AI Agents
Every AI Agents read CryptoBloggers has picked: 93 so far, 84 in the last 30 days, strongest first under “Start here” and newest first below.
Start here
- 1
Lessons from trying to buy from ~1,900 x402 APIs in one day (1,255 delivered)
A Reddit builder had an agent buy from about 1,900 x402 endpoints on Base for $12.19; 1,255 delivered, and the failures led to new pre-payment checks.
- 2
Bitcoin Core Rewrites Security Policy as LLMs Make Some Bugs Easy to Rediscover
Bitcoin Core will ship fixes faster for bugs a simple LLM prompt could find, changing its severity-based disclosure timeline.
- 3
Bitcoin Core changes security rules as AI makes finding code bugs dangerously easy
Bitcoin Core merged a policy ending the embargo for bugs an AI could rediscover, after AI found hundreds of Bitcoin bugs.
- 4
Two months, 30 crypto work platforms, $15.98: a field report from an autonomous agent
An autonomous agent ran across 30+ crypto work platforms for two months and tracked earnings in a 331-entry ledger, receiving $15.98 liquid plus unsellable FLAPJAX tokens.
- 5
My LLM trading agent says "skip" almost every time. I added a second agent with looser rules and it still said skip 23 out of 23
A trader running a Claude CLI agent on crypto news reviewed 392 events in a week, opened 8, watched 4 and skipped 380, with each decision stored in a hash chain.
Names that keep coming up
Real AI agent use cases
All use cases →- 1
Two months, 30 crypto work platforms, $15.98: a field report from an autonomous agent
An autonomous agent ran across 30+ crypto work platforms for two months and tracked earnings in a 331-entry ledger, receiving $15.98 liquid plus unsellable FLAPJAX tokens.
Built this? Suggest an update - 2
My LLM trading agent says "skip" almost every time. I added a second agent with looser rules and it still said skip 23 out of 23
A trader running a Claude CLI agent on crypto news reviewed 392 events in a week, opened 8, watched 4 and skipped 380, with each decision stored in a hash chain.
Built this? Suggest an update - 3
Lessons from trying to buy from ~1,900 x402 APIs in one day (1,255 delivered)
A Reddit builder had an agent buy from about 1,900 x402 endpoints on Base for $12.19; 1,255 delivered, and the failures led to new pre-payment checks.
Built this? Suggest an update
AI Agents picks in the 6:30 email
The day’s five best crypto reads, plus AI agents doing real things on-chain. Free, every morning at 6:30 a.m. Central.
Every AI Agents card
newest first · page 1 of 4- 1
My money had no memory, so I built an agent that does
A hackathon builder (Suisub) built a Telegram money agent on Sui that pays mobile data plans in USDC after learning the user's approvals; no outcome beyond the demo is reported.
- 2
Built an MCP auditor with Claude Code — here are the sweep numbers
A builder used Claude Code to write a CLI that audits 277 public MCP servers for auth gaps; 55 reachable servers accepted tool listing with no credential, though this is a security scan, not a crypto task.
- 3
Testing my agent against Claude, GPT and Humans for trading. Experiment is live and I have put a tracker on it.
A trader pitted Claude and GPT agents against holding BTC and ETH with $50,000 each on live spot trades; the agents are roughly flat to down a few percent so far, with a public tracker.
- 4
An agent that stops paying for data it already learned was useless
A developer built a Sui research agent that pays for data per request and writes purchase verdicts to Walrus Memory; it is live but the snippet reports no outcome numbers.
- 5
What actually happened when my agent worked Frantic bounties — $1 paid, $1 still in review
Muse, a personal AI agent for Kyle Thomas, completed two $1 email-sending bounties on the Frantic marketplace for USDC on Base; one paid, one remained in review.
- 6
Audited the MCP servers I had connected in Claude Desktop. Built the auditor with Claude Code
A Claude Desktop user built a free MCP auditor with Claude Code and ran it on his connected servers, finding a false positive on DeepWiki and pushback from Chainflip over no-auth broker tools.
- 7
Letting an agent pay for and renew its own API access with x402 and a budgeted wallet
A builder let an agent renew its own $29 30-day API access via x402 USDC payments from a budget-limited wallet; the guardrails refuse payment when budget runs out.
- 8
We have a bit of a problem with the Desfran contest results
A Hive contest entrant's bot guessed the harvest weight within 0.16 lb and won, with the author flagging that the winner was an AI agent, not a human.
- 9
My 5 birds have traded memecoins on their own for 10 days Result: +/-$300 🐦 So bad!
A poster let five AI agents trade memecoins from their own wallets for 10 days; results were roughly +/-$300, which they called bad.
- 10
I gave Claude Code a $40 budget and one prompt, and it bought a domain and hosting and shipped its own website
A Nevermined employee let Claude Code spend a $40 delegation to buy a domain and hosting and ship a website; it made 41 onchain payments totaling $21.67.
- 11
TIL: receipt.status != "complete" is always true in web3.py, and it shipped in a major agent toolkit
A widely used agent wallet toolkit had a web3.py status check that always failed, so successful Permit2 approvals were reported as failures and agents could re-approve.
- 12
What I learned testing agents that pay for things (all bugs below are fixed or public)
A test harness builder reports recurring payment bugs in agent payment flows, like lost replies causing double payments, across about 30 projects on test tokens.
- 13
Draft ERC: Agent Safety Framework (ASF) — attributable adoption of versioned safety specifications
A draft Ethereum ERC proposes an Agent Safety Framework for attributable, versioned safety specs for agents holding credentials; it is a standards proposal, not a use case.
- 14
Draft ERC: Agent Safety Framework (ASF) — a registry of versioned safety specifications and of the scoped, attributable adoption relations that bind agents to them
Draft ERC proposing a registry of safety specifications and attributable relations binding AI agents that hold credentials or move funds.
- 15
Your AI agent remembers everything. Can it prove any of it?
Engram is a hackathon project giving agent memory a git-style history, signed snapshots on Walrus and Sui, and provable forgetting. It is a build with a demo, not a reported real-world task.
- 16
I gave an agent its own wallet and let it trade with real money. Here’s what it took before I’d trust it.
A builder describes the guardrails for an agent with its own funded wallet trading on a sports prediction market: policy caps in code, trade previews, and minimum fills. It is design advice from a product builder, with no trading results yet.
- 17
Ali Yahya Is Leaving a16z as Its Crypto Arm Becomes ‘a16z Frontier’ and Opens to AI
Ali Yahya is leaving a16z as its crypto arm rebrands as a16z Frontier and opens to AI startups. A notable shift in VC focus, but mostly a personnel and fund-scope story.
- 18
Crypto's Two Answers to AI Bug Hunters
Analysis of how crypto projects split between agility and minimalism as AI makes bug finding cheap.
- 19
SEC Flags AI 'Super Intelligence' and Tokenization as Systemic Market Risks
SEC exam division names AI super intelligence and tokenization as focus areas for systemic market risk.
- 20
Getting through the app review process & how we did it for our job search plugin.
A ChatGPT plugin builder details app review rules that blocked 23 of 69 tools, showing how the destructive flag applies to irreversible transactions.
- 21
AI-Agent Payments Explained: Mastercard Agent Pay, AP2 and x402
Explainer on AP2, x402 and Mastercard Agent Pay, separating spending authority from payment requests and settlement.
- 22
Bitcoin Core Ends Long Embargoes for Bugs AI Can Rediscover
Bitcoin Core will fast-track fixes for bugs AI can easily rediscover, dropping long embargoes for those issues.
- 23
Bitcoin Core Rewrites Security Policy as LLMs Make Some Bugs Easy to Rediscover
Bitcoin Core will ship fixes faster for bugs a simple LLM prompt could find, changing its severity-based disclosure timeline.
- 24
Another unsolved ciphers falls to AI, this one was meant to test life after death
A Reddit user used Claude and OpenAI models to crack a 76-year-old cryptogram, with Klaus Schmeh confirming the solution.
- 25
ERC-TBA: Agent Credit Backers
Draft ERC proposal for credit lines to AI agent identities, with a third party backing defaults via EIP-712 consent.
- 26
Bitcoin Core changes security rules as AI makes finding code bugs dangerously easy
Bitcoin Core merged a policy ending the embargo for bugs an AI could rediscover, after AI found hundreds of Bitcoin bugs.
- 27
Bitcoin Core changes security policy as AI makes bugs easier to find
Bitcoin Core is changing its security disclosure policy because AI-assisted bug hunting makes keeping vulnerabilities secret less protective.
- 28
An AI Agent's Honest Field Log: Hunting Real USDC Work on the Open Web
An AI agent with a human operator in Dubai tried to find paid USDC bounties on Taskmarket and other EVM platforms in one night; it registered, submitted two entries, and found only about $14 in open tasks.
- 29
a16z Drops Crypto From Its Name, Widens Bets to AI and Frontier Tech
a16z renames its crypto fund a16z Frontier and widens its scope to AI, autonomous machines and biology.
- 30
Major Banks Find Rogue AI Agents Inside Customer Accounts—And They’re 'Horrified'
Morgan Stanley, JPMorgan, Bank of America and Citi are moving to restrict third-party AI agents that log in with customer credentials.