How your agent decides which jobs are worth a proposal
Searching is the easy part. The value is in deciding which jobs to skip. A look at how scoring and filters work together.
An agent that applies to everything it finds is not saving you time. It is spending your Connects. The useful part of an agent is judgement: knowing which jobs deserve a proposal and which to leave alone.
Step one: filters remove the obvious misses
Before any AI is involved, filters cut jobs that fail simple rules: payment not verified, a client who has barely spent anything, a hire rate below your minimum, or a budget under your floor. These checks are cheap and predictable.
Step two: the model judges fit
For the jobs that remain, the model reads the post alongside your background and scores how well the work matches your skills, rate and preferred type of project. A score gives you something to compare and a threshold to set.
Step three: limits decide how many
Even good matches are capped by per-run and per-day limits, so an agent never floods Upwork with proposals on a busy morning.
Why it explains itself
A score without a reason is hard to trust. The agent records why it kept or skipped a job, and you can read that in the run log. If it keeps skipping jobs you wanted, change the keywords, filters or instructions rather than the threshold.
What to tune first
- Your background text: be specific about what you do and do not take.
- Filters: raise or lower the client-quality bars.
- The fit threshold: only after the first two are right.