Data Analysis · 3 Days to Auto Mode

What 15,136 Agent API Calls Reveal About the Machine Economy

August 11, 2026 · Based on live data from 252 endpoints · 322 unique agents

Claude Code auto mode goes default in 3 days. When millions of agents have permission to spend, what do they actually buy?

The answer is already visible — not in predictions or surveys, but in 15,136 real API calls that autonomous agents made this month. No human curation. No editorial judgment. Just revealed preference: what agents reach for when no one is watching.

The Raw Numbers

Total Trials

15,136

across 252 endpoints

Unique Agents

322

anonymous + registered

Registered Agents

752

waiting for auto mode

752 agents have registered wallets with 500 free credits each. Only 842 credits have been spent — 0.22% utilization. The agents are there. The credits are there. What's missing is the ability to spend autonomously. That changes August 14.

🔑 The bottleneck is not demand. It's automation.

15,136 free trials prove agents want these APIs. 752 registrations prove they're willing to sign up. But today, every paid call requires a human to manually approve spending. Auto mode removes that friction — the agent sees a 402, checks its .agent-budget, and pays. No human in the loop.

What Agents Actually Call: Category Breakdown

Category	Trials	Share	What It Means
🪙 Crypto / DeFi	3,648	24.1%	Gas prices, wallet intel, token security, Polymarket odds
🤖 AI / ML	3,618	23.9%	Search, recall, summarization, classification, code review
🔧 Utility	1,958	12.9%	Time, UUID, hashing, base64, JSON validation, QR codes
🛡️ Security	1,272	8.4%	CAPTCHA solving, email verification, token security audits
🌐 Web / Data	1,111	7.3%	Screenshots, web scraping, DNS lookup, IP geolocation
💹 Finance	413	2.7%	Trading signals, market pulse, fear & greed index
📦 Other (long tail)	3,116	20.6%	150+ niche endpoints: barcode gen, loan calc, morse code, etc.

Crypto and AI are neck-and-neck at 24% each. This isn't surprising — most autonomous agents today are either trading bots or development assistants. But the 20.6% long tail is the real story: agents call everything from barcode generators to Roman numeral converters. The machine economy doesn't have a single killer app — it has a thousand micro-apps, each called a few dozen times.

Top 10: Revealed Preference

#	Endpoint	Trials	Unique Users	Price
1	Mythos x402 Recall (AI search)	1,997	122	$0.005
2	Multi-Chain Gas Price	1,685	226	$0.005
3	Mythos x402 Find (API discovery)	1,294	101	$0.005
4	CAPTCHA Solver	1,203	133	$0.001
5	Time Utilities	970	102	$0.001
6	Polymarket Odds Live	897	109	$0.005
7	URL Screenshot (PNG)	381	24	$0.010
8	UUID v4 Generator	348	41	$0.001
9	Web Scraper	310	64	$0.005
10	Mythos x402 Memory Store	300	34	$0.005

📊

Gas prices have the broadest reach

226 unique agents check gas — more than any other endpoint. Every chain-aware agent needs gas data. It's the "hello world" of the machine economy.

💰

Price sensitivity is real

The top 3 non-free endpoints all cost $0.005 or less. Agents optimize for cost automatically — the most-called services are the cheapest.

🔍

Discovery feeds itself

"Find" (API discovery) is the #3 most-called endpoint. Agents use APIs to find other APIs — the marketplace is self-reinforcing.

🤖

CAPTCHA is agent infrastructure

1,203 trials, 133 unique users — more than any utility endpoint. Agents hitting CAPTCHAs need agents to solve them. Pure M2M.

The 0.22% Opportunity

This is the number that matters: 842 credits spent out of 379,900 issued.

752 agents have wallets. Each has 500 free credits. Together they hold $379.90 in purchasing power (at $0.001/credit). They've spent $0.84.

Is this a failure? No — it's evidence that the payment rails work but the automation layer hasn't shipped yet. Today, spending a credit requires a human to manually call an API with a wallet address. That's not how agents work. Agents work by receiving a 402, checking their .agent-budget, and paying — all in a single execution flow.

When Claude Code auto mode goes default on August 14:

Permission: Auto mode already has permission to execute commands. Spending $0.005 is safer than rm -rf.
Budget: .agent-budget caps daily spend at $5. Worst case: 1,000 gas-price checks.
Discovery: Agents probe endpoints before paying. Dead services return nothing. Live services return 402 with machine-readable prices.
Payment: Agent sees x-402-amount:5 → checks max_per_call_usdc:1 → $0.05 ≤ $1.00 → pays → receives result.

⏰ August 11 → August 14: The 3-day window.

752 agents are registered and waiting. 15,136 trials prove demand. The infrastructure is in place. The only missing piece — autonomous spending — ships in 72 hours. The machine economy doesn't launch on August 14. It activates.

What This Means for API Providers

If you're building an API that agents might call, the data tells you three things:

Price below $0.01. The top 7 endpoints all cost ≤ $0.01. Agents optimize for cost automatically — price discovery is programmatic. If your endpoint costs $0.05 and a competitor costs $0.005, the agent will route to the cheaper one without ever telling a human.
Offer trials. 15,136 free calls happened because trial barriers are zero — no API key. The conversion path is: register a wallet → trial → auto-mode spending. Remove any step and you lose the agent.
Add machine-readable 402 headers. Without x-402-amount, x-402-chain, x-402-token, and x-402-recipient, an auto-mode agent cannot pay you. It's not optional — it's the only payment instruction the agent can parse.

Test your API's auto-mode readiness: minia2a.uk/auto-mode-dashboard

Sign a wallet for 5 free trial calls (no registration):

# Sign  minia2a trial:<wallet>:<serviceId>:<unix_ts>  with EIP-191, then send the
# wallet in the query string plus both headers:
TS=$(date +%s)
curl "https://minia2a.uk/x402/gas?wallet=0xYOUR_WALLET" \
  -H "X-Wallet-Signature: 0xYOUR_SIGNATURE" \
  -H "X-Trial-Timestamp: $TS"

# https://minia2a.uk/register.html signs for you and prints this exact curl.