI builds your site", you are buying senior work, delivered faster on the parts where AI is genuinely leverage, with the gates that keep it correct. The AI services pages cover what gets built with AI for you; this page was about how it is used by the practice, both governed by the same rule: disclosed, gated, and never at your data's expense.