Human–AI Operating Model · Token Planner
Describe the work. Get the token budget.
Say who does what, how often, how urgent it is and how much freedom the AI gets, or paste a Markdown document such as a work inventory or a statement of work. The agent matches the work to the delivery catalogue and returns a governed monthly token budget for the people doing it, and the human time each accepted item still needs.
Token Planner agent
Describe the work, or upload a document
A sentence gives one portfolio. A Markdown document is divided into one line of work per section and one per kind of work a section names. What you type is read by the Token Agent service in AWS and is not stored by the page.
Try:
Upload a Markdown document
drag a .md file here, or
drag a .md file here, or
read by the Token Agent service, not stored · one line of work per section · the total is the sum of the lines
Token Agent service · local reader
Agent settings — local reader or Claude
Either way, the figures are the model's. Claude is used only to read the text against the catalogue; its answer is checked and then priced exactly as the local reader's would be. The reading runs in the Token Agent service in AWS, which holds its own access to the model, so there is no key to type here. This choice is kept in this browser.
Settings
Set the parameters before you estimate
These are used for every estimate and kept in this browser. Where the text you give states a figure, such as a team size, an urgency or a first-pass yield, the text wins.
The people
leave blank to take it from the text; one person if the text does not say
leave blank to use the rate of the role the agent reads
The AI
a yield stated in the text wins
Defaults for every line of work
used where the text does not say
never above the ceiling governance allows for that kind of work
How demanding is the task?
How does this person work with AI on it?
Saved in this browser as you change them.
The answer