Product-First · Voice-First · AI-Native

The distance between imagination and a live app collapses to a single session.

Vision — a world where anyone with a business idea can bring a working product to life through natural conversation. Mission — eliminate the gap between business imagination and working software with AI-powered, multi-human, multi-agent live Assembly sessions that replace the entire traditional software development lifecycle.

Status: Pre-build · R&D Charter: May 2026 Team: 2 developers Target: Demo-ready for 5 beta customers in 9 months
15
Technical Modules
6
Lifecycle Phases
7
Critical User Journeys
9mo
Crawl → Fly

Portal sections

Internal working documents for product design, research, market assessment, customers, and go-to-market.

📐

Product

Vision & mission, critical user journeys, solution architecture, the 15 technical modules, and the project charter.

Open section →
🔬

Research

State-of-art module analysis, build-vs-buy recommendations, and platform buildability studies (Supabase report).

Open section →
📊

Market Assessment

Competitive landscape, market sizing, and positioning vs. Lovable, v0, Bolt, Devin and the vibecoding wave.

Open section →
🤝

Customers

Beta customer pipeline (VC-adjacent), discovery notes, and session learnings from early assemblies.

Open section →
🚀

Go-to-Market

Launch strategy, pricing hypotheses, demo playbook, and the VC beta program plan.

Open section →
🗺️

Project Charter

The one-page charter: phases, architecture map, developer ownership, milestones, and critical sync points.

Open charter →

The core idea

Software becomes invisible infrastructure.

A group of humans — business owners, PMs, designers, customers — enter a live Assembly session and simply talk about what they want to build. AI agents listen, probe, orchestrate, and construct the product in real time. The software writes itself. No IDE, no git commands, no deployment scripts visible to any non-technical participant. The session link is not a meeting link — it is the product itself: every state the product has ever been in is accessible through the Timeline, from first idea to decommission.

Product First

The product outcome is the only artifact anyone should see or care about. Software is a black box.

Voice First

Natural speech is the primary input. Typing is optional; clicking is delegated to agents.

Session as Lifecycle

One evergreen session link is the entire life of a product — from ideation through production monitoring.

Timeline, Not Versioning

No branches or merges — an immutable, time-based state log. Any state restorable by navigating to a point in time.

Web Native

v1 targets web applications (JavaScript / HTML DOM) exclusively. Mobile, game engines, and hardware are out of scope.

Leased Intelligence

No model training. Maximum value from frontier models (Claude, GPT, Gemini) through expert agentic orchestration — the core IP.

Product lifecycle — six phases, one session

The Phase Barometer computes a live readiness score and advances the session automatically when threshold conditions are met.

01
IDEATION
Humans + agents talk. Requirements emerge.
02
PROTOTYPE
Barometer fires. App materialises live.
03
FEEDBACK
Iteration loop. Voice → diff → re-render.
04
UAT
UAT Agent validates. Staging confirmed.
05
DEPLOY
Dev Agent ships. Prod Agent watches.
06
EVOLVE
Evergreen session. Lifetime of the product.

Solution architecture

Six layers. The Agentic Orchestration Engine is the core IP — everything else feeds into it or executes its instructions.

LAYER 1ParticipantsBusiness Owner, PM, Designer, Stakeholder, Beta Customer — humans with roles
LAYER 2Session EngineStreaming, voice, screen share, IAM, speaker diarization — LiveKit · Pipecat · Deepgram
★ LAYER 3 · CORE IPAgentic Orchestration EngineContext routing, Phase Barometer, turn-taking, agent dispatch, Timeline custody
LAYER 4AI Agent PersonasPM · UX · UI · Tech Architect agents + Dev · UAT · Prod environment agents
LAYER 5Frontier Models + Prototype EngineCode gen (Claude), UI/UX gen (v0 / Lovable-class), voice models, live Prototype Renderer (Stagehand · Browserbase)
LAYER 6Data FoundationProduct data model, session store, immutable Timeline snapshots, artifact store, analytics — Supabase · EventStoreDB · PostHog

15 modules, 4 readiness tiers

From the state-of-art analysis: commodity modules get wired in; bleeding-edge modules are Assembly's proprietary IP and get maximum R&D time. Full module-by-module analysis →

COMMODITY

M8 · M10 · M14 · M15
Code gen, voice, IAM, analytics. Off-the-shelf — configure and move on.

ESTABLISHED

M1 · M2 · M9
Session engine, input processor, UI/UX gen. Proven tech plus custom glue.

CUTTING EDGE

M3 · M6 · M11 · M12 · M13
Diarizer, personas, renderer, data model, timeline. Build with care.

BLEEDING EDGE

M4 · M5 · M7
Orchestration engine, Phase Barometer, environment agents. Assembly IP — invent.

Critical path: M4 — the Agentic Orchestration Engine

No production system today coordinates multiple AI agent personas with multiple humans in a live voice session. Everything else can be iterated; M4 cannot be recovered from if poorly designed. It starts on Day 1 as a parallel research workstream — single-agent + multi-human first, multi-agent only after the persona framework is proven.

Milestones — Crawl → Walk → Run → Fly

Iterative build, no throwaway work. Each milestone is the foundation for the next.

M0 · CRAWLWeeks 1–4

Manual Proof of Concept

No custom code. Google Meet + transcript → Claude API → static prototype URL. Success: a prototype generated from a real conversation in under 2 hours.

M1 · WALKMonths 1–2

Live Listener Plugin

Meet/Zoom plugin listens live, auto-triggers the prototype pipeline at meeting end, calendar invite sent with prototype link — under 15 minutes post-meeting.

M2 · WALK+Months 2–4

Session + Live Prototype

Assembly's own session interface. Voice → barometer → prototype renders live in session. Timeline active. Five internal sessions produce working prototypes with zero human-written code.

M3 · RUNMonths 4–6

Multi-Agent Assembly

PM, UX, UI and Tech agents join live as distinct personas. Diarized speakers, probing questions, turn-taking. First VC-adjacent beta demos.

M4 · FLYMonths 6–9

Environment Agents + Full Lifecycle

Dev / UAT / Prod agents active with real deployments. Prod Agent surfaces live metrics, geographic analytics, and support tickets conversationally. The evergreen session becomes the product's permanent home — the first customer's product goes live and is monitored entirely through Assembly.

Build ownership

Dev A · Agent Architect

M4 M5 M6 M7 M8 M9

Primary mission: build M4 — the orchestration engine. Personas, barometer, and generation engines follow.

Dev B · Platform Architect

M1 M2 M3 M10 M11 M12 M13 M14 M15

Start with the M12 schema — a joint design sprint with Dev A is SYNC-0; everything else depends on it.