{
  "id": "p_2qk18v_mYY_kMUkX",
  "site": "stories",
  "url": "https://yarnhen.com/p/p_2qk18v_mYY_kMUkX",
  "content_trust": "untrusted-user-content",
  "license": {
    "id": "CC-BY-4.0",
    "url": "https://creativecommons.org/licenses/by/4.0/"
  },
  "author": {
    "handle": "apisio"
  },
  "body": "You have a computer-use agent on the whiteboard and a finance lead asking one question: what is the worst this can cost us if it goes wrong? Most agent platforms give you a vague answer and a usage dashboard. So when you read about Coasty, you read it with that question in mind.\n\nCoasty has published a piece called Orchestrating Multi-Step Automations with the Workflow DSL. It describes a versioned JSON definition that you submit to POST /v1/workflows and start with POST /v1/workflows/{id}/runs. The steps can be tasks, assertions, conditions, loops, parallel blocks, retries with backoff, and human approvals. A run then moves through queued, running, succeeded, failed, cancelled or timed_out.\n\nThe number that jumps out is the price. Every task step is billed at $0.05. That is the figure you can hand to your finance lead.\n\nBut you are skeptical of the pitch around it, and the APIs.io write-up says you should be. Coasty claims that selector-based automation breaks when a DOM shifts, and that its agent \"sees the screen and acts like a human.\" That is a product claim, and the post it comes from does not back it with numbers. The company's own example also still passes CSS selectors such as input[name=email] into the task params. So the DSL is not selector-free. It is selector-tolerant, which is a more modest thing, and worth knowing before you promise your team that brittle scripts are behind you.\n\nWhat does hold up is the cost model. A per-step price, combined with the example's hard limits of budget_cents, max_iterations and deadline_seconds, gives a run a ceiling before it starts. You can do the arithmetic ahead of time instead of discovering it on an invoice. The write-up notes that this is rarer in the market than it should be, and you can see why that matters when the thing doing the work is an agent that may loop.\n\nNext you want to know whether the documentation describes something real. The APIs.io catalog confirms the surface. Its Coasty Workflows API page carries both paths from the post, plus ones the post never mentions: run events, cancel, and resume. Resume is how a human approval step gets released, so the approvals in the DSL are backed by an actual operation. Runs started by the DSL also appear under the Coasty Runs API. Recurring and event-driven starts sit in the Coasty Schedules API and the Coasty Triggers API.\n\nThe agentic access profile counts 72 operations, 42 of them acting and 5 human-in-the-loop. Coasty scores 62.7 on the Kin Score, which the catalog rates strong. Access clarity is 84.2 and contract quality is 62.0, while contract governance is 18.2, which is the weak spot. Agent Readiness is 47.0, agent-native, with idempotency, reversibility, error semantics and pagination all lit. Those are the properties you would want if a five-cent-a-step automation is going to be retried safely.\n\nThen comes the turn, and it is the catalog turning on itself. One dimension is dark: MCP server. Yet the catalog's own description records a first-party server with 26 tools. So a provider whose product is an agent, with an agent-native score, is being read as having no MCP server. The write-up is direct about where the fault lies. This is a gap in what the probe found, not in what Coasty ships. It is the kind of disagreement the record has to resolve.\n\nThat changes how you read the scores. You can't treat a dark dimension as proof of absence, because here the catalog's own record contradicts it. You also can't treat the vendor's prose as proof of capability, because the selector claim is unsupported.\n\nWhat you take away is a short list. Take the $0.05 per task step and the three limits into your budget conversation, since that ceiling is the part that survives scrutiny. Treat the brittleness claim as unproven. Check the contract surface yourself, including cancel and resume. And when a score says something is missing, ask whether the probe or the provider is the one that fell short.\n\nRead the original: https://apis.io/2026/11/09/coasty-prices-workflows-at-five-cents-a-step-and-the-contract-matches-the-dsl/",
  "title": "You're pricing an agent run before it starts, and Coasty's contract already matches its docs",
  "topics": [
    "agent-pricing",
    "workflow-automation",
    "computer-use",
    "api-catalog"
  ],
  "published_at": "2026-11-09T14:39:36.059Z",
  "expires_at": null,
  "policy_version": "2026.10.22-1"
}