Free tool · for anyone running AI agents
What are your AI agents really spending on?
Scheduled routines usually drive agent usage more than chat does. In this example, two fast timers make up 92% of an estimated $890 a month in work, against a $100 plan. Slower schedules, quiet hours and a cheap “anything new?” check would cut that to about $82. Plan prices are verified. Per-run costs are your own estimate.
The formula
Runs times cost per run, added up
runs a month (timer) = runs per hour × 24 × 365 ÷ 12 around the clock
= runs per hour × 12 × 5 × 52 ÷ 12 working hours
runs a month (calendar) = 30.4 daily, 21.7 weekdays, 4.33 weekly
routine cost = runs × (cost per run + subagents × medium cost, heavy only)
sessions = sessions per workday × 21.7 × session cost
+ heavy sessions per week × 4.33 × (heavy cost + subagents × medium cost)
work vs plan price = (routines + sessions) ÷ what your plans cost
pre-check saving = runs × no-change share × (cost per run − light cost)
event-trigger saving = timer routine cost × no-change share
suggested setup = timers at most hourly, working hours, one effort lighter,
pre-check on no-change runs, same-schedule checks mergedEvery dollar here is the cost of the work at the per-run prices you enter, close to what it would cost at API rates. Subscription plans include an allowance instead, so read “work vs plan price” as a pressure gauge, not a bill.
Worked example
One person, one plan, five routines
- An inbox check every 5 minutes, around the clock, runs 8,760 times a month. At $0.04 a light run that is $350, and 90% of those runs find nothing new.
- A leads watch every 15 minutes at medium effort runs 2,920 times, or $467.
- A morning briefing, a research digest with two subagents and a weekly report add $39. Six medium sessions a workday and three heavy ones a week add $33.
- Total: $890 of work against a $100 plan, about 8.9× the plan price. At that ratio, running into usage limits is likely.
- Move both timers to hourly during working hours, drop them one effort level, add a pre-check and merge them into one run. The same work comes to about $82, and the briefing, digest and report don’t change.
Starting values
Where every default comes from
| Input | Default | Basis |
|---|---|---|
| Claude Pro / Max / Team | $20 · from $100 · $25 or $125 a seat | SourceAnthropic Monthly list prices. Pro is $17 a month billed annually, Team Standard $20 and Premium $100 billed annually. Max is “from $100” with 5x or 20x Pro’s usage, so enter your own price for 20x. |
| Claude usage limits | No fixed number | SourceAnthropic Limits reset on a rolling five-hour window, paid plans add weekly limits, and there is no fixed message count. Usage credits at standard API rates are available on paid plans. |
| Cursor Pro / Pro+ / Ultra / Teams | $20 · $60 · $200 · $40 a user | SourceCursor Monthly list prices. Cursor’s Pro plan lists Grok Bot access. |
| Grok Bot | No separate price | SourcexAI docs Included with paid individual Cursor plans and Cursor Teams, or a linked SuperGrok, SuperGrok Plus or SuperGrok Heavy subscription. Weekly usage, with optional on-demand usage billed from model and token cost. |
| ChatGPT, SuperGrok and other plans | Enter your price | Exampleyours We couldn’t confirm these on an official pricing page today. |
| Cost per light / medium / heavy run | $0.04 · $0.16 · $0.65 | Workahead estimateyour estimate Token counts are Workahead guesses (15k/1k, 60k/4k, 250k/15k in/out), priced at Anthropic’s published Sonnet 5.5 API rates of $2 and $10 per million tokens. Replace with your own. |
| Routines, sessions, subagents, no-change shares | See form | Exampleuse yours A one-person setup with two fast timers. Use your own schedule. |
Your own numbers
Find them in your settings
List every scheduled routine with its timer. Most agent apps show them in a routines, tasks or schedules page. Then look at the last week of runs and count how many ended with “nothing new” or “no change.” That share is the biggest lever in most setups.
For cost per run, note your usage meter, run one routine by hand, and check it again. Do that once each for a light, medium and heavy routine. On Claude, the meter is in Settings > Usage.
If you’d rather someone else did this, the AI Agent Usage & Cost Audit works from your screenshots and exports and returns a written fix list in about three business days.
Questions
Before you trust the number
Is this my actual bill?
No. It prices the work at the per-run costs you enter, which is roughly what it would cost at API rates. Flat subscription plans don’t bill that way. They include a usage allowance and stop or switch to paid extra usage when you hit it. Use the estimate to see which work uses the most, and use the optional last-month bill to compare with what you actually paid.
Where do the per-run costs come from?
They are your estimate. The starting values are Workahead examples: a light run of about 15,000 input and 1,000 output tokens, a medium run of about 60,000 and 4,000, and a heavy run of about 250,000 and 15,000, priced at Anthropic’s published Claude Sonnet 5.5 API rates of $2 per million input tokens and $10 per million output tokens. Real runs vary a lot with tools, browsing, files and the model. The best number is your own: run one routine once and check how much your usage page moved.
What are the usage limits on Claude plans?
Anthropic says every plan has limits that reset on a rolling five-hour window, paid plans add weekly limits, and there is no fixed message count because usage depends on conversation length, model and features. Max gives 5x or 20x Pro’s usage per five-hour session. Team Premium seats give 5x a Standard seat. When you hit a limit you can wait, move up a plan, or turn on usage credits at standard API rates. Claude Code shares the same pool. Checked October 11, 2026.
How does Grok Bot usage work?
xAI’s Grok Bot FAQ says Grok Bot is included with every paid individual Cursor plan and the Cursor Teams plan, or with a linked SuperGrok, SuperGrok Plus or SuperGrok Heavy subscription. Plans include weekly usage, and eligible accounts can add on-demand usage billed from model and token cost. Routines run on a schedule or, where supported, after an event. Cursor lists Pro at $20 a month, Pro+ at $60, Ultra at $200 and Teams at $40 per user.
Why don’t you list ChatGPT or SuperGrok prices?
Because we couldn’t confirm them on an official pricing page today. Prices there vary by region and change often. Choose “Another plan” and type in what you pay.
Why does a 5-minute timer cost so much?
Because it runs 8,760 times a month around the clock, and most of those runs find nothing new. Even at $0.04 a run that is $350. The same check every hour during working hours runs about 260 times.
Can I add the savings levers together?
Not one by one, because they overlap. A slower timer has fewer no-change runs left to stop. The suggested setup applies them in order: hourly at most, working hours only, one effort level lighter, a cheap pre-check, then merging checks on the same schedule. It leaves out event triggers because not every tool supports them.
What do subagents add?
Each subagent is extra work running alongside the main one. The calculator charges each subagent as one medium run, on heavy routines and heavy sessions only. If your agent fans out on smaller tasks too, raise the medium cost or the subagent count.
Does anything I type get stored?
No. The math runs in your browser. Routine names stay on your screen and are not put into share links. Analytics records that the tool was used, never the numbers.
If the numbers surprise you
Get a written fix list for your own setup.
The $295 AI Agent Usage & Cost Audit reviews your routines, settings and usage exports, then tells you what to cut, slow down or merge, with a 30-day plan. No passwords needed.
Workahead is independent and not affiliated with Anthropic, Google, Invoca or BrightLocal. Results are planning estimates built from the numbers you enter, not measured results for your business.