Contents

Introduction0%

↑

00Introduction

Act IDiagnosis

Group 1The pipeline evolves in the dark 01You cannot freely experiment with agentic pipelines 02Engineering processes cannot be compared fairly 03The true cost of an accepted result cannot be determined

Group 2Compute is tied to people and laptops 04Compute capacity is fragmented across people 05Parallelism hits the local runtime ceiling

Group 3The tracker no longer sees the work 06The issue tracker no longer reflects a task’s lifecycle 07Intermediate results do not become issue artifacts 08Developers became agent managers—without management discipline 09Actual human work is not reflected in issues

Group 4The executor and harness are isolated 10The actual executor is inaccessible to the rest of the team 11The harness is inaccessible to other people and teams 12Access, security, and accountability are not formalized

Group 5Experience does not become project memory 13Skills are not a shared, managed asset 14A failed session cannot be reproduced and examined 15Agent memory is shaped by a personal set of tasks

ΣRoot cause

Act IIMultica 01Agent platform 02System entities 03Versioning the stack

Act IIIAdoption 01Digital executors in Jira 02AI review as a system 03What a shared system does not solve 04Phased rollout 05New scenarios 06Conclusion