# How Forward Deployed Engineering is done at Factory — Eno Reyes

Source: [AI Engineer](https://www.youtube.com/watch?v=wpOA-UXynoM)  
Feed7 permalink: https://feed7.dev/p/how-forward-deployed-engineering-is-done-at-factory-eno-reyes-0zgscmd  
Published: 2026-07-28T22:00:06.000Z  
Trust: Source Linked (source_linked)

## Why Included

Factory’s deployed-engineering model treats agent adoption as workflow design: instrument the path from signal to deploy, build validators, and tie autonomy to measurable business outcomes.

## Source Summary

Factory frames software delivery as a loop from incoming signals through planning, code changes, validation, and deployment. Its own codebase has roughly **15–20% autonomy**, while a constrained internal legal workflow is described as effectively fully autonomous.

## Practical Implication

Builders should invest in agent readiness before longer runs: bounded tasks, explicit completion criteria, dense verification signals, observable workflows, and an ROI link to business goals. The engineer’s role shifts toward maintaining the system that produces software.

## Agent-Ready Context

Factory frames software delivery as a loop from incoming signals through planning, code changes, validation, and deployment. Its own codebase has roughly **15–20% autonomy**, while a constrained internal legal workflow is described as effectively fully autonomous.

Builders should invest in agent readiness before longer runs: bounded tasks, explicit completion criteria, dense verification signals, observable workflows, and an ROI link to business goals. The engineer’s role shifts toward maintaining the system that produces software.

Autonomy depends on what can be verified. Factory says visual terminal defects such as flicker still prevent a closed loop in its core harness, so **100% autonomy** appears more attainable first in constrained internal tools than in broad product surfaces.

## Connected Context

Feed7 judgment across 262 accumulated Signals:

This adds an operational constraint to software-factory claims: autonomy rises only where completion can be verified densely and reliably. Factory’s contrast between limited autonomy in its broad codebase and effective autonomy in a constrained legal workflow supports bounded tasks, observable loops, and business-linked validators, while visual defects that remain hard to verify mark a concrete barrier to closing the loop.

- [From Signal to PR: Anatomy of a Self-Improving Agent — Jason Lopatecki, Arize](https://feed7.dev/p/from-signal-to-pr-anatomy-of-a-self-improving-agent-jason-lopatecki-ariz-1i6f0ba) — Both turn operational signals into agent work, with Factory’s model emphasizing validators and business outcomes around the signal-to-change loop.
- [What Does Done Even Mean? Agents and Paperclip's Liveness Model - Dotta, Paperclip](https://feed7.dev/p/what-does-done-even-mean-agents-and-paperclip-s-liveness-model-dotta-pap-0lx8wfc) — Factory’s demand for explicit completion criteria reinforces the view that done must be established through evidence and verification rather than declared by the agent.
- [Loop Engineering from First Principles — Kyle Mistele, HumanLayer](https://feed7.dev/p/loop-engineering-from-first-principles-kyle-mistele-humanlayer-1nuq7gf) — The recommendation for bounded, verifiable work supports incremental loops and explains why broad autonomous runs remain limited.
- [AI Agents for Performance: Ship Faster, Pay Less — Rajat Shah, Netflix](https://feed7.dev/p/ai-agents-for-performance-ship-faster-pay-less-rajat-shah-netflix-1c2tvq2) — Netflix’s fixed orchestration, canaries, and engineer ownership are a concrete example of the dense validation signals Factory says autonomy requires.

## Context Map

- Layer: agent
- Domains: coding
- Topics: harness-engineering, agent-reliability, observability

## Uncertainty

- Autonomy depends on what can be verified. Factory says visual terminal defects such as flicker still prevent a closed loop in its core harness, so **100% autonomy** appears more attainable first in constrained internal tools than in broad product surfaces.

## Agent Instruction

Use this item as source-backed context. Do not invent claims beyond the linked source. If this item conflicts with another source, call out the conflict.