AI-SDLC v1.0
A short position paper
Purpose
This paper proposes AI-SDLC v1.0, a software development lifecycle designed for a world where AI performs most implementation work and humans own intent, constraints, and decisions.
It does not promote a tool, framework, or methodology brand.
It describes a structural shift in how software is built.
The historical bottleneck
Traditional SDLCs evolved to address a single dominant constraint: human execution.
Historically:
- writing code was slow and expensive,
- Refactoring carries a high risk,
- coordination between people was costly,
- Late errors were hard to correct.
Processes such as Waterfall and Agile are optimised around these realities.
Their practices exist to manage execution cost and coordination risk.
What AI changed
AI collapses the implementation cost.
Today, systems can:
- generate and rewrite code quickly,
- refactor large codebases cheaply,
- produce tests and infrastructure automatically.
Execution is no longer the primary bottleneck in many environments.
The new bottleneck
When execution becomes cheap, a different constraint dominates:
- unclear intent,
- weak constraints,
- untested assumptions,
- delayed or misleading feedback,
- Poor decisions made without evidence.
Speed amplifies outcomes, both good and bad.
Building faster does not help if the direction is wrong.
The SDLC must therefore optimise for decision quality, not task throughput.
Design goals of an AI-native SDLC
An AI-native SDLC should:
- make intent explicit before execution,
- treat constraints as enforceable system rules,
- allow implementation to be replaced easily,
- enforce quality and safety automatically,
- rely on observed runtime behaviour,
- preserve clear human accountability.
These goals define AI-SDLC v1.0.
AI-SDLC v1.0 lifecycle
AI-SDLC v1.0 is a closed decision loop, not a linear delivery pipeline.
1. Intent
A human defines:
- the problem,
- the desired outcome,
- non-negotiable constraints,
- key assumptions,
- stop or change conditions.
2. Specification
The intent is expressed in a machine-readable form describing:
- behaviour and interfaces,
- data and state,
- non-functional requirements,
- acceptance scenarios,
- required observability.
3. Automated implementation
AI generates:
- code,
- tests,
- infrastructure,
- instrumentation,
- supporting artefacts.
Humans review outcomes rather than hand-crafting implementation.
4. Automated enforcement
Quality, security, performance, and cost limits are enforced automatically.
Failures block progression.
5. Deployment
The system is deployed incrementally and safely.
6. Observation
The running system produces evidence through:
- user and system behaviour,
- reliability and performance data,
- operational cost.
7. Decision
A named decision owner chooses to:
- continue,
- adjust,
- stop,
- or pivot.
The decision closes the loop.
Unit of progress
In AI-SDLC v1.0, progress is measured by decisions informed by evidence, not by the number of tasks completed.
Progress means:
- uncertainty was reduced,
- assumptions were tested,
- direction became clearer.
Accountability
AI-SDLC v1.0 does not remove human responsibility.
Humans remain accountable for:
- intent,
- constraints,
- decisions,
- outcomes.
AI executes. Humans decide.
Conclusion
AI changes the economics of software development.
When execution is cheap, decision quality becomes the dominant factor.
AI-SDLC v1.0 aligns the SDLC with this reality by treating intent, constraints, evidence, and accountability as first-class elements, and by positioning AI as the primary executor rather than the primary decision-maker.