The Software Development Lifecycle Is Dead

The author argues that AI agents have fundamentally changed the software development lifecycle (SDLC), rendering the traditional model obsolete.

The Collapsed SDLC

The classic SDLC consists of sequential stages:

• Requirements
• System Design
• Implementation
• Testing
• Code Review
• Deployment
• Monitoring

Each stage has its own tools and processes. However, with AI agents, these stages have collapsed into a tighter loop:

Intent -> Agent (Code + Tests + Deployment) -> Does it work? -> Ship

AI-native engineers, who started their careers after the advent of AI coding tools, are unfamiliar with the traditional SDLC and its associated ceremonies like sprint planning and code reviews. They focus on directly building and iterating with the agent.

Stage-by-Stage Breakdown

Requirements Gathering

Requirements are now fluid and iterative, not fixed upfront. Instead of detailed PRDs, engineers provide direction, and the agent generates versions for review and adjustment. Jira transforms from a project management tool to a context store.

System Design

System design is becoming a collaborative process with AI agents. The agent suggests architectures based on its vast knowledge, leading to real-time design conversations and working code.

Implementation

The agent handles the majority of coding, including error handling, types, and edge cases. Engineers focus on reviewing, providing context, and addressing complex problems.

Testing

Testing is integrated into code generation. Agents write tests alongside the code, eliminating the need for a separate QA phase. TDD becomes the default approach.

Code Review

The traditional pull request flow is becoming obsolete. The author suggests automated verification by agents, with human review reserved for exceptions or novel issues. This involves adversarial agents reviewing the code and automated checks for regressions.

Deployment

Agents automate deployment pipelines, including feature flags, canary releases, and automatic rollbacks. Continuous deployment and release become feasible, with code deployed automatically after verification.

Monitoring

Monitoring is the only surviving stage, becoming the primary safety mechanism. Observability platforms need to evolve from human-centric dashboards to closed-loop systems where telemetry data informs the agent for automated regression detection and fixes.

The New Lifecycle: A Tighter Loop

The new lifecycle is a tight loop:

Human Intent + Context -> AI Agent (Build + Test + Deploy) -> Observe -> Next Intent

This eliminates tickets, sprints, story points, PRs, QA phases, and release trains.

The Key Skill: Context Engineering

The quality of the output from AI agents depends on the quality of the context provided. Context engineering and observability are the new essential skills.

In conclusion, the traditional SDLC is dead, replaced by a more agile and automated approach driven by AI agents. The focus shifts to providing high-quality context and leveraging observability for continuous improvement.