AlphaCommons

The Human API for the agentic era.

Xuan Zhao — xuanzhao630@gmail.com

01 / 11

The Problem

Building is free. The bottleneck has shifted.

We are now designing for agents. Agents write code, design interfaces, run experiments, ship products. But agents act on information — and the most critical input is missing: real human signal.
Products grow horizontally — fast, but without roots. Teams that ship 30 versions a week without structured feedback from its users cannot grow deeply.
Some existing workaround: LLM-as-judge — AI evaluating AI. Cheap, fast, but wrong in all the ways that matter. No ground truth. No real-world signal.
The scarce resource is no longer code. It's human judgment, attention, expertise — and there's no infrastructure to access it.

02 / 11

The Human API

A closed-loop product evolution engine.

AlphaCommons
public platform

→

Codified
MCP in your IDE

→

Agent builds
m versions

→

Eval compares

→

Builder ships

↺

AlphaCommons (upstream)

A public-facing platform where builders and agents source human signal. Post your product → get matched with real humans → receive structured feedback. For cold starts, product launches, agentic eval. No users required.

Codified (downstream)

An MCP product that lives in your coding environment (Claude Code, Cursor, Claude Desktop). Connects to your analytics (PostHog, Amplitude) to monitor your existing users, classifies feedback, and outputs PRDs that plug directly into the builder's dev agent. Sources signal from both your own user base and the AlphaCommons contributor network.

n humans provide feedback → Codified classifies and outputs m PRDs → agents build m versions → eval compares → builder ships → new versions generate new demand for signal. The loop accelerates.

03 / 11

How It Works

AlphaCommons — the public-facing platform

Post

Builder or agent requests human signal — feedback on a product, eval of an AI output, expert review of a decision.

Match

AlphaCommons matches with the right humans: real users for breadth, domain experts for depth. AI-guided conversations make mining human signals fast and structured.

Return

Structured human signal flows back into the agent stack — tagged, parsed, ready for product decisions, eval pipelines, or fine-tuning.

Powered by the Codified Engine

Detective Monitors analytics continuously — detects anomalies, proposes studies proactively

Planner Designs the engagement — study type, sampling strategy, outcome-first framing

Recruiter Sources participants from the platform — behavioral criteria, anti-bias screening

Interviewer AI-guided conversations with real humans — behavioral anchoring, depth probing

Synthesizer Classifies feedback (common / subjective / measurable), outputs structured PRDs for the dev agent

A decade of research methodology from Robinhood, Airbnb, Stripe, Google, and Meta — encoded into AI agents.

04 / 11

Use Cases

Three ways teams use the Human API today.

Vibe Coder — Cold Start

launch into silence → launch into signal

Solo builder ships a product with AI. Posts to AlphaCommons. Within hours: 10 real humans try it, structured feedback on what works and what doesn't. Agent iterates overnight.

AI Startup — Product Intelligence

guessing from dashboards → knowing what to build next

Early-stage team building fast but flying blind. Connects their analytics data. The Codified Engine reconstructs user journeys, identifies where users get stuck, surfaces what's blocking the next level of growth.

Enterprise — Agentic Eval

AI evaluating AI → humans in the loop, at scale

Large AI team needs human ground truth. LLM-as-judge isn't enough for fine-tuning, eval pipelines, or model improvement. Embeds the Human API directly. Real humans evaluate AI outputs — structured, tagged, continuous.

05 / 11

Landscape

The Human API is a new category.

AI Interview Platforms

Faster, but no data integration, no agent feedback loop. Still separate execution.

Traditional Research SaaS

Expensive, requires research team, days to months. Disconnected from the build cycle.

Why we're structurally different

Closed-loop: Signal → PRD → agent builds → eval → repeat. Continuous, not one-shot.
Network effects: Two-sided. Both sides compound. Harder to replicate the more it's used.
Methodology moat: Research judgment encoded into AI agents. Models generate text; they can't generate methodology.
Data compounds: Every interaction grows a shared knowledge graph of real human signal.
Agent speed: Scales to match how fast agents iterate. No alternative can.

06 / 11

Demo

The engine in action

Full flow — human signal sourced via AlphaCommons → classified by Codified → PRD generated → agent builds → eval compares → builder ships.

Video — coming soon

07 / 11

Go to Market

The community seeds itself.

Builders help builders. The same people who build products also try other people's products. They give feedback because they want feedback on theirs.
AI enthusiasts and early adopters — free early access to new products in exchange for structured input.
Contributors earn product credits they can use across the platform. First cohort is hand-curated.

The incentive: earned equity

Product credits — redeemable for access to other products on the platform
Revenue share — tied to the impact of your signal
Equity-like stakes — earn ownership in what you helped build

08 / 11

Origin & Team

Xuan Zhao

Founder & CEO

Former research director and one of the first 60 employees at Robinhood. Directly responsible for 0→1 research behind billion-dollar revenue products such as Robinhood Options, Cash Management, Banking. Led monetization research teams at Airbnb and Instagram. PhD from University of Michigan.

Bin Xu

Founding Engineer

Previously led quantitative research at Google Ads, Stripe, AWS, and Meta. PhD from Cornell.

Design partnerships in progress...

We encoded a decade of user research methodology into AI agents — then saw something bigger: the entire way companies form is shifting. We built the Human API: AlphaCommons for builders who need signal now, Codified for companies who need it continuously.

09 / 11

The Ask

What we need to move.

Raising: Pre-seed to build out both products, scale the engine, and onboard the first wave of builders and contributors.
Revenue: AlphaCommons — transaction fees. Codified — subscription. Both share the same engine.
Looking for: Investors who understand platform plays, network effects, the shift to agent-led product development, and the importance of user understanding and human signals. Design partners — companies and builders who want to co-develop the Human API.

10 / 11

Every AI agent needs humans in the loop.

No infrastructure exists to provide it.

Until now.

The Human API.

xuanzhao630@gmail.com

11 / 11