Agentic Engineering

A much nicer term for building products using AI Agents than the almost derogatory term Vibecoding

The bottleneck in AI-assisted development is not code generation. It is specification quality.

faros research findings 2026

findings
- AI is now the standard, not the experiment
- Thrashing and cognitive load are increasing
- Output is up. Not all of it sticks.
- Code changes are getting larger and more complex
- Code quality before merge is declining
- The workflow is slowing down at every stage
- Poor quality code is reaching production
recommendations
- 1. Measure your incident-to-PR ratio and your monthly code change volume against monthly incidents.
- 1. Define a review policy. Then enforce it as a gate.
- 1. Set PR size guidelines for your coding agents and enforce them.
- 1. Build quality gates into the development environment, not the review queue.
- 1. Use workflow stage times as input to a continuous retrospective process.
- 1. Watch work restarts as a signal about agent context quality.
- 1. Distinguish between human and agent review comments and act on both differently.
- 1. Build an automated investigation loop around code churn.
- 1. Repurpose engineering capacity. Do not reduce it.
- 1. Build a context engine for your engineering environment.
sources https://pages.faros.ai/hubfs/AI_Engineering_Report_2026_The_Acceleration_Whiplash_Faros.pdf