I recently came across this definition:
The Agentic Development Lifecycle (ADLC)—also known as the Agent Software Development Lifecycle—is a software engineering methodology designed for building, testing, deploying, and governing AI agent systems rather than traditional deterministic code.
That sentence made something click. A lot of people (myself included, until recently) have been casually treating ADLC and ASDLC as interchangeable. They’re not.
Once you separate them, the confusion starts to dissolve.
The Three Terms, Cleanly Separated
| Term | Full Name | What it is | Primary focus | What is being built? | Who does most of the work? |
|---|---|---|---|---|---|
| SDLC | Software Development Lifecycle | The classic methodology for building software | Deterministic systems | Traditional applications & services | Humans (tools assist) |
| ADLC | Agent / Agentic Development Lifecycle | A specialized lifecycle for creating AI agents | Probabilistic, tool-using, adaptive systems | AI agents and multi-agent systems | Mixed (humans + agents), with heavy emphasis on evaluation & governance |
| ASDLC | Agentic Software Development Lifecycle | The traditional SDLC performed with AI agents as workers | Any software, delivered agentically | Applications, services, infrastructure… or even agents | Agents execute; humans set intent, guardrails, and approve |
Why the Mix-Up Happens
Both terms contain the words “Agentic” and “Development Lifecycle,” so people collapse them. On top of that:
- Some writers use “ADLC” loosely to mean “the new way we develop software now that agents exist.”
- Others (IBM, Arthur.ai, Glean, Atlan, etc.) use ADLC more precisely: the lifecycle for building agents themselves.
- Meanwhile, “Agentic SDLC” / ASDLC has become the clearer label for “we’re using agents to ship regular software.”
The result is a muddy conversation where people argue past each other.
Why the Distinction Actually Matters
If you’re building an agent, you care about different things than if you’re using agents to build a normal product:
- Non-determinism and evaluation harnesses become first-class concerns.
- Observability, containment, and continuous behavioral monitoring matter more than classic “green tests.”
- The “done” state is fuzzier. Agents keep running and adapting after deployment.
Treating that work with a regular (or even agent-assisted) SDLC often leads to brittle agents that look good in demos and then drift or fail in production.
Conversely, if you’re just trying to ship a web app or an internal tool faster with coding agents, you don’t necessarily need the full ADLC machinery. You need a well-governed ASDLC.
Quick Mental Model
- SDLC → Building normal software the old way
- ADLC → Building agents
- ASDLC → Building anything (including agents) with agents
Once you start using the terms this way, a lot of the current discourse becomes much clearer.
Have you noticed the same mixing of ADLC and ASDLC in your circles? I’d love to hear how your team is currently labeling these ideas.
Top comments (0)