Blog

• Closing the books on stablecoin payout bots: the CSV your accountant will accept

  9/20/2026

  Most writing about agent spend is about runaway loops and 3am bills. This one is about a duller problem that costs more hours: it is the end of the month, your payout bots have sent a few hundred USDC transfers across Base and Solana, and someone has to turn that into rows a bookkeeper will accept. If you have done this by hand, you already know where it goes wrong. Here is the shape of it, and what to fix first.

• Why we only ever ask for a public address

  9/19/2026

  AgentCap never asks for a seed phrase, a private key, or a wallet connection — because none of those are needed to read outbound spend. A public address is enough, and it's the only thing we'll ever ask for.

• How to set a spend budget for an AI agent wallet without touching its keys

  9/19/2026

  You set a budget by telling AgentCap a ceiling and a public address — nothing about the agent's keys, signing logic, or wallet software changes. The budget is a read-side alert, not a spending limit the chain enforces, and that gap is the thing to understand before you rely on it.

• What x402 payments look like on-chain, and what they don't tell you

  9/19/2026

  An x402 payment settles as an ordinary USDC transfer — same Transfer event, same fields, same lack of a memo — the x402 handshake that negotiated the price and endpoint happens entirely off-chain, over HTTP, before the transaction is ever broadcast.

• Three ways an autonomous agent quietly overspends, and how we caught each one

  9/19/2026

  Every overspend pattern we've seen so far traces back to one of three causes: a retry loop that re-pays for the same resource, a budget set once and never revisited as usage grew, or a shared wallet where one agent's spend gets misread as another's — and only the third one is something our attribution model can't fully catch on its own.

• Why we don't ask you to connect a wallet (and what we lose by not doing it)

  9/19/2026

  Wallet-connect would get us a few things a public address can't: an authenticated identity check, in-app signing for a future auto-upgrade flow, maybe a smoother "add this agent's wallet" onboarding step. We've decided those wins aren't worth the attack surface for a tool whose entire job is watching money, and this post is the honest accounting of what we're giving up.

• Reconciling agent USDC spend against an on-chain balance: our method and its gaps

  9/19/2026

  We reconcile by summing every outbound Transfer we've ingested for a wallet and comparing that running total against the wallet's current on-chain USDC balance plus inbound transfers — if the arithmetic doesn't close, either our ingest missed something or a transfer happened on a rail we don't watch yet.

• Base vs Solana for agent fleets: throughput, finality, and what breaks at 10 wallets

  9/19/2026

  At ten tracked wallets making frequent small payments, the practical difference between Base and Solana shows up less in raw throughput — both handle that volume trivially — and more in finality time and RPC rate limits, both of which affect how fast our ingest can catch up after a burst.

• What a $0.04 agent payment actually costs on Base vs Solana

  9/17/2026

  The two transfers we use in our own demo data moved $0.10 and $0.04 in USDC on Base. Base's gas at the time of those transactions ran a few cents in ETH — meaning the fee on the smaller payment was in the same order of magnitude as the payment itself. That ratio is the actual number to watch when you're routing high-frequency agent spend.

• Attributing USDC transfers to agents when the chain doesn't tell you who paid

  9/12/2026

  A USDC Transfer event carries a from address, a to address, and an amount — nothing that says which of your agents sent it. We attribute a transfer to an agent by matching the event's from address against the wallet you registered for that agent, and that match is only as good as your one-wallet-per-agent discipline.