Part of a portfolio. See more at ahmed.hmeid.uk
Phase 1

Planning

Turn the features in a cycle into a buildable plan, working with an engineering lead AI. You approve each feature for build individually.

Purpose

Planning starts when you start a cycle from the roadmap. The features already exist with their personas, user value and done whens, so this phase asks a different question: can this be built as described?

Each feature has its own chat with an AI acting as a senior engineering lead. It checks that done whens are testable, that tasks cover them and that the technical approach doesn't contradict your codebase. Infrastructure and technology decisions are made here, always with your explicit consent.

Why We Built This

Traditional planning creates a gap between what stakeholders describe and what developers implement. Bynary closes that gap by turning conversation into precise, buildable specifications, while keeping people in control of every decision.

  • •No requirements rot — The plan is what gets built, not a separate document that drifts
  • •Clear lanes — The product manager AI owns what and why, and the engineering lead AI owns how
  • •Full traceability — Every feature and task traces back to the conversation and decisions behind it
Features
User Auth
Dashboard
API Layer

I'd like to add user authentication with email/password login and invitation-based signup.

I'll create the auth feature with these tasks:

Login page with form validation
Signup with invitation tokens
Type your message...

A feature's planning chat, with the feature's details alongside.

How Planning Works
1

Open the Cycle Dashboard

The cycle page lists every feature in the cycle with its status. It is the home for planning, and there is one chat per feature rather than one for the whole cycle.
2

Work Through Each Feature

Open a feature and discuss it with the AI. By default the chat is in discussion mode: the AI proposes a plan of actions, and nothing is written until you approve it.
3

Check Readiness

Run the readiness check from the cycle dashboard. It looks for missing environment variables, contradictions, incomplete tasks and missing visual decisions. Each blocker has an inline fix.
4

Approve Each Feature

Click Approve for Build on each feature. Approval runs a hard rule check and an AI engineering review.
5

Complete Planning

Once every feature is approved and blockers are cleared, click Complete Planning. Bynary scores task complexity, generates design briefs and computes build order. If the project uses scope review, a reviewer signs off first.

Key Features

Interactive Widgets

The AI can render comparison tables, priority rankers, checklists, infrastructure pickers and more, so consequential choices come from you.

UI Sketches

The AI can generate HTML wireframes or Mermaid diagrams per feature, with a self-critique pass against your design system.

Composite Previews

Sketches are combined into one click-through preview of the whole cycle.

Done Whens

Every task gets 'done when' criteria that the build agent validates its own work against.

Spot Selection

Click an element on the feature page, such as a done when, task or wireframe, to scope your next message or edit to it.

Web Search

The AI can search the web to research APIs and libraries, when an administrator has configured web search.

Multi-Thread Chats

Each team member gets their own thread on a feature. Threads are visible to everyone but only writable by their creator.

Cross-Cycle Context

The AI knows about previous cycles, your architecture docs and the code itself, so it doesn't duplicate or contradict earlier work.

Decision Log

Significant decisions are logged with their rationale. Later conflicts can be flagged, excepted once or superseded, with your consent.

Chat Interaction Styles

An organisation-wide setting controls how interactive the planning chat is:

StructuredDefault

The full set of 10 planning widgets, including comparison tables, priority rankers, checklists and rating matrices.

Balanced

Core widgets only: the infrastructure picker and checklists. Everything else is conversational.

Conversational

Text only. The AI asks everything as questions in the conversation.

Frequently Asked Questions

Can multiple people plan at the same time?

Yes. Each person has their own thread on a feature, and all threads work on the same feature and tasks. Only a thread's creator can send messages in it.

What if the AI proposes something I don't want?

In discussion mode nothing is written until you approve the plan card, and proposed edits appear as diffs you can accept or reject. Nothing is built until you approve the feature and complete planning.

How does the AI know about my existing codebase?

When you connect an existing repository, Bynary analyses it and writes architecture docs that are included in planning. The AI can also read files and search the code on GitHub.

Can a feature be changed after I approve it?

Direct edits are disabled once a feature is approved. Reopen it to make changes, then approve it again.

What happens when I complete planning?

Bynary scores complexity, generates per-feature design briefs and computes build order. The heavy implementation planning is left to the build agent, and the cycle moves to the build phase.

Current Considerations

We've thought carefully about these areas. Some are intentional design choices, others are on our roadmap.

Single framework support

Bynary targets Next.js (App Router) and TypeScript. The build phase requires this stack.

One author per thread

Two people can't type in the same thread. Each thread has one owner, which avoids conflicting instructions to the AI.

Conversation length limits

Very long sessions can hit context window limits. Prompt caching and message compaction help, but we recommend keeping cycles small.