# Today’s Job Market Runs on Automation — So I Built a System to Stay Deliberate In It > Source: > Published: 2026-08-05 09:08:36+00:00 While AI has made individual job-search tasks dramatically easier, what I was missing was a way to keep the decisions, evidence, learning and relationships coherent across a search that lasted months — so I builtCareer-Campaign, a human-governed, AI-assisted system that does exactly that. In early 2025 I found myself out of a job, looking for the next one in a market where not only are available roles themselves scarce but also where companies are running their hiring process with ever more automation & AI systems. Your application is parsed, scored, and ranked by software before a person ever opens it — and increasingly it’s filtered out before a person opens it at all, frustratingly for reasons you will likely never get to see. You can do everything right and simply vanish into a system that never explains itself. Of course there is the traditional route of working with career search coaches and recruiters. However one approach that has also become quite popular among job seekers is to fight automation with MORE automation. [Job applicants are winning the AI arms race against recruiters](https://www.economist.com/business/2026/01/12/job-applicants-are-winning-the-ai-arms-race-against-recruiters) There are a number of both paid services and open source tools that can harvest job posting info from multiple sites , fire off applications at scale, tailor a resume to a keyword list in one click and apply to two hundred roles a week on your behalf. Quite honestly — it has become an arms race of sorts because to deal with the flood of AI applications, employers are just applying stricter filtering. I went a different way. I built a system to run my search — NOT a bot to run my applications for me. This article is what I learned about the difference, and the one thing no system can do for you. If you want to skip ahead, here’s the repo that I co-developed with Claude : Like most people, I started with the obvious tools. Plain chatbot conversations came first — ChatGPT. For a single, well-defined task, they’re excellent: *tighten this paragraph, critique this résumé against this job description, draft three versions of this message.* It’s a once-off and well bounded task. The trouble is a job search isn’t once-off. It may run for multiple months, across dozens of roles at different stages, and a fresh chat remembers none of it. Every session starts from zero. You spend the first ten minutes re-explaining who you are and what you’re looking for. I “graduated” to ChatGPT Projects, which let me anchor a workspace on fixed references — resume templates, my search preferences, the kind of role I was chasing. While this was an improvement, two things still broke down: Long conversations drifted where the further a thread ran, the more the model lost the thread, contradicting things it had agreed earlier. ChatGPT also couldn’t progressively update the files it worked from — the reference material sat there frozen, while my actual search moved on week by week. The knowledge base and new info from the live search would often fall out of sync unless I was very disciplined in manually keeping them current. At the same time , I also tried a few point solutions — specialist services that each solve one slice of the problem such as : ATS resume drafting and screening tools and even LinkedIn’s own AI tools (NB For premium paid subscribers only). Some of these were genuinely useful but they all handed the output back to me for me to integrate : carry the resume insight into the cover letter, remember what I’d learned about this company when I spoke to such and such person. The stitching was mostly manual or in custom spreadsheets and tedious work. That was a real gap. Not any single task — the *continuity* across all of them. No memory, no single source of truth, and me doing the integration by hand. There are already a few good tools for running a more systematic job searches with both “memory” and the ability to self-orchestrate it’s own workflow. So I’d be remiss if I didn’t at least mention one of the most developed open-source examples - [Career Ops](https://santifer.io/career-ops-system), which combines role evaluation, company research, tailored CV generation, pipeline tracking and other parts of the search into a multi-AI agentic assisted workflow. ** However I still ended up building my own, because the problem I was trying to solve was slightly different**. I wasn’t primarily trying to automate more of the search. I wanted a system that would preserve the reasoning behind it: what I was targeting, why I was pursuing or dropping a role, what evidence supported the claims I was making, which channels were actually producing conversations, and what the accumulated record suggested I should do differently next. I also wanted a search I couldn’t lie to myself about. A chatbot is helpful ,forgetful, and agreeable by design - it will cheerfully tell you a mediocre role is worth a shot and remember none of it next week. I wanted the reverse: files that I own, a durable history of what I did, how it went, etc that I could put in front of a career coach and say *here is exactly what is happening.* So I built the system around the files, and used AI as the fast, disciplined worker *inside* it — never as the place the search lived. The system is essentially a folder of files. Inside: a versioned Markdown tracker, two CSVs (applications and outreach), a per-company research directory, a single-file HTML dashboard, and various instruction files that tell the AI how to behave in every session. Under the hood the files are divided into: You can customize how you want to work with it where during the initial set-up , it will ask you up front how hard you’re actually searching — full-time between roles, a deliberate hour a week while you’re employed, or a monthly check-in when you’re only open to the right thing — and creates a working rhythm to match. Each working session can be customizable as it runs on a set of rails based on the skills I mentioned earlier. For example a session could look like : *(Note : **Applying for roles is one of several things Career-Campaign is designed for but not the whole point of it**. The above is just an example as you could also run different types of work sessions —one that’s purely a networking push (finding and reaching the right people), a step-back “campaign review” that first checks your records are sound, then asks what’s actually converting and what to change, or a session spent writing something public in your own voice. )* Three parts do more than their share of the work. As an overarching principle: **AI discovers and drafts; the human decides and acts.** No auto-apply. No auto-outreach. That’s the deliberate line between a system and a bot. One honest note on how it got built, because it’s a trap worth naming: the parts with obvious mechanics get built first, and the parts that address your real constraint get built last. I built scanning, scoring, and the dashboard early — they had clean, satisfying mechanics. The people-finding side, which turned out to matter most, I built later and thinner. When I cleaned the system up to share it, I promoted that side to a first-class feature, because it’s the part I’d tell someone else to start with, not finish with. To be clear — automation isn’t necessarily a bad thing. Career Campaign tries to remove the tedium of web-searching, extracting information, comparing roles, drafting, generating the dashboard. But the distinction I eventually cared about was not *manual versus automated*. It was whether automation removed work or removed judgement. “Blind” automation often falls into a trap of *set and forget*. You configure it, point it at the job boards, and it sprays applications while you get on with your life. It optimises the one number it can see — applications sent — but it tells you nothing about whether any of it is working. This system is a *thinking partner*. It doesn’t just do the work; it makes you look at the work. Three things fall out of that, and they’re the things a bot structurally can’t give you: The payoff of all this isn’t speed, though drafting a tailored resume and cover letter did drop from two hours to ten minutes. The payoff is that at any moment I knew what was true about my search — and that grounding helps prevent sycophantic behavior or hallucinations in my interactions with it. For all the discipline the system gave me, the dashboard kept telling me the same thing, and it was humbling: the responses I got did not come from the volume. They came from people. Every real conversation I reached traced back to someone who knew what I could do and was willing to say so — a forwarded link, an introduction, a quiet word to someone hiring. I want to be careful here, because it would be easy to draw the wrong lesson. This was, in large part, *my* constraint specifically. I was searching in a market where I hadn’t built deep roots, and even at home my network was oddly insular — years inside one large organisation, where most of the people I dealt with were colleagues under the same roof. I never invested enough in external relationships beyond it. That’s a personal gap, not a flaw in the method. Someone with a broad, well-tended network is in a completely different position — and the system is built to help them use it, through exactly the people-finding features I described. But the general truth underneath my particular gap holds for almost everyone: the roles that matter are disproportionately filled through people, often before they’re ever posted, and no amount of application volume substitutes for someone on the inside willing to vouch for you. This is precisely where a system beats a bot. A bot sends more applications into the channel that’s already saturated. A system points you at the warm paths, tells you honestly that they’re the ones converting, and keeps you disciplined about working them week after week — the slow, unglamorous, high-leverage work that has no deadline and is the first thing to slip without one. My own search resolved the way the dashboard had been quietly predicting all along. The role came through warm contacts who knew me and were willing to advocate for me to the people who mattered. The door was opened from the inside. That isn’t the system failing. It’s the system doing exactly its job: it made me deliberate instead of frantic, it kept me honest about where my effort was actually landing, and it kept pointing me back at the right relationships until I acted on them. So build the system. Let it make you deliberate, and let it keep you honest. Then use the part of it that matters most — the part that helps you find and reach the people — and go do the human work it can prepare but can never do for you. *This templated system — ***Career Campaign ***— is now available on GitHub* The setup process starts with some mechanical installation configs but more importantly, there is an initial ‘interview’ / data gathering stage where it will work with you to understand and record your targets, align on a scoring rubric, collect evidence, and frame your positioning. Note this step **does take a bit of effort and time but **it’s the part you should **definitely not skip or rush**. This is because the output becomes the ‘evidence layer’ that the system then grounds itself in rather than making assumptions or guesses for any future activity. *(One caveat worth stating plainly: what’s on GitHub isn’t the exact config I used for my own search as that version had my history, my contacts, my assumptions baked into every default. What I’ve published is a refined version that can better fit other users whose job search and background looks nothing like mine. The thinking and the lessons are lived-in - If you browse the commits you’ll see that tidying happening in real time)* **If this resonates, I’d like to hear what you’ve found to be the real constraint in your own search. In my experience, it’s rarely the thing the tools are built to fix.** [Today’s Job Market Runs on Automation — So I Built a System to Stay Deliberate In It](https://blog.devgenius.io/todays-job-market-runs-on-automation-so-i-built-a-system-to-stay-deliberate-in-it-24e0eb224e9c) was originally published in [Dev Genius](https://blog.devgenius.io) on Medium, where people are continuing the conversation by highlighting and responding to this story.