SDLC explorer

Where each pattern earns its place in delivery

A pattern is only useful somewhere. This map puts the engineering patterns on the software lifecycle: nine stages, drawn as a ring because real delivery loops back. Pick a stage to see which patterns help, why, who decides, what to watch for and how to tell it worked. Every relevance level is editorial and explained in words; there are no invented percentages.

9 stages · 3 patterns · 16 relevance cells, each with a rationale

Start with a pattern

Circle size is the number of patterns relevant to the stage. The ring is a reminder that delivery loops; nothing here is a sequence to follow once.

Stage 1 of 9

Discover and define

Finding out what problem is worth solving and whether an agent belongs in the answer.

Common tasks

Research synthesis across interviews, tickets and logs
Competitive and market scans
Problem statements and opportunity sizing
Deciding whether an agent is needed at all

Human roles and approval points

Product lead frames the problem and owns the decision
Researcher checks synthesis against the raw material
Domain expert validates what the agent claims to have found

Security concerns

Untrusted web and document content enters the context
Confidential research material must not leave the system
Fabricated citations

Evaluation methods

Citation checks against the source material
Expert review of a sample of syntheses
Comparison with a human-written synthesis

Patterns that help here

Orchestrator and workerscentral to this stage
Research across several sources is the canonical use: Anthropic’s own research system is this pattern.
Example A discovery brief that pulls from interviews, support tickets and competitor docs, each by a worker, with attribution.
Single-agent tool loopuseful here
A research loop over search and read tools is the simplest agentic research setup.
Example Summarising what five support tickets have in common, with every claim traced to a ticket.

How the levels are set

3, central: the stage's core work is this pattern, and a primary source describes it in that role. 2, strong: the pattern fits a recurring task in the stage; the fit is argued on the pattern page. 1, useful: it can help, with a simpler alternative often enough. Patterns can sit in several stages, and a stage can have none from a family. The matrix is reviewed when a pattern is added or re-scored; changes are listed on What's new.

The stages

1. Discover and define

Finding out what problem is worth solving and whether an agent belongs in the answer.

Single-agent tool loop (1/3): A research loop over search and read tools is the simplest agentic research setup.
Orchestrator and workers (3/3): Research across several sources is the canonical use: Anthropic’s own research system is this pattern.

2. Requirements and planning

Turning intent into scoped, testable work: stories, acceptance criteria, plans.

Stateful graph with branching and human approval (2/3): The plan node and the reviewer node are a reviewed plan before any work starts: the requirement is restated, scoped and approved by a person, and the approval is recorded in state.

3. Architecture and system design

Choosing the structure: what is deterministic, what a model decides, where humans step in.

Orchestrator and workers (2/3): Option analysis benefits from independent workers investigating each option without anchoring on the others.
Stateful graph with branching and human approval (2/3): Deciding where the explicit structure belongs is an architecture decision; this pattern is the shape to compare the loop against.

4. Development and implementation

Writing, changing and reviewing code with agents in the loop.

Single-agent tool loop (3/3): Coding agents are this loop: read, run, edit, re-run. It is the unit every other development pattern is built from.
Orchestrator and workers (2/3): Large changes split by module; coding assistants use subagents for verification so the main context stays clean.
Stateful graph with branching and human approval (3/3): The change-request graph is the shape of an agentic development workflow with review built in.

5. Testing and validation

Finding out whether the system, and the agent inside it, does what was asked.

Single-agent tool loop (2/3): A loop that can run the suite can also write a test, run it, and iterate until it exercises the case.
Orchestrator and workers (2/3): Test generation per module or per requirement parallelises cleanly and the results are naturally structured.

6. Security and compliance

Keeping the agent, its tools and its data inside the lines, and proving it.

Stateful graph with branching and human approval (2/3): Approval on a defined path with a recorded decision is an audit trail by construction.

7. Deployment and release

Shipping safely: staged rollouts, approvals, rollbacks.

Stateful graph with branching and human approval (3/3): Release workflows need exactly this: conditional paths, a person who approves, durable waits, idempotent actions.

8. Operations and monitoring

Watching the system and the agent run, and knowing what to do when they misbehave.

Single-agent tool loop (2/3): Incident triage is a loop over read-only tools: logs, metrics, recent deploys. The budget and the read-only tool set keep it safe.
Orchestrator and workers (1/3): Incident investigation across logs, metrics and deploy history can fan out, though a single read-only loop is often enough.
Stateful graph with branching and human approval (2/3): Runbook automation with approval before each destructive step and resume after an interruption.

9. Maintenance and evolution

Keeping the system current as models, tools, dependencies and requirements change.

Single-agent tool loop (2/3): Dependency and API upgrades are a loop over the same tools: bump the version, run the tests, read the failures, fix, re-run. The test suite is the stop condition.

Discover and define

How the levels are set

The Agents & Humans Briefing