The latest experiment in agent-native infrastructure comes from OpenJob, a public job database that refuses to build its own agent. Instead, it provides a standardized llms.txt file that instructs existing tools like Cursor, ChatGPT/Codex, and Hermes to handle job searches, postings, and profile management. The maintainer, Bill Zuo, emphasizes a zero-friction approach: no new apps, no accounts, and no tracking. It’s a pragmatic shift from UI-heavy platforms to data-driven prompts that work within the developer’s current workflow.

The Honest State of the Index

Before diving into the prompts, the project discloses its actual capacity. As of November 21, 2026, the index contains 33,437 rows, but only about 14,000 were posted in the last 60 days. The prompts explicitly instruct agents to filter for this freshness to avoid stale data. Crucially, the system does not automate the application process itself; it returns the employer's own source_url for applications. This distinction is vital for builders to understand: OpenJob is a discovery layer, not an application proxy.

Three Prompts, One Interface

The entire user experience is condensed into three copy-pasteable instructions. To find a job, users paste a prompt that directs the agent to read the llms.txt file and search the dataset for roles matching specific criteria like location and remote status. For hiring managers, a second prompt drafts a job post and opens a pull request to the softprobe/openjob GitHub repository. These posts are Markdown files that auto-merge if validated, ensuring the board remains clean. A third prompt converts a resume into a talent profile via API or remote MCP, requiring explicit user approval before publishing.

Data Governance and Expiry

OpenJob enforces strict data hygiene to maintain utility. Job posts expire after 60 days, preventing the accumulation of outdated listings. Talent profiles are licensed under CC BY 4.0 and expire after 180 days, with a hard rule against including personal contact information like emails or phone numbers. Users receive a private deletion token upon profile creation, which is shown only once. This design choice minimizes PII exposure while leveraging open licensing for broader agent accessibility.

Key Takeaways

  • OpenJob relies on llms.txt instructions rather than custom agents, integrating with tools like Cursor and ChatGPT.
  • The database contains ~14,000 active listings out of 33,437 total rows, with strict 60-day expiry for job posts.
  • Job applications are not automated; the agent provides the employer's direct source_url for users to apply manually.
  • Talent profiles are public under CC BY 4.0, expire in 180 days, and strictly exclude contact information.
  • Hiring posts are managed via GitHub pull requests, with auto-merging enabled for validated Markdown files.

The Bottom Line

OpenJob proves that agent-native infrastructure doesn't require a new app ecosystem; it just needs clean data and clear instructions. By leveraging llms.txt to delegate work to existing agents, it offers a frictionless alternative to bloated job platforms, though its reliance on manual application links limits its automation potential.