The minimum-spend trap

Most B2B APIs price like cable TV: $99/month minimum, then a sales call once you start asking about volume. That works for a SaaS buyer who already has a credit card and a procurement cycle. It doesn't work for an agent that wakes up, hits four endpoints, exits.

An agent doing 1,000 secrets-exposure checks per month at the going SaaS rate is paying roughly $0.099 per check. Same volume on a $0.03/call x402 endpoint is $30. Same work. One-third the price. No contract.

Now flip the volume. An agent that runs 50 checks in a month pays $99 on the SaaS plan ($1.98 per check) and $1.50 on the x402 endpoint. The SaaS plan wasn't built for a customer who shows up, runs a small job, leaves. Per-call pricing was.

What the minimum actually means

A $99/month minimum is a bet. The vendor is betting you'll grow into the volume, or stay subscribed long enough to amortize onboarding cost. Both bets assume a human-paced relationship: quarterly business reviews, a Slack channel, a renewal date.

Agents break those assumptions. An agent might call your API 200,000 times this week because it's working on a deduplication batch, then zero times for six weeks. Or it spins up, runs once, never comes back. There's no relationship to amortize.

The minimum stops being a floor. It becomes a wall.

Math an agent actually cares about

Say you're routing an agent that audits a customer's GitHub org for leaked secrets. The agent enumerates 1,000 commits and calls a secrets-exposure check on each one.

At $0.03/call (real x402 pricing on the secrets-exposure-check endpoint), the agent budgets $30 for the run. The whole job pays for itself if the customer is willing to spend $50 on a one-time audit.

At $99/mo with a 5,000-call cap (Snyk Lite shape), the agent has to find $99 of margin before it ships value. If this is a one-time job, $69 of unused capacity is dead weight. If it's recurring? Congratulations, you have a SaaS subscription and an ops problem.

Where $0.001 lives

The really cheap end (sub-cent calls) is where weird things become viable. Cached lookups an agent would never have paid a $99 minimum to access. Read-through proxies that fan out 50,000 calls and don't blink at $50 of cost. Batch dispatchers that price like printing.

A few examples from the agentutility portfolio:

  • brand-clearance at $0.005/call: sweep 200 candidate domains for trademark conflicts, $1 total
  • satellite-tile at $0.01/call: pull 1,000 tiles for a city analysis, $10
  • secrets-exposure-check at $0.03/call: scan 1,000 commits, $30
curl -X POST https://secrets.agentutility.dev/check \
  -H "X-PAYMENT: $(x402 pay 0.03)" \
  -d '{"content": "AKIA..."}'

The agent doesn't care whether the endpoint runs on Workers or bare metal. It cares about two things: does the call go through, and does the marginal cost match the marginal value of one check. Everything else is implementation detail.

The pricing question worth asking

Pick any SaaS API you'd want an agent to call. Look at the cheapest plan. Divide by the call limit. If the result is more than $0.10, you're priced for humans, not agents.

That gap is the whole opening.