AI Workflows & "Shape & Ship"

Welcome to the core of the AI Workflow Curriculum. Here, we discuss the mental models, methodologies, and orchestration patterns required to build applications successfully in pair-programming environments with AI agents.

Incurable curiosity drives our stack, but structure is what allows us to deliver high-quality code. At Incrementic, we refer to this cycle as the Shape & Ship methodology.

---

What is an AI Workflow?

An AI Workflow is a structured, repeatable sequence of actions combining human direction, automated tools, and LLMs (Large Language Models) to accomplish developer tasks.

Unlike simple chat interfaces, modern AI workflows leverage agentic frameworks capable of reading files, executing shell commands, analyzing codebases, and self-correcting errors.

graph TD
    A[User Intent] --> B(Shape Phase: Context & Rules)
    B --> C(Steering: LLM + Custom Skills)
    C --> D(Ship Phase: Code & Test)
    D --> E{Validation Passes?}
    E -- Yes --> F[Deploy / Done]
    E -- No --> G[Self-Correction Loop]
    G --> D

---

The "Shape & Ship" Methodology

Developing software alongside agentic coders requires a shift in how we structure tasks. Rather than writing raw code ourselves, we shape the constraints and guidelines, then let the AI ship the implementation.

Phase	Developer Role	AI Agent Role	Key Artifacts
Shape	Define scope, architectural guidelines, constraints, and success metrics.	Analyze existing files, query APIs, search documentation, suggest schemas.	SKILL.md, brand.json, config schemas.
Ship	Review code changes, approve shell commands, perform integration tests.	Generate source code, compile, resolve lint errors, run local test suites.	Component files, tests, build outputs.

Phase 1: Shaper (Human)

As a Shaper, you establish boundaries. Agents are extremely fast at writing code, but they lack human intuition about product requirements or aesthetic taste.

• You provide the brand constraints (e.g., Incrementic Brand Guide).
• You provide the architectural patterns (e.g., using uv over pip for Python tool management).
• You write or feed machine-readable AI Skills to keep agents aligned.

Phase 2: Shipper (Agent)

As a Shipper, the agent executes the instruction set. It reads the files, generates the diff blocks, runs the compilers, and validates that the solution matches the provided constraints.

[!TIP] Avoid Micromanagement: Instead of telling an agent how to write every line of code, shape the input files, dependencies, and lint rules, then let the agent figure out the syntax and structure.

---

Core Principles

When orchestrating AI Workflows, we adhere to the five core company principles:

1. Plain over Clever: Write explicit prompts and readable code. Clever tricks confuse models. Plain instructions produce reliable outcomes.
2. Curious, Forward-Moving: Always explore the next optimal pattern.
3. Brief: Keep instructions short, concise, and focused.
4. Honest: Provide direct feedback. If the agent's code is buggy, state the error clearly without padding.
5. People over Pixels: Keep the developer experience at the center. AI workflows should reduce friction, not add administrative overhead.

---

Next Steps

Now that you understand the foundational Shape & Ship methodology, proceed to the next parts of the curriculum:

• Using AI Skills: Learn how to load and feed predefined guidelines to coding agents.
• Creating AI Skills: Learn how to draft and validate your own machine-readable guidelines for any project.