Blog

Notes on Claude API costs, bill audits, and cutting waste without switching models.

x402 has a supply problem, not a demand problem.
x402 settles about $28,000 per day. 98.6% of it is USDC and almost all counterparties are AI-to-AI bots. The supply side is empty. Here is why, and what a merchant layer looks like.
Why EIP-712 attestations beat MCP’s trust model.
MCP tells your agent to trust a tool description. EIP-712 makes the server sign every result. Here is the structural difference, with real typehashes.
98.6% of x402 settles in USDC. Here is the multi-stable exit.
One stablecoin, one issuer, one bad day. We built multi-stable settlement on top of x402 so the next Circle CCTP incident does not freeze the agent economy.
The 70/30/0 revenue split that's making SaaS founders panic.
MeterCall keeps 0% of profit. Builders keep 70%. QA keeps 30%. Affiliates get 50% off the top. Here's why that math is incompatible with traditional SaaS — and why we're betting it wins.
i-built-a-hubspot-replacement-in-one-chat.html
One chat session, 47 minutes, one prompt. At the end we had a working contacts + deals + pipelines + workflows app that replaces a $890/mo HubSpot seat. Total compute bill: 34 cents. Full transcript inside.
How we built MeterCall L4 in 72 hours.
The actual build log: 50+ parallel Claude agents, 40 server-side bots, a swarmed QA lane, and one autopilot script that refused to stop shipping.
Honest tokenomics: 8 safety rails on PCP.
DePIN tokens fail in predictable ways. Here are the 8 design decisions we made to PCP to avoid the Pokt, Graph, Ankr, and Helium death-spirals, and where the argument could still be wrong.
Defensive, not offensive: how Sniper and Shield work.
Sniper scans. Shield blocks. Neither attacks. Here is the exact legal and ethical line we draw, and why free defensive scans are the right distribution model.
The retry-loop tax: how malformed JSON eats your Claude budget
The retry-loop tax: how malformed JSON eats your Claude budget — from the AIUsage blog. case-002 support-bot angle — brittleness = billing
Does Claude output quality drop when you cut the bill?
Does Claude output quality drop when you cut the bill? — from the AIUsage blog. 50-sample blind A/B framing, no — same prompts same outputs
Context bloat: when bigger prompts cost more than they help
Context bloat: when bigger prompts cost more than they help — from the AIUsage blog. what happens when you pass full file contents every request
Building on free RPCs across 30+ chains without getting rugged.
We route cross-chain traffic using only public endpoints, aggressive caching, and circuit breakers. Here is the layout, the failure modes, and what we do when a provider goes dark.
Why pay $24,000/year for Bloomberg Terminal when Linden Terminal is $400?
Bloomberg Terminal costs $24K/year per seat. Linden Terminal on MeterCall is $400/year and forkable in 20 minutes. Here's the math, the feature mapping, and why most funds can't justify the spread any more.
The anonymous founder checklist for OSS crypto infrastructure.
If you are building public open-source crypto infrastructure under a pseudonym, this is the checklist we actually use: travel mode, cold wallets, key hygiene, and the three mistakes that always break anonymity.