Agentic Developer HandbookUnderstand
Agentic System Readiness Assessment
What is the smallest AI/agentic architecture this problem actually needs? Bring one use case. The assessment asks a fixed order of questions, starting with the simplest possible answer — ordinary software — and only moves toward a model, tools, retrieval, memory, skills, or an agent runtime when an earlier answer makes that necessary.
It is willing to conclude "you probably do not need an LLM", "you need an LLM, but not an agent", or "a deterministic workflow is simpler than an agent runtime". Those are good outcomes.
Canonical version: assessments/agentic-system-readiness/README.md and the fillable ASSESSMENT.md template. Answers stay in this browser only.
Understand
Use case
Start from the problem, not from an AI feature list. Short answers are fine.
Decide
Simplest solution first
Before any AI capability: could ordinary software do this? Do not introduce an LLM merely because this is an AI handbook.
Decide
Capability decisions
The AI capability questions do not apply while that answer stands. Record what you are deliberately not building on the result page, or return to the simplest-solution questions if the answer changes.
Answer in order. Each yes makes a capability a candidate (CONSIDER); each no leaves it out (NOT JUSTIFIED). A justified No is worth exactly as much as a justified Yes.
Decide
Decision authority
Choosing capabilities is half the design. The other half is what the model's output is allowed to do. The model may reason, generate, recommend, and propose; the application owns validation, authorization, execution, budgets, and stopping.
No model is justified in your answers, so nothing is influenced by a model. The questions still have ordinary answers — authoritative data, side effects, authorization — and writing them down is optional here.
The concept in full: docs/model-vs-decision-authority.md
Decide
Runtime boundary
Production questions for later
Ask these now so the design leaves room for the answers. They are Milestones 9–11 (evaluation, observability, security) and are not implemented here.
Review
Result
Derived from your answers by transparent rules. Candidates to consider, not "the correct architecture" — and not universally correct for anyone else's use case.
Use case
Simplest alternative considered:
Capabilities
| Capability | Status | Why |
|---|
Likely shape
Decision authority boundary
Rejected complexity
Unresolved questions
Reflect
Worked examples and reflection
Four filled-in assessments, from "no model at all" to a bounded agent runtime. Each is one defensible reading of its scenario, not an answer key. The full documents live in the repository.