HADLC Engineering · 2026
Agentic Development Flow

Aligning humans and agents, one phase at a time.

The HADLC way — Human-Agent Development Life Cycle.
Human-Centric → Agent-Oriented → Breakdown → Construction → Operations
01
The Problem

A raw requirement handed straight to an agent goes sideways.

You get plausible-but-wrong output, scope creep in the gaps, and specs that drift from the code they were meant to describe.

Humans align on what before agents formalize how.
HADLC extends AI-DLC by making the human / agent boundary explicit.
02

Five phases, one explicit boundary.

Work is spec-bounded, not time-bounded. Unit ≈ Epic, Task ≈ Story, Bolt ≈ Sprint — the rhythm, not the unit of work.

01 Human-Centric Inception Agree on the problem Human
02 Agent-Oriented Inception Formalize into OpenSpec AI
03 Breakdown Decompose into Tasks AI
04 Construction Propose, validate, repeat AI + Human
05 Operations Not yet defined Deferred
The work ladder
ProposalPhase 1
→
UnitsPhase 2
→
TasksPhase 3
→
CodePhase 4
03
01
Phase 01

Human-Centric Inception

PO, PM, RD, UX, and Architect agree on the real problem — in / out of scope — before any tooling.
Lead: Human · no AI in the room
Activities
01Understand the real problem behind the request
02Clarify scope — what is in, what is explicitly out
03Align on how requirements will be fulfilled
Output
Refined initial spec
Tools
brainstorming idea-refine
04
02
Phase 02

Agent-Oriented Inception

The agent joins — with human-validated context in hand — and formalizes intent into OpenSpec.
Lead: AI · human validates
Activities
01Formalize the refined spec into OpenSpec format
02Identify Categories — api, ux, auth, data
03Human validates intent; agent revises. Iterate.
Output
Units — Categories with spec.md
Tools
OpenSpec workflow spec-driven-development extract-behavior
05
03
Phase 03

Breakdown

Units become Tasks. Made an explicit phase because decomposition is where scope creep hides.
Lead: AI · validate before code
Activities
01Define the API contract first
02Define the UX — screens, components, flows
03Generate a clickable prototype + per-Task specs
Output
Tasks — API, UX & technical specs
Tools
hadlc-breakdown api-designer hadlc-ux-design openspec-prototype
06
04
Phase 04

Construction

The core loop: AI proposes, human validates, repeat — until each Task is done, not until the clock runs out.
Lead: AI + Human
Activities
01Domain & logical design, validated at each step
02Generate code with tests
03Five-axis review gate before merge
Output
Working, tested, reviewed code
Tools
writing-plans executing-plans test-driven-development code-review vortex-inspect
07
The Refinement Loop

Fix the spec, not the symptom.

When downstream work reveals an upstream problem, go back to the source. Local patches accumulate as tech debt.

Discovered at
Issue
Go back to
Construction
Spec contradiction, infeasible design
→Breakdown · Tasks
Breakdown
Missing requirement, scope gap
→Agent-Oriented · Units
Agent-Oriented
Ambiguous intent, conflicting goals
→Human-Centric · Proposal
08
Take This With You

Phase → tool cheat sheet.

Phase Lead Output Tools
01Human-Centric
 Human Refined spec
brainstormingidea-refine
02Agent-Oriented
 AI Units
OpenSpec workflowspec-driven-developmentextract-behavior
03Breakdown
 AI Tasks
hadlc-breakdown
04Construction
 AI + Human Tested code
writing-plansexecuting-planstest-driven-developmentvortex-inspect
—Review gate
 Human Reviewed code
code-reviewrequesting-code-review
↻Refinement loop
 Human Fixed specs
OpenSpec workflow/openspec:archive
09
Getting Started

Three steps to run your first flow.

01
Set up context
Write a CLAUDE.md — tech stack, conventions, and domain terms the agent should assume.
02
Pick a pilot task
Well-defined, medium complexity, low risk if it takes longer than expected.
03
Iterate & capture
Turn what works into reusable skills. Refine specs when issues surface.
Specs are the source of truth. Code follows specs.
Questions & discussion
10