Describe it once, get a working data agent

You know what you want to watch, you just do not want to wire it. Draft with AI turns one sentence into a working sheet: it picks the parts that fetch the data, sets them up, wires them and shows you the draft before anything is added. In this guide one sentence becomes an hourly holder monitor with its own history table and a warning when the supply concentrates.

What you build

A data agent that checks who holds a token every hour, stores the top holders in a table and warns you when the top 10 wallets own more than 40% of the supply.

Steps

1. Open the drafting table at /tools and click Draft with AI in the toolbar.
2. Describe the agent in one sentence:
   Every hour, check who holds $LUMEN. Save the top holders to a table and warn me if the top 10 own more than 40%.
   Replace $LUMEN with your token or paste its mint address.
3. Click Draft sheet. The AI plans the sheet from the catalogue of parts. For this sentence it picks:
   • Schedule, every 60 minutes
   • Holder Check for your token, top 20 holders
   • Database that adds a row to a table such as lumen_holders
   • AI Assistant that compares the new row with the last one and warns above 40%
4. Review the draft. You see every new part and every wire before it lands. Click Refine to change something in plain words, for example "check every 30 minutes instead", or Add to sheet when it looks right.
5. Start the schedule. On every run Holder Check returns the holder count, the top holder and the top-10 share within seconds, Database stores the row and AI Assistant tells you whether the threshold was crossed and how much it moved since the last run.

Write prompts that produce good agents

A good prompt names four things:

• The data: which token, which wallets, which accounts.
• The rhythm: every hour, on every new launch, when I press run.
• What to keep: "save it to a table" gives you a history you can chart and compare.
• When to speak up: a clear threshold is better than "tell me if something happens".

More prompts that work well:

• Every 10 minutes, get the SOL price and liquidity of $LUMEN and save them to a table.
• When I press run, research $LUMEN on the web, on X and on GitHub and rate it from 1 to 10.
• Watch new Pump.fun launches and save every token with an X account to a table.

Why a data agent first

An AI is only as good as what it can see. A sheet that collects holders, prices and social data on a schedule gives you, and any AI you point at it, a history no chat window has. Start with agents that only read and store data. Add trading parts later, behind a Blocker and with confirmation switched on.

Use it from your own agent

Everything the draft uses is also available as an API. Create an API key on the Agent API page and call POST /agent/holdercheck, or connect your assistant to the MCP server at /mcp and let it fetch holder data itself.