Model prices dropped an order of magnitude. Tool prices didn't. The same web search still costs anywhere from $0 to $0.005 per call depending on who answers it, and for search-heavy agents that spread is most of the bill.
These posts put numbers on the problem: rack-rate comparisons, cost observability, retry overhead, and the unit economics of shipping a product where every user action fans out into tool calls.
Not a dashboard project. Eight numbers you can compute from a call log in an afternoon, and the specific fix each bad number points at.
Dashboards tell you what the month cost. A receipt tells you what this run cost, which provider earned it, and what the same result would have cost at rack rate.
Every retry is a repurchase at full price with worse odds than the first attempt. Here's the playbook: budgets per task, hard no-retry rules, and fallbacks.
One click becomes forty API calls before anyone looks up. The budgeting discipline that survives that fan-out, worked through with real search rates.
Inference got an order of magnitude cheaper and everyone noticed. The per-call price of search, scrape, and browse barely moved, and almost nobody did.
Everyone budgets for tokens. Then a teardown shows search at 54% of task cost. The math behind why research agents multiply search spend.
Your invoice knows what you spent by provider and month. It has no idea what one agent run cost. Stamp the cost on the response envelope and the question answers itself.
No replatforming, no procurement cycle. Seven changes you can ship by Friday that attack the parts of an agent bill nobody is looking at.
Five providers, one capability, a price range from free to half a cent. An analysis of what you're actually paying for when your agent searches the web.