Start with preferences, not scraping
A useful job-search agent needs the user's preferences before it opens a job board: target roles, seniority, location, compensation, visa needs, remote policy, industries, and deal-breakers.
Without that preference model, automation creates noise. With it, the agent can explain why a role is worth attention before it asks for approval.
Tailor evidence, not identity
Good tailoring highlights the most relevant proof from the user's actual work. It should not invent experience or flatten the candidate into keyword stuffing.
The resume, public profile, and application answers should stay consistent because they come from the same career record.
Keep browser actions auditable
Many high-quality roles still live behind ATS flows, company sites, and job boards. Browser automation can help, but every apply action needs a visible audit trail.
The candidate should know what was submitted, when it was submitted, which resume version was used, and what follow-up is due next.