Two losing paths.
Modern software delivery sits between two pressures. Adopt AI without redesigning the process around it, and you burn people out, ship less stable work, and can’t defend the gains you thought you had. Keep your traditional SDLC discipline and skip AI, and you protect the craft while competitors working at a different pace pull away.
Both are losing paths. Cadence is the third one, and most orgs aren’t taking it.
The point of this handbook
AI alone won’t do it. Process alone won’t either. Modern delivery needs both.
This handbook is the operating model, not the manifesto. It spells out the lanes, the events, the roles, and the signals, and what each one produces.
AI without process: the human cost.
AI tools dropped into unchanged workflows fail in predictable ways, and the people inside the system pay for it. The evidence is already in.
Felt speed isn’t real speed
Experienced engineers using AI coding tools are often slower than their AI-free baseline, even as they feel faster. METR’s 2025 randomized controlled trial measured a 19% slowdown for senior open-source developers — who believed they’d sped up by about the same margin.
That gap is expensive. It fuels productivity claims no one can back up, hides what’s actually slowing the team down, and leaves people anxious about a pace the numbers never confirm.
The quality cost downstream
Even when AI speeds up output, that output arrives less stable. Google’s 2024 DORA research found higher AI usage tracks with throughput gains and stability losses. More shipped, more broken. The tax doesn’t disappear — pods pay it later in incident response, hotfixes, and rework, well past the dashboard that showed the win.
What it does to the people
Run AI without redesigning the process and the pattern repeats. People work longer hours fixing AI-generated rework. Role boundaries blur. Engineers watch their craft get redefined faster than the org can support, and start to wonder where they fit. The Stack Overflow 2024 Developer Survey shows a wide gap between how often developers use AI tools and how much they trust the output. The daily experience and the executive narrative aren’t the same story.
Burnout is the predictable result of asking people to absorb a transformation with no new way to work through it. Most of it is preventable. The problem isn’t AI. It’s AI shipped into a system never redesigned to receive it.
Process without AI: the resilience cost.
The opposite mistake — doubling down on traditional process while refusing AI — loses too. Just on a slower timeline.
What process alone earns
Scrum, lean, and DevOps deliver real but bounded gains. They plateau at the velocity humans alone can sustain. AI tools alone follow a similar curve from the other direction: modest, useful gains that hit a ceiling without a surrounding system to compound them. Either approach on its own hits a wall.
What process alone can’t do
The harder problem is the assumption these approaches were built on. Scrum, lean, and DevOps treat humans as the rate-limiting step. They assume the bottleneck is human cognition and human coordination. That held for thirty years. It’s becoming wrong.
AI-native competitors aren’t iterating faster the way a fast Scrum team beats a slow one. They’re working at a different pace altogether: parallel investigations, parallel drafts, parallel evaluation, continuous shipping. Process discipline can’t match that pace no matter how mature it gets, because what changed is the pace, not the technique.
Slowly, then suddenly
For a while this looks fine. The process-mature, AI-cautious org keeps shipping reliably, quality stays good, the team stays happy. Then the gap with AI-native competitors crosses a threshold, usually somewhere between “we noticed they shipped that faster” and “their feature set has lapped ours.” The decline isn’t gradual. It compounds.
Holding to traditional process without AI isn’t the safer path. It’s a slower decline.
Why both, why now.
AI alone caps out at about the same ceiling as process alone. Combined, they compound — but only when the process is redesigned around AI capability and AI runs inside the discipline of that process.
The compounding doesn’t come from stacking two improvements. It comes from one system designed for both: specs that humans and agents both read, quality gates that catch AI failure modes before they ship, roles that are honest about what humans should still own, events that pause to sense-check what AI is producing, and maturity tracked on both the AI and the process side.
Practical, not theoretical
Most orgs haven’t gotten this right — not because the idea is hard to grasp. It usually arrives as a manifesto: “Combine AI with strong process.” Teams read it, go back to their desks, and have no idea what to actually do on Monday.
Cadence is the operating model underneath the manifesto. It names the lanes work flows through, the pods, the four scaled events and what each produces, the roles and what each decides, the signals that drive prioritization, and the maturity dimensions the org tracks. None of that is a principle. It’s a practice you can run.
What this handbook does.
The chapters that follow describe Cadence in working detail. Read in order, they build the framework piece by piece. Read out of order, each chapter still stands on its own.
- 02Three lanesHow the org is grouped: Align, Deliver, Enable.
- 03PodsStable, small, AI-powered units of delivery.
- 04RolesThe minimum roles, and the SDLC maturity they own.
- 05SignalsThe data feeding every decision about what to work on.
- 06ArtifactsThe named work products that flow through the pipeline.
- 07Scaled eventsFour moments where work, learning, and direction reconnect.
- 08AI Guild & MaturityHow AI competency moves and matures across the org.
One more thing before the framework starts. Everything here is opinionated and concrete. Where there’s a tradeoff, this handbook makes a call instead of describing both options. If a section doesn’t match how your org works today, treat that as the gap to close, not proof the framework is wrong.