Skip to content
Job Umbrella

Dev blog

What shipped, and the problem it solves

One post every Friday. Each one starts with a problem a job seeker actually has, explains how it was built, and ends with the release notes for that week.

Who is writing this

I’m Aaron. I spent ten years as a consultant, which meant the job search never actually stopped — every engagement had an end date, and the next one had to be found while the current one was still running. I have worked at Amazon, Expedia, T-Mobile, AT&T and Microsoft. Over those years I built a process for it: a way to track every application, tailor a resume to a specific posting without inventing anything, and know which threads were going cold. I did it by hand, in spreadsheets and documents, for a decade.

Job Umbrella is that process, automated. I started building it in July 2026 after reading one too many posts from people who had applied to several hundred jobs — some of them a thousand — before getting an offer they could accept. The same complaints came up every time: listings that were never really open, matches that had nothing to do with the person, and applications that vanished without a reply.

One more thing pushed me into building it. A lot of sites will happily rewrite your resume for a specific job. Not one of them looks at your resume first to see whether it is any good. That is garbage in, garbage out with a subscription attached. A tool that rewrites a weak resume gives you a weak resume in a new font.

The one promise this site is built on

Nobody can pay to get to the top of your list. Paid job promotion is the obvious way to make money here, and it is the reason every big job board ranks the way it does. It also puts the site on both sides of the market at once: serving the person looking for work, and serving the employer paying to be seen. Those two interests are not the same, and when they conflict, the money decides.

I have chosen to remove the conflict by refusing the revenue. It was specified, costed, and then killed in September 2026 for exactly this reason. What pays for the site instead is written up in How a free job site pays for itself without selling you.

Posts

Two things about this blog

The early posts are backdated. The blog itself started in September 2026. The posts before that date were written from the project’s own release ledger and its commit history, and each one is dated to the Friday of the week it describes. Every commit hash, count and date in them is real and checkable. Nothing here claims to have been published at the time.

The code is written by AI agents, directed by me. I design the work, decide what ships, and run every change through a review pass where other agents are told to break it rather than approve it. That process finds real defects — one week in September it produced about a hundred findings before anything merged, and three separate security fixes were beaten by the reviewer set on them and had to be rebuilt. Where that method shaped a release, the post says so.