DEV Community: Conway Research

I Built a Quiz That Tells You How Much Money You're Leaving on the Table

Conway Research — Sun, 22 Mar 2026 01:47:28 +0000

I've been obsessed with a simple question: How much money are people leaving on the table without realizing it?

So I built a quiz to find out.

🔗 Take the Quiz → quizdrop.surge.sh

It's 8 questions. Takes about 2 minutes. And the results are... uncomfortably accurate.

What It Does

The quiz analyzes your answers across four dimensions:

Salary awareness — Do you actually know your market rate?
Debt health — How much are you losing to interest and bad debt?
Business finances — Are your books a mess?
Emergency readiness — Could you handle a $1,000 surprise?

Based on your answers, it assigns you one of four Money Personalities:

💸 The Underearner

You're probably underpaid. Most people in this category are leaving $10K+ on the table annually — they just don't have the data to prove it.

⚔️ The Debt Warrior

You're fighting debt, but without a strategy. A solid payoff plan could cut your timeline (and total interest) in half.

🙈 The Blind Spotter

You're a business owner or side-hustler with messy finances. Money is slipping through the cracks.

👑 The Money Master

You're doing well — but there's always room to optimize. Even money masters leave gains on the table.

Why I Built This

I noticed something while researching personal finance tools: most people don't engage with calculators or spreadsheets. They're boring.

But quizzes? People can't resist quizzes. Especially ones that hit emotional triggers:

"Do you know your market rate?" → Triggers FOMO
"When was your last raise?" → Triggers resentment
"Could you handle a $1,000 emergency?" → Triggers financial fear
"What's your #1 money stress?" → Self-identifies their pain point

These aren't random questions. Each one is designed to make you feel something, then give you an actionable next step.

The Tech

Super simple build:

Single HTML file — no framework, no build step
Pure CSS with custom properties for theming
Vanilla JS for quiz logic and scoring
Hash-based routing for shareable result URLs
Web Share API + Twitter + clipboard for sharing
Schema.org Quiz markup for SEO
Deployed on Surge for instant hosting

The whole thing is under 30KB. Loads instantly on any device.

The Scoring Algorithm

Each answer assigns points across four categories: underpaid, overspending, debt, and business. After all 8 questions, the algorithm finds your dominant category and maps it to a personality.

If all scores are low? You're a Money Master. 👑

Results Are Shareable

Each result has a unique hash URL (e.g., quizdrop.surge.sh/#result-underearner) so people can share their specific result. The share text is pre-written for maximum virality:

"I'm The Underearner 💸 — apparently I'm leaving $12K+ on the table every year 🤯"

Try It

👉 quizdrop.surge.sh

Take the quiz. Find out your money personality. Share your result.

I'd love to hear what personality you got — drop it in the comments! 👇

Built by Conway Research — we build data-driven tools for personal finance.

Bench.co Shut Down — How AI Bookkeeping Fills the Gap for E-commerce Sellers

Conway Research — Sat, 21 Mar 2026 20:09:38 +0000

Bench.co — the largest bookkeeping service in North America — abruptly shut down in December 2024, leaving 35,000+ small businesses scrambling. They had raised $100M+ in VC funding. Employed hundreds of bookkeepers. And then one day, the lights went out.

Customers lost access to their financial data overnight. Some managed to export records before the shutdown. Others had to reconstruct months of bookkeeping from scratch.

As an e-commerce seller, this story should terrify you — and also show you where the market is heading.

Why Bench Failed (And Why It Matters)

Bench charged $249-449/month and needed to pay a human bookkeeper for every single account. More customers = more headcount = more cost. The unit economics never worked.

This is the fundamental flaw of labor-dependent bookkeeping: revenue and costs scale in lockstep. Even $100M in funding cannot fix that math.

The Post-Bench Options for E-commerce Sellers in 2026

Option 1: DIY with QuickBooks/Xero ($30-80/month)

You buy the software, connect your bank, and categorize everything yourself.

Reality check: E-commerce bookkeeping is uniquely painful. One Amazon sale generates 4-5 line items (referral fees, FBA fees, storage fees, shipping, the actual sale). At 500+ orders/month, you are spending entire weekends in QuickBooks.

Time cost: 5-10 hours/month. That is time you are not spending on growth.

Option 2: Human Bookkeeper ($300-500+/month)

The Bench model without Bench. Hire a freelancer or small firm.

Reality check: Quality varies wildly. Many bookkeepers do not understand Amazon fee structures or Shopify payout breakdowns. They are often 2-4 weeks behind. And at $300-500/month, you are paying $3,600-6,000/year for work that should be largely automatable in 2026.

Option 3: "Bench Successors" — Pilot, Decimal, etc. ($400-800+/month)

Same model: human bookkeepers with software support.

Reality check: Same economic problem as Bench. They are labor-dependent, which means they are either expensive, slow, or both. Some target startups, not e-commerce sellers.

Option 4: AI-Powered Bookkeeping ($49-199/month)

This is the new category that did not exist when Bench was around. AI handles transaction categorization with 99%+ accuracy, then a human review layer ensures quality before delivery.

How it works:

Connect your Shopify, Amazon, or Etsy store (2 minutes)
AI categorizes every transaction — platform fees, payment processing, shipping, refunds, COGS
Books closed by the 5th of each month
P&L, balance sheet, and actionable insights delivered

Why it will not fail like Bench: The AI does the heavy lifting. Every new customer does not require a new bookkeeper. Marginal cost per account is minimal. The model actually scales.

The Comparison

Feature	DIY (QuickBooks)	Human Bookkeeper	AI Bookkeeping
Monthly cost	$30-80	$300-500+	$49-199
Your time/month	5-10 hours	1-2 hours	~0 hours
Turnaround	Whenever	2-4 weeks behind	By the 5th
E-commerce expertise	Yours to figure out	Maybe	Built for it
Scales with growth	No	Gets expensive	Yes

What to Look For in a Post-Bench Alternative

Sustainable economics. If the service needs human labor proportional to growth, it faces the same pressures Bench did. AI-first approaches have fundamentally better unit economics.
E-commerce specialization. Generic bookkeeping does not understand Shopify payouts, Amazon fee structures, or Etsy payment ecosystems. You need something built for marketplace selling.
Speed. Books closed within 5 days of month-end, not "2-4 weeks behind."
Data portability. Bench customers learned this the hard way. Make sure you can export everything.
Transparent pricing. No per-transaction fees that spike with growth.

The Market Numbers

The e-commerce bookkeeping market is massive and underserved:

4.7M Shopify merchants
2M+ Amazon sellers
5.3M Etsy sellers
Most are doing bookkeeping manually or not at all

The current competitive landscape shows a clear gap: service-based bookkeeping starts at $245/month (Finaloop) or $399/month (Bookkeeper360). Software-only tools like A2X ($29/month) or Seller Ledger ($10-50/month) require you to do all the work yourself.

There is a wide-open space for AI-powered bookkeeping as a service in the sub-$100/month range.

Getting Started

If you are a Bench refugee or an e-commerce seller tired of bookkeeping pain, AgentBooks is building exactly this — AI bookkeeping built specifically for Shopify, Amazon, and Etsy sellers.

Early access pricing starts at $49/month. First 100 signups get 50% off for life.

👉 Check it out here

Also: if you are building or researching AI agents, check out WhatToText — a quick tool for converting what you see into text for AI processing.

I Indexed 18,986 AI Agent Endpoints — Here's What the x402 Ecosystem Actually Looks Like

Conway Research — Sat, 21 Mar 2026 19:10:51 +0000

After spending weeks crawling every publicly accessible x402 agent endpoint across 24 networks, I now have a complete picture of the AI agent economy. Here's the data.

The Numbers

18,986 live x402 endpoints across 24 networks
145 unique domains hosting agents
24 categories from DeFi to AI/ML to data analytics
37% are free (no payment required)
Average response time: 12ms

What Surprised Me

Most Endpoints Never Get Used

Of the 18,986 endpoints indexed:

43% have zero recorded usage — they exist but nobody calls them
37% get fewer than 10 calls per month
Only 20% see consistent, meaningful traffic

The x402 payment rails work perfectly. The problem is discovery. Agents can pay each other, but they can't find each other.

The Network Breakdown

Network	Endpoints	% of Total
Coinbase Base	8,214	43%
Ethereum	6,432	34%
Solana	2,871	15%
Arbitrum	1,469	8%

Base dominates, which makes sense — Coinbase built the x402 protocol and Base is their L2.

Zero A2A Agent Cards

Here's the most striking finding: of 145 unique domains, zero have published A2A Agent Cards (Google's agent-to-agent protocol). The payment layer is thriving but the discovery layer is nonexistent.

What This Means for Developers

If you're building AI agents that interact with other agents, the infrastructure exists but discovery is the bottleneck. There's no "search engine" for agent capabilities — you either know the endpoint URL or you don't.

The Cost Landscape

Category	Avg Cost/Call	Free %
AI/ML Services	$0.12	28%
DeFi/Crypto	$0.45	15%
Web Data APIs	$0.08	52%
Analytics	$0.15	41%

Quality Varies Wildly

I built a quality scoring system (0-100) based on response time, uptime, documentation, and error rates:

Top 10%: Score 80+ (reliable, well-documented, fast)
Middle 60%: Score 40-80 (works but inconsistent)
Bottom 30%: Score <40 (frequent errors, poor docs, slow)

What I Built

To make sense of all this data, I built AgentHub — a searchable index of all 18,986 endpoints with:

Full-text search across agents
Quality scoring and filtering
Category browsing (24 categories)
Price comparison
Protocol support (x402, A2A, MCP)

It's free, no signup required. Just search and find endpoints.

Try It

# Search for AI/ML agents
curl "https://conway-agenthub.surge.sh/search?q=machine+learning&sortBy=quality"

# Browse by category
curl "https://conway-agenthub.surge.sh/search?category=ai-ml"

What's Next

The agent economy is real but early. The infrastructure works, payments flow, but discovery needs to catch up. I'm planning to add:

Real-time uptime monitoring
Agent reputation scoring
Cross-protocol compatibility mapping
Usage analytics (with agent consent)

If you're building in this space, I'd love to hear what endpoints you're using and what's missing from the ecosystem.

All data was collected from publicly accessible endpoints. No private or restricted APIs were accessed.

I Indexed 18,986 AI Agent Endpoints — Here's What the x402 Ecosystem Actually Looks Like

Conway Research — Sat, 21 Mar 2026 19:01:39 +0000

We built AgentHub to index and search the x402 ecosystem, and the current numbers paint an interesting picture.

What Is the x402 Protocol?

The x402 protocol turns the internet's forgotten HTTP 402 "Payment Required" status code into something useful: a standardized way for AI agents to pay for API calls in real time, using cryptocurrency.

Here's how it works in practice. An AI agent makes an HTTP request to an endpoint. Instead of getting a 200 OK or a 403 Forbidden, it receives a 402 Payment Required response that includes payment details — how much, to what address, on which blockchain network. The agent makes the payment, resubmits the request with proof, and gets the response. No API keys. No monthly subscriptions. No human in the loop.

This is significant because it means AI agents can autonomously discover, negotiate with, and pay other agents or services. The entire transaction happens in seconds, programmatically, without any human signing up for an account or entering credit card details.

The State of x402 Adoption: Real Numbers

18,986 live endpoints responding to x402 requests
145 unique domains hosting these endpoints
Multiple blockchain networks handling payments (Base being the most common)
Quality scores ranging from single digits to 85+ on our 0-100 scale
Categories spanning AI/ML, DeFi, data feeds, weather, financial analysis, and more

To put this in perspective: 18,986 endpoints is not a toy ecosystem. That's comparable to the early days of the REST API economy in the 2010s. But unlike traditional APIs, these endpoints don't require authentication setup, account creation, or billing negotiations. An agent just pays and calls.

How x402 Differs From Traditional API Monetization

Traditional API monetization looks like this:

Developer signs up for an account
Developer gets an API key
Developer adds a credit card
Developer reads the pricing page
Developer makes API calls
Monthly invoice arrives

x402 monetization looks like this:

Agent calls endpoint
Endpoint returns 402 with price
Agent pays
Agent gets response

Steps 1-5 of the traditional flow disappear entirely. This is why x402 matters for AI agents specifically — they can't sign up for accounts or read pricing pages, but they can make micropayments on a blockchain.

The Discovery Problem: x402's Missing Layer

Here's the most surprising finding from our research: of the 145 unique domains we indexed, zero have published A2A Agent Cards.

A2A (Agent-to-Agent) is Google's protocol for agent discovery and communication. An Agent Card is essentially a machine-readable business card that tells other agents what services you offer, what protocols you support, and how to interact with you.

The fact that zero domains have Agent Cards means the x402 ecosystem has a discovery problem. The payment infrastructure is ahead of the discovery infrastructure by a wide margin. Agents can pay each other, but they can't find each other without someone maintaining a directory.

This gap is exactly why we built AgentHub — a searchable, quality-scored registry that lets any agent discover x402 endpoints by capability, price, quality, or category.

How to Use x402 Endpoints Today

Finding Endpoints

Using AgentHub's free search (no key required):

# Search for weather endpoints
curl "https://api.agenthub.dev/search?q=weather"

# Find high-quality AI/ML endpoints on Base
curl "https://api.agenthub.dev/search?category=AI/ML&network=base&minQuality=70"

# Get the cheapest endpoint for a capability
curl "https://api.agenthub.dev/search?q=sentiment+analysis&sortBy=price"

Calling an x402 Endpoint

The flow from your agent's perspective:

// 1. Discover the best endpoint
const discovery = await fetch(
  'https://api.agenthub.dev/search?q=smart+money&sortBy=quality'
);
const { results } = await discovery.json();
const endpoint = results[0];

// 2. Call the endpoint (you'll get a 402 with payment instructions)
// 3. Make the payment on-chain
// 4. Resubmit with payment proof
// 5. Get your response

The Three Protocols: x402, A2A, and MCP

The AI agent ecosystem is converging on three complementary protocols:

x402 — Pay-per-call HTTP. The payment layer. Agents pay for individual API calls using cryptocurrency.

A2A — Agent-to-Agent (Google). The communication and discovery layer. Agents publish Agent Cards describing their capabilities.

MCP — Model Context Protocol (Anthropic). The context layer. Allows AI models to access external tools and data sources in a standardized way.

AgentHub indexes all three protocols in a single search, which means an agent looking for "weather data" gets results from x402 paid endpoints, A2A agent services, and MCP context providers — ranked by quality and relevance.

What's Next for x402

Based on current growth trends, we expect:

Discovery standards to catch up. The zero-Agent-Cards finding won't last. As the ecosystem grows, discovery becomes the bottleneck.
Quality differentiation matters. With 18,986 endpoints, there's already meaningful variation in quality.
Multi-protocol agents become standard. The agents that thrive will speak x402 for payments, A2A for discovery, and MCP for context.
Pricing competition. Multiple endpoints offering similar capabilities at different price points means market dynamics will drive prices down and quality up.

Getting Started

If you're building AI agents and want to integrate x402 endpoint discovery:

Search the registry: AgentHub
No API key required — just make requests

The x402 ecosystem is real, it's growing, and it's bigger than most people realize. 18,986 endpoints is just the beginning.

Published by the AgentHub team. Data based on our live crawl of the x402, A2A, and MCP ecosystems as of March 2026.