Product

Prodigy decides what to build next, turns it into tickets, dispatches agents, and ships the code.

The product has four parts: deciding the next opportunity, turning it into assignable ticket packages, dispatching validated tasks to coding agents, and tracking the work all the way to shipped code. For a deeper walkthrough, watch the overview on the home page or reach out on Contact.

Deciding what to build next

Signal ingestion

Feedback and usage from your existing stack.

Prodigy ingests from the tools your team already uses—no migration required. Feedback connectors normalize unstructured text; usage connectors bring quantitative event streams. Everything lands in one ranked signal pool.

Connects to your stack

FeedbackZendeskIntercomHubSpotTypeformGongSlackNotion

UsageAmplitudeMixpanelDatadog

Dispatches to your agents & boards

AgentCursorClaude CodeCodexWebhook

BoardJiraLinear

Scoring — v1 spec

Deterministic ranking. AI overlay on top clusters.

A versioned scoring formula combines four signals. The same inputs always produce the same output—no black box, no drift. An optional AI layer (Claude, Gemini, or OpenAI) refines only the top-ranked clusters; if it fails, the heuristic score is used automatically.

35%

Frequency

Share of total evidence across clusters

30%

Business impact

Account tier weight, normalised 0–1

20%

Recency

Exponential decay, τ = 45 days

15%

Urgency

Mean negative sentiment across cluster feedback

Opportunity score = 0.45 × pain + 0.30 × reach + 0.25 × usage urgency. Usage urgency joins qualitative clusters to quantitative event streams using a ±7-day time window and tier-filtered denominators.

Opportunities as tickets

Ticket packages — schema v1

Every opportunity becomes structured, assignable work.

When an opportunity is approved, Prodigy generates a persisted ticket package: metadata, ordered tasks, implementation prompts, acceptance criteria, dependency chains, and evidence links back to the original feedback. Packages are versioned as prodigy.ticket_package.v1.

Ranked Problem Clusters

Every cluster scored by business impact, urgency, frequency, and confidence—so prioritisation starts from evidence, not opinion.

Feature Concepts & Specs

Evidence-backed opportunities with UX flows, acceptance criteria, data model implications, and decision rationale ready for engineering.

Rollout & Risk Plan

Dependency mapping, rollout phases, migration notes, and failure risks generated before a single line of code is written.

Agent-Ready Task Bundles

Structured dispatch payloads and Jira ticket exports agents and engineers can pick up immediately—no retranslation required.

Export & push

Jira, Linear, JSON, Markdown, or CSV.

Push ticket packages directly to Jira or Linear with fields synthesised from stored tasks and plan context. Download bundles as JSON (machine-readable), Markdown (human-readable), or CSV (one row per task with pipe-separated dependencies).

JiraLinearJSON exportMarkdown exportCSV export

Traceability

From feedback to PR, every step linked.

feedback_id→insight_id→opportunity_id→task_id→PR

Every ticket export includes a links block with evidence[] mapping each task back to the insight and feedback it came from.

Dispatch to coding agents

dispatch.v1 contract

A validated payload every agent can consume.

Each task is converted to a dispatch.v1 JSON payload validated by a Pydantic contract before send. The payload carries everything a coding agent needs to start without a follow-up meeting.

task_id

Stable UUID for deduplication and callbacks

plan_id

Parent scope for all tasks in a plan

repo + base_branch

Where to branch from (defaults to main)

prompt

Full implementation instruction for the agent

depends_on

Ordered dependency list for sequential execution

branch_name

Pre-named feat/{key}-{slug} branch

callback.url

Where the agent posts its result on completion

context

Plan-level context JSON passed through to agent

Supported agents

Cursor, Claude Code, Codex, or any webhook.

Prodigy resolves the provider from the plan context and routes to the configured webhook. Local autocomplete mode handles development environments without a live endpoint.

CursorClaude CodeCodexWebhook

Feedback loop

Callback → result → plan status.

Agent runs→Opens PR→POST /dispatch/results→Plan status updated→Portal ROI updated

Callbacks are idempotent. A Slack notification is sent on plan completion and the portal ROI snapshot is updated automatically.

Shipping the code + tracking ROI

Client portal

Every workspace ships and sees what changed.

The Prodigy client portal is a secure, multi-tenant workspace where each team sees their shipped changes, ROI snapshots, and pipeline health—all tied back to the original evidence.

ROI tracking

Metric snapshots tied to every shipped change. See delta from baseline over a 14-day window, per product, per workspace.

Pipeline health

Stage success rates, average duration, dispatch reliability, and execution velocity across every pipeline run.

Bug ops dashboard

New bugs by severity, deployed fixes, mean time to resolution, handoff success rate, and AI adoption metrics.

API key management

Workspace-scoped keys with usage telemetry, last-used timestamps, audit trail, and one-click revoke.

Isolation

Workspace-scoped, row-level secure.

Every portal request carries a workspace JWT. Supabase row-level security ensures no data leaks between tenants. Prodigy API calls append workspace_id as a query parameter and require a valid bearer token on every route.

Why Prodigy

Built for decisions, not just data views.

Analytics Dashboards

Show what happened.

Prodigy tells you what to build next—and writes the spec.

IDE Copilots

Help write code faster.

Prodigy decides which code is worth writing and dispatches the task.

Static PM Docs

Capture assumptions manually.

Prodigy generates evidence-backed direction from live signals.

Method transparency

Prioritisation logic your team can inspect.

Insight ranking combines frequency (35%), business impact (30%), recency (20%), and urgency (15%) from both feedback and behavioural analytics signals. Opportunity score further weights pain (45%), reach (30%), and usage urgency (25%).

Every insight includes source traceability and decision rationale. When AI reasoning is enabled, it refines only the top-N clusters and falls back to the heuristic if the model returns an invalid response—so outputs are always deterministic and auditable.