Is AI Making You an Average Engineer?

The podcast discusses the transformative impact of AI on software development, emphasizing both its potential to accelerate development velocity and the challenges it introduces in maintaining code quality and system reliability. AI tools can significantly increase output, such as boosting pull request volume, but this rapid pace risks introducing technical debt, redundant code, or regressions to average-quality practices. To counteract these risks, organizations must implement strong safeguards - such as platform engineering, automated controls, observability systems, and feature flags - and focus on faster incident recovery rather than attempting to prevent all failures. The discussion highlights that while AI can assist with tasks like testing, documentation, and code reviews, human oversight remains critical for design decisions and long-term system health.

A key theme is the need for cultural and organizational shifts to harness AI effectively. This includes moving from luck-based to systematic code quality enforcement, involving junior engineers in architectural discussions and ADRs (Architecture Decision Records), and fostering learning through rapid feedback loops rather than prolonged pre-deployment cycles. Observability plays a central role in enabling safe AI adoption by making systems legible and allowing teams to detect and learn from issues quickly. The conversation also explores future directions, such as proactive AI that prevents bugs before they occur, multidisciplinary collaboration in reliability engineering, and the importance of shared ownership of system reliability across teams rather than relying solely on specialized roles like SREs.

• What if you enforced AI-generated code quality by codifying your team's best practices into automated review rules?

  • Move: Identify 3 - 5 high-impact coding standards (e.g., comment verbosity, test coverage, error handling) and implement them as linter rules or AI-powered PR checks using tools like GitHub Actions or Sourcegraph.
  • Why Now?: AI is accelerating output, but without constraints, it regresses code to the statistical average - eroding hard-won quality. The time to lock in your edge is before AI dilutes it.
  • Expected Upside: Maintain or raise code quality while scaling velocity; reduce human review load by 30 - 50% on routine hygiene issues.

• What if you shifted all junior developer onboarding to include real production changes with AI-assisted observability guards?

  • Move: Pair each new solo dev with an AI-augmented feedback loop: require every first PR to include a Honeycomb span or OpenTelemetry attribute, validated by an AI bot before merge.
  • Why Now?: Fast feedback loops are the fastest path to skill growth, and AI lowers the cost of instrumentation. Waiting for "perfect" code delays learning and weakens system legibility.
  • Expected Upside: Accelerate developer ramp-up by 2x; increase production observability coverage by default, not retrofit.

• What if you offloaded 80% of reactive incident response to AI agents - but only after implementing proactive blast radius controls?

  • Move: Deploy feature flags with automatic rollback triggers (via LaunchDarkly or Flagsmith) and train an AI agent to detect anomalies in metrics/logs, then roll back if thresholds are breached - escalating only if unresolved.
  • Why Now?: AI increases deployment volume, which increases incident count. Relying on humans for 2 a.m. firefighting is unsustainable; now is the time to automate containment.
  • Expected Upside: Reduce incident severity by 5 - 10x; reclaim 10+ hours/month of engineering time currently spent on out-of-hours toil.

• Implement AI-assisted observability by integrating automated span and attribute generation in your codebase to detect regressions early and improve incident recovery speed.
• Define and document explicit coding standards (e.g., testing requirements, comment length) to guide both AI-generated code and onboarding, ensuring consistent quality in solo development workflows.
• Use feature flags with automated rollback triggers for new AI-generated changes to reduce blast radius and minimize out-of-hours firefighting as a solo operator.
• Offload repetitive code review tasks (e.g., linting, boilerplate validation) to AI tools while reserving human judgment for design-level decisions and architecture alignment.
• Break down development tasks into small, well-scoped units before delegating to AI agents to maintain control, improve output quality, and create learning feedback loops for iterative improvement.