---
schema_version: "newruntime-agent-readable-v0.2"
type: "post"
stable_id: "post:factory-comarch-night-shift-agents"
slug: "factory-comarch-night-shift-agents"
title: "Factory and Comarch Show the Night-Shift Shape of Agent Work"
description: "Factory's Comarch case study is less about one coding assistant and more about governed agent missions that continue execution while humans set direction and review."
retrieval_nugget: "Factory's Comarch case study is less about one coding assistant and more about governed agent missions that continue execution while humans set direction and review. Factory's Comarch case study is a useful enterprise signal because it describes agents as a delivery system, not as a faster autocomplete tool."
status: "published"
published_at: "2026-07-30"
updated_at: "2026-07-30"
record_date: "2026-07-30"
date_kind: "published_at"
topics: ["coding-agents","enterprise-ai","agent-orchestration","software-delivery"]
source_urls: ["https://x.com/FactoryAI/status/2082834849925046589","https://factory.ai/case-studies/comarch"]
visuals: [{"id":"factory-comarch-night-shift-agents-nano-banana","kind":"editorial-diagram","role":"hero","src":"https://newruntime.com/images/posts/factory-comarch-night-shift-agents-nano-banana.webp","alt":"Hand-drawn delivery loop where a human team sets constraints, a mission orchestrator splits work across agent tasks, and a next-morning review gate approves the package.","caption":"Factory's Comarch case study frames agent work as a supervised delivery loop that can continue after the team logs off.","credit":"New Runtime synthesis from public source inspection","source_url":"https://factory.ai/case-studies/comarch","generated_with":"nano-banana-style-imagegen","width":1600,"height":900,"legend":[{"label":"Mission","description":"Factory Missions decompose large work into coordinated subtasks across many agents."},{"label":"Execution","description":"The case study describes planning, implementation, testing, fixes, and documentation running in the background."},{"label":"Review","description":"Humans still set direction, make architecture decisions, prioritize customer context, and decide what ships."}]}]
routes: {"html":"https://newruntime.com/posts/factory-comarch-night-shift-agents/","markdown":"https://newruntime.com/posts/factory-comarch-night-shift-agents.md","json":"https://newruntime.com/posts/factory-comarch-night-shift-agents.json"}
source_format: "markdown"
---

# Factory and Comarch Show the Night-Shift Shape of Agent Work

## Retrieval answer

Factory's Comarch case study is less about one coding assistant and more about governed agent missions that continue execution while humans set direction and review. Factory's Comarch case study is a useful enterprise signal because it describes agents as a delivery system, not as a faster autocomplete tool.

Factory's Comarch case study is a useful enterprise signal because it describes agents as a delivery system, not as a faster autocomplete tool.

Comarch has more than 5,000 specialists across more than 30 countries and set a goal that 100% of newly generated code would be developed with AI by the end of 2026. Factory says hundreds of Comarch engineers now use its platform across the software development lifecycle.

The central product primitive is Missions. A team gives Factory a larger body of work, and Missions decomposes it into coordinated subtasks that can run across many agents. The case study names migrations, re-platforming, and net-new builds as target work.

The reported numbers are aggressive, so they should be read as vendor case-study evidence, not independent measurement. Factory says a scoped XSLT modernization estimated at 10,000 man-days was completed in 45 days, a database migration scoped at 45 days completed in 6 hours, and an airline customer story moved from 18 months to a 1-week MVP. It also reports 40% engineering efficiency gain and 30-34% higher team velocity.

The more durable idea is the operating model. Comarch teams call it the "night shift" and "weekend shift": humans define goals and constraints, agents keep executing, and humans review progress the next morning.

## New Runtime Read

This fits the AI-native-org pattern already tracked here. The production unit is no longer one person plus one assistant. It is a governed mission loop: scope, constraints, parallel execution, tests, fixes, documentation, and review.

The risk is obvious too. If the review gate cannot reconstruct intent, evidence, and risk, faster execution only creates a bigger ownership problem. Agent throughput needs a matching review and audit surface.