The Predictability Loop

A Control Layer for Enterprise Software Delivery in the Age of AI
1. Auflage 2027
ISBN: 979-8-8688-3418-9
Verlag: APRESS L.P.

Buch, Englisch, Format (B × H): 155 mm x 235 mm

Keep software delivery predictable in the age of AI with a practical operating model built for engineering leaders. AI coding assistants have made development faster, but speed alone does not finish work; it shifts the constraint downstream into review, integration, and testing, where rising output quietly hides commitments becoming unsafe.

The Predictability Loop is the control layer for that problem: a closed practice you run to observe the signals the delivery system actually emits, interpret what changed, choose the least-harmful intervention, validate whether the system improved, and communicate the remaining uncertainty honestly to the people making commitments. It runs on established flow measurement, but its real subject is the layer above the instruments: which signal to act on, which fix to try first, how to prove it worked, and when to distrust the number. That is the difference between a dashboard and an operating model.

Around that loop, the book builds the enterprise layer the AI era demands. It shows you how to diagnose where AI has shifted the constraint, build an early-warning view of delivery risk, set service-level expectations based on real history, and defend a forecast to the board in language a non-statistician can follow. It tackles the questions that decide whether AI adoption pays off: governing AI use across many teams, managing the verification burden it creates, telling real productivity gains from theatre, and reasoning about the economics and cost of delay at portfolio scale. The methods come from running AI-accelerated delivery in practice, including guarded use of generative AI under human review - built, not theorised.

You Will:

· Apply the Predictability Loop - observe, interpret, intervene, validate, communicate to keep delivery predictable as AI accelerates development.

· Use the four delivery signals - work in progress, throughput, cycle time and work item age with service-level expectations and probabilistic forecasting to make delivery decisions you can defend.

· Identify and address AI-shifted bottlenecks across code review, integration, testing, and deployment pipelines before they cost you a commitment.

· Distinguish genuine productivity gains from misleading metrics and develop reliable approaches for measuring the impact of AI-assisted development.

· Run the four gates - data quality, stability, forecast-readiness and distrust, so a number has to earn its way into a promise.

This book is for: VPs of engineering, engineering directors, CTOs and heads of platform and delivery who own delivery predictability. Delivery leads and engineering-effectiveness practitioners who operate the instruments will find every artefact in fillable form in the appendices.

Zielgruppe

Professional/practitioner

Autoren/Hrsg.

Lobus, Roman

Fachgebiete

• Mathematik | Informatik EDV | Informatik Informatik Künstliche Intelligenz Maschinelles Lernen
• Mathematik | Informatik EDV | Informatik Programmierung | Softwareentwicklung Software Engineering

Inhaltsverzeichnis

Part 1 - Why AI needs a control layer.- Chapter 1 - How AI moved the constraint. Establishes the spine: AI accelerates code, the constraint moves into the flow, and output dashboards cannot see it. Introduces the loop as the control layer.- Chapter 2 - Signals you can trust. Sets up the instruments and gets workflow data clean enough to act on - the gauge layer the rest of the book steers with.- Part 2 - Reading an AI-accelerated system.- Chapter 3 - Where did the constraint go? Diagnose the bottleneck when the shape of work changes, including AI-shifted patterns: review queue saturation, integration backlog, and approval friction.- Chapter 4 - What’s at risk right now? Build an early-warning view from work-item ageing, breach probability, and tiered alerts.- Chapter 5 - Are we keeping our promises? Set service-level expectations from real history, track compliance, and predict in-flight breaches.- Part 3 - Acting on the system.- Chapter 6 - Choosing the least harmful intervention. Apply the intervention hierarchy - WIP before capacity before process - and map each pattern, including AI-shifted ones, to the fix that addresses it.- Part 4 - Rigour you can defend to the board.- Chapter 7 - Are we stable enough to commit? Use process behaviour charts to judge whether the system can support a commitment.- Chapter 8 - Did the system actually improve? Prove whether a change helped, using baselines rather than gut feel.- Chapter 9 - Commitments without false certainty. Give realistic dates with explicit confidence using Monte Carlo on the team’s own throughput.- Part 5 - Running the control layer at enterprise scale.- Chapter 10 - When to distrust your analytics, and your AI. Decide when a number is unsafe to present and use AI to support interpretation without inviting hallucinated diagnosis.- Chapter 11 - Governing AI across the engineering organisation. Set what stays under human review, manage the verification burden AI creates, and write policy that holds across many teams.- Chapter 12 - The economics of AI-accelerated delivery. Reason about the cost of the moved constraint, real return versus productivity theatre, and cost of delay at portfolio scale.- Chapter 13 - Running the loop across the portfolio. Read portfolio flow health and cross-team dependencies, and aggregate without averaging to a meaningless number.- Chapter 14 - Keeping the system honest. Detect and prevent metric gaming and sustain the practice past initial enthusiasm.- Appendices.- A - Artefact toolkit. The worksheets and dashboards each chapter produces.- B - Statistical methods reference. The mathematics kept out of the prose.- C - Guarded-AI prompt templates and companion repository.

Über Autor(innen)

Roman Lobus is VP of Agile Product and Portfolio Management at Singapore Airlines, where he redesigned the end-to-end software delivery lifecycle across more than 120 teams. He replaced velocity-based reporting with a delivery-signal system — queue monitoring, work-in-progress aging, bottleneck detection, and structured intervention reviews — now extended with guarded GenAI analysis under human review. His work focuses on how engineering organisations stay predictable as AI accelerates development and shifts the constraint downstream into review, integration, and testing.

He has spent more than 20 years in enterprise technology, leading delivery and transformation across telecommunications, finance, oil and gas, pharmaceuticals, logistics, and government. That cross-industry experience, much of it in regulated environments, informs the operating model at the centre of this book. He writes and speaks on flow metrics, delivery predictability, and AI-accelerated software delivery. He is the author of The Art of Creating Self-Organizing Teams: Agile Team Coaching from a Journeyman to an Expert. More at lobus.works and linkedin.com/in/lobus.