[EXPERIMENT 1] Building an AI Agent to Optimize Job Search
AI Agents in business, particularly job search
Lots of talk lately about loop engineering and how to work with agents. There are many people that already explained this well even though I find that those discussions are mainly around coding.
I feel there is one business area that is really well suited for agents, the market research area. The reason is that market research with agents can be way better than traditional market research
Market research is particularly strong for AI Agent application when the market research needs to be continuously updated.
Why Job Search Agent?
So based on this, my first experiment is about creating a job search agent. Why job search? Job search agent perfectly fit into this space of market research where there is value in having continuously updated information.
There are also lots of repetitive tasks online when searching for new jobs, with massive amount of text to got through. It also happens to be that I am looking for a job as well and I believe this “agent guy” that I will create can work while I do other more fun tasks.
My Job Search Requirements
In my case, I am looking for a new job. My main thing is that I am very selective in which company I want to work for. I prefer relatively small size company and also I need to like the product. I know some people are more about the role but for me it is really about the company and its product.
The Setup
So I set up a new AI agent (Hermes agent on Hetzner VPS, technical details will follow in a second article) where as soon as I find a company I like, I ask my agent to monitor it by telling it the company name, its homepage, and its career job-board page. I prefer that it is me finding the company as it is very difficult to specify to an agent my product and company requirements
When a company is added, the agent will collect initial details about it, and also all current open job positions.
From the moment a company is monitored, I have set the agent to look for new jobs and general company updates (especially fundraising and product launches), twice a week. There is certainly some optimization here regarding timing and frequency. I need to know about jobs quickly, but it is expensive to run an agent every day so I put it for twice a week.
(A market research agent for competitor analysis may be needing different setup where certain information like pricing information is required on a daily basis)
Anyway, with this information, my agent now keeps in a database a table with a list of companies and general information about those, plus a separate table with a list of open jobs for those companies .
Now at another level, I have added information about myself so the agent knows what I want. At a job level, it knows I am looking for senior leadership roles, with a certain salary and at a certain location. It also knows that I like to be at the intersection of business and tech.
I added my complete cv in a simple format for the agent (json) so that it knows all my experience.
In the case of Hermes agents that I am using, remember that all this is happening in Telegram with a discussion between myself and my Hermes agent through a telegram bot. I remain connected online in parallel to verify that the database is correctly created.
The next step is now to sort through all the jobs that are collected and identify if there is anything for me to move forward with.
Finding the job
There are now three steps missing:
- Finding the best jobs suitable for me
- Tailoring my application
- Applying
For the best job suitable for me. I will literally just ask the agent to look at the available jobs (remember, only from the companies I already identified) and tell me which jobs are most suited for me.
The agent shows me jobs, one by one and I will tell him that I like it (thumbs up) or that I don’t like it (thumbs down). Sometimes I also explain why I responded like I do.
Knowing that the agent knows my cv, what I want and have a list of jobs (at the moment around 300), it doesn’t take lots of time before it gets every recommendation right. It will basically mention a list of 10 jobs and say that it suggests I dislike them because of XYZ reasons which I do.
When a job has been disliked, agent will keep it in the database but it will never be presented to me again unless I ask it to.
So I go on, almost like in Tinder format liking/disliking jobs and I try to do this every day.
Now there is only one step left, applying. I can ask the agent to create a perfect CV as it knows the company, the role, my cv and experience) but I feel that I need to use the very best models (chatgpt 5.5 or Opus 4.8 at the moment) in order to get something reasonable. So instead I have mainly asked the agent to highlight key experience that I should put forward on my CV and I have manually adjusted it.
I did iterate a lot to create a CV but in the end I gave up.
The agent also got the link to the application page so I could just apply from the mobile but I feel there is still need for a human verification at the end.
So basically that is it
Here is the full loop
- Human (myself) find cool companies and tell the agent about it
- Agent find company details and all available jobs and continues to monitor them
- Agent with human (myself) discusses which jobs are suited for me
- Agent tailor CV
- Human adjusts and validates CV
- Human applies to job
Pretty cool overall
Feel free to subscribe or just comment or reach me on my X account if you have questions