engineering with ai

Engineering with AI

AI coding agents are part of my daily engineering workflow — not as a replacement for engineering judgment, but as an accelerator for the well-defined parts of it: codebase analysis, implementation, refactoring, tests, and documentation.

Every generated change goes through the same gate as hand-written code: review, automated tests, type checks, security checks, and architectural validation. Nothing reaches production on the strength of a prompt.

The workflow

  1. Understand

    Before any generation, the context has to be explicit: the architecture, the constraints, the business requirements, and the shape of the existing codebase. An agent given vague context produces confident, wrong code.

  2. Contextualize

    The agent receives explicit project rules: architecture constraints, coding standards, and the expected outcome. In a monorepo, this includes application-specific rules that differ between services.

  3. Generate & explore

    Agents produce implementation alternatives, first-pass code, refactoring steps, and repetitive changes. They are particularly effective for structured multi-file modifications that follow a clear pattern.

  4. Review

    Every diff is read by a human. Architecture decisions, error handling, and edge cases are checked — hallucinated APIs and plausible-but-wrong decisions are caught here.

  5. Test

    Generated changes run through the same automated suites as everything else: unit and integration tests, type checks, linting, and security checks.

  6. Validate

    Behavior, performance, backwards compatibility, and production constraints are verified before merge. A change that passes tests but degrades performance is not validated.

  7. Ship

    Merging goes through the normal Git and CI/CD processes. AI assistance never short-circuits the delivery pipeline.

Tools in daily use

Claude Code · Codex / OpenAI · Cursor · AI coding agents · MCP tools

Applied: Everabyte

On Everabyte, this workflow covers progressive monorepo refactoring across the Go, Node.js, and Next.js services, assisted test generation, technical documentation, migration preparation, API contract review, and CI/CD assistance — with every generated change reviewed, tested, and validated against architectural, security, and production requirements before merge.

Read the Everabyte case study →

AI accelerates implementation. Engineering judgment remains human.