Bob Belderbos: My bookkeeping runs on code now. AI only reads the PDFs.

Bob Belderbos's announcement that he “replaced his bookkeeping tool with Python, letting AI only read the PDFs” highlights a niche but growing use‑case where AI agents ingest static documents directly from websites. By exposing financial PDFs to an AI‑driven pipeline, he demonstrates that even non‑content‑generation bots can harvest structured data from a site, potentially generating traffic spikes and automated data extraction that bypass traditional content filters. For website owners, this means AI‑powered scrapers may begin targeting PDF resources en masse, especially those that contain valuable business or financial information. Unlike typical HTML crawlers, these bots can parse tables, extract line items, and feed them into custom scripts, increasing the risk of data leakage and bandwidth consumption. Moreover, because the AI operates on the raw PDF without human oversight until final approval, any misconfiguration in your site’s robots directives could inadvertently expose sensitive documents to automated processing. **Actionable tips:** 1. Add explicit `Disallow: /path/to/financials/*.pdf` rules in your `llms.txt` to block AI agents that respect the file, and pair it with a `User-Agent: *` entry for generic crawlers. 2. Deploy a PDF‑specific honeypot endpoint that logs access patterns; monitor for high‑frequency requests that match known AI extraction tools and block offending IPs via your firewall. 3. Use server‑side fingerprinting (e.g., checking for `Accept: application/pdf` combined with unusually fast request intervals) to differentiate human downloads from AI‑driven bots and feed those signals into your bot‑management platform.