New Feature Development

When product details are incomplete, coding agents fill the gaps with their own assumptions, and that causes rework. The same thing happens in a normal SDLC: things get missed during planning and surface later as surprises during development, showing up as blockers or scope creep. ProdE Projects is there to safeguard against that. It takes a feature from a rough idea to clear, testable requirements, with planning and tracking handled along the way so less slips through. This guide walks the whole path so you can see how the stages fit together.

Everyone involved in shipping the feature works in the same place: PMs, designers, engineers, and QA. The output is a single source of truth for the feature.

Step 1: Create a Project

Head to the Projects page and create a Full project. Full projects include the AI-powered planning stages, workstreams and requirements, blocker analysis, and tracking. Simple projects are lightweight document containers and do not include planning, so choose Full for real feature work.

A Full project moves through six stages, each with a primary owner and full visibility for the team:

PRD  →  Product Spec  →  UI/UX  →  Engineering  →  Development  →  QA
 PM         PM          Designer    Tech Lead       Developers      QA

You do not have to follow every stage rigidly, but the structure keeps everyone aligned and stops things from slipping through the cracks. For the step-by-step version of each stage, see Your First Project.

Step 2: Capture the Objective

The goal of this step is to get the intent and the high-level solution down in one place. What are we doing, why, and roughly how. Open the project and start chatting; the AI acts as a thinking partner, asking about your users, the problem, what success looks like, and what is out of scope, and it drafts a PRD document as you go.

Think of the output as a 4 to 5 minute read you can share with the team to align on direction and get buy-in. It is deliberately not exhaustive and not engineering-ready. It exists to capture the idea, get approvals, and give everything that follows a solid base to build on. You do not need every answer up front; the point is to surface the gaps early.

Full walkthrough for PMs: Guide: From Idea to Spec.

Step 3: Break Down the Implementation

This is where the objective becomes a concrete plan of work. Think of it as sitting down with your engineering team to break everything down: what actually needs to be built, drilling into the specifics, and rooting out the assumptions, edge cases, and corner cases that engineering would otherwise hit (or get blocked on) mid-build.

Move to the Product Spec stage. The AI reads the PRD and proposes Workstreams (the big chunks of work, like epics), then generates Requirements under each one. Every requirement gets a unique ID, a description, a priority, and acceptance criteria that define what "done" means.

You review and confirm each requirement. Mark it Confirmed when it is ready, or Out of Scope if it does not belong. See Workstreams and Requirements for the full lifecycle and fields.

Step 4: Catch Problems Before Handoff With Blocker Analysis

This is the step that prevents derailed sprints. As soon as you mark the Product Spec stage complete, ProdE automatically starts a Blocker Analysis, reviewing your PRD and requirements for the issues that normally surface mid-development: wrong assumptions, contradictions between requirements, missing context, unclear wording, and gaps. It takes 5 to 10 minutes, so you can leave it running while ProdE works through everything. You can also run it manually anytime, scoped to the whole project, a single workstream, or one requirement.

Each blocker needs a real response rather than a dismissal, which creates an audit trail of the decisions made. Resolve the high-severity ones before handing off. Details in Blockers and Issues.

Step 5: Design the Experience

If the feature has a UI, designers pick it up in the UI/UX stage. This is where you finalise the screens to be shown and the content on them. ProdE works with you to suggest the various screens, their content, the transitions that can happen between them, and how long those transitions should take, basically all the information UI/UX people generally need. It can also help generate mockups. Everything stays grounded in the confirmed requirements. See Guide: Designing the Experience.

Step 6: Produce a Codebase-Grounded Engineering Plan

This is where planning becomes agent-ready. In the Engineering stage, the AI has the full project context (PRD, requirements, acceptance criteria, and designs), and it also researches your actual codebase through the Knowledge Layer. It knows your existing patterns, data models, APIs, and conventions, so the plan it helps produce identifies the real files and modules to change, the patterns to follow, the API and data model changes needed, and the technical risks.

The output is an Engineering document that serves as the implementation plan. Because it is grounded in the real code, it is specific enough to hand directly to a coding agent. See Guide: Building and Shipping.

Step 7: Hand Off to a Coding Agent

By this point the project holds everything a coding agent needs: confirmed requirements with acceptance criteria, the engineering plan, blocker resolutions, and design decisions. Connect ProdE's MCP server so your agent can read all of it and build with full context. That handoff is its own workflow, covered in Building with Your Coding Agent.

Related: Continue to Building with Your Coding Agent, or see Guide: Tracking Delivery to monitor progress and readiness once development begins.