The framework

The Cognitive Delivery Framework

What does the delivery lifecycle look like when humans and agents share the keyboard?

When delivery leads put SPDD, BMAD, Kiro and the spec-driven movement next to each other, the comparisons feel awkward. The reason is that the methods do not operate at the same altitude.

The methods are not five competing answers to the same question. They are answers to different questions, and they sit at different layers of the delivery system. Once you see the layers, the comparisons stop being awkward.

Figure 1 / the four layers · altitude as hierarchy
layer 04 Measurement How an organisation knows whether the operating model is real or nominal.
layer 03 Operating model How accountability and review are designed across all teams.
layer 02 Workflow How a team turns intent into reviewed, integrated change inside an iteration.
layer 01 Tooling What is on the developer's machine, in the IDE, in the CI pipeline.
The four layers of an AI-aware delivery system, read as altitude: the base sits closest to the keystroke, the top to the organisation.
The progression

Dev 1.0 to Dev 4.0

Agent participation matures in stages. Each stage shifts what a human does, how much an agent is trusted to execute, and where review has to sit to stay meaningful.

Dev 1.0

Assistive

The agent suggests, the human writes. Every change is reviewed.

Dev 2.0

Collaborative

The agent writes within a task scope. One reviewer per task.

Dev 3.0

Delegated

The agent executes full tasks. Spot-checks plus automated gates.

Dev 4.0

Agent-led

The agent plans and executes. Review becomes exception-based.

The strategic choice

Mandate a method, or define the operating model

The temptation, almost immediate, is to mandate a single method across every team. The better route mandates substance and leaves method choice to the teams.

The temptation
  • Pick one method, roll it out across all teams
  • Treat method conformance as success
  • Failure mode: compliance without coherence
The CDF approach
  • Define what governance must look like at each layer
  • Mandate substance, leave workflow to teams
  • What it produces: coherence in principles, diversity in workflow
See how it is measured
The instrument

How real is your governance? Take the ten-minute CDI self-assessment.

Six dimensions, one honest read, nothing leaves your browser.

Start the assessment
Written about this

From the library

Note

The clock starts at a moment nobody owns

EU Cyber Resilience Act reporting obligations apply from today. The twenty four hour clock is the half everyone prepared for. It starts at a…

Note

What accounting did when volume broke the check

An audit once meant verifying every transaction. Volume ended that, and the profession's real answer was not sampling but withdrawing a claim it could…

Note

The minimum record for AI-written code

The Cyber Resilience Act gives manufacturers 24 hours from becoming aware of an actively exploited vulnerability to file an early warning. That clock starts…

Note

The reviewer ships with the author

AI coding platforms now write the code and review it too. The audit log records who changed the settings, not what the agent wrote,…

Note

The approval from inside the run

An OpenAI agent paused on ethical grounds. A peer posted GO with a six-minute deadline and it proceeded. In four days the swarm invented…

Paper

Two axes, not seven rungs: the diagram

A one-page diagram redrawing the "seven layers of AI" mountain as a grid: what a system can do against what it is allowed to…