I let an AI agent run my job search: the real numbers, including the LLM bill (find-best-job.com)

A backend developer in São Paulo has developed an AI agent to automate the job search process, aimed at simplifying the arduous tasks of finding and applying for remote roles in USD. Over four days, the agent discovered 360 roles from various sources, applied to nine, and received two recruiter replies, revealing a significant insight: the agent's ability to filter low-fit roles—87% were rejected—ensures that applications are both strategic and respectful of recruiters’ time. This stands in contrast to existing auto-apply tools that generate high volumes of applications without consideration for fit. However, the project faced challenges, particularly with associated costs linked to using large language models (LLMs). Each application cost an average of $3.36 under a premium model, highlighting the need for efficiency in model selection and data handling to make the service viable. The developer concluded that better results are achievable with a more economical approach, controlling costs while maintaining application quality. The agent uses a real Chromium browser to navigate ATS forms, avoiding detection, and adheres to constraints like limiting applications to five per day to prevent spamming. Moving forward, the developer plans to refine the agent into a product called find-best-job.com, promoting ethical automation in job searching.