I Built an AI Job Tracker While Job Hunting: The AI Was the Easy Part | HackerNoon

TEXT START: How aggregating 11 job APIs into one schema turned out to be harder than any of the LLM features.

The Dissection

This is a competent engineering build log and an accidental labor-market autopsy. Its real subject is not API normalization. It is the conversion of a graduate’s exclusion from the labor market into a respectable productivity project.

The author correctly identifies the external world as the hard part: fragmented providers, inconsistent schemas, rate limits, authentication, deployment friction, and unreliable data. The LLM is easy because the author controls the input and output contracts. That is sound engineering.

But SearchWork only accelerates the search for scarce jobs. It does not create jobs, increase the candidate’s bargaining power, or produce an asset the candidate controls. It turns job hunting into a more efficient competition among increasingly automated applicants. The spreadsheet is replaced by a system; the structural trap remains.

The Core Fallacy

The article mistakes ease of construction for durable economic value.

That the AI features were easy to build is not evidence that AI is harmless. It is evidence that cognitive task packaging is becoming cheap. CV extraction, matching, ranking, and cover-letter generation are precisely the kinds of functions that become interchangeable once models can perform them reliably through structured interfaces.

The genuinely difficult work—integrating hostile, fragmented external systems—may provide temporary leverage, but it is not automatically a moat. Job boards can standardize APIs, aggregators can absorb the function, and coding agents can increasingly maintain the adapters. The remaining friction is being industrialized too.

The deeper error is the conclusion that a portfolio piece “actually about you” is a survival strategy. In a collapsing junior market, demonstrating that you can assemble commodity services does not make you economically necessary. It makes you legible to a filtering system for a little longer.

Hidden Assumptions

• Junior technical employment will remain available at sufficient scale for better search tooling to matter.
• More applications, better matching, and tailored cover letters will convert into employment rather than intensify applicant saturation.
• Employers will reward this project as a meaningful differentiator instead of classifying it as another wrapper around commercial APIs and models.
• API integration, prompt contracts, async orchestration, and deployment plumbing will remain scarce human capabilities rather than agent-maintained infrastructure.
• The job-board data is sufficiently accurate, current, permitted, and complete to support consequential decisions.
• The model’s reliability, pricing, access, and behavior will remain stable.
• Individual optimization can overcome a system in which employers automate selection faster than candidates can optimize applications.
• Building a useful tool for oneself creates ownership or control. In this case, the critical assets remain elsewhere: the models, job inventories, platforms, distribution, and hiring decisions.
• The current crisis is mainly a search-friction problem rather than the early stage of productive participation collapse.

Social Function

This is partial truth wrapped in transition management and prestige signaling.

The technical lessons are real: normalization layers matter, failure isolation matters, structured outputs beat prose parsing, and token handling is better than careless localStorage use. The article is not fraudulent.

Its ideological function is different. It translates systemic disposability into an individual engineering assignment: if the market rejects you, build a better funnel. That preserves the belief that sufficient initiative can restore employability, while leaving the ownership of productive infrastructure untouched. It is a clean, modern form of anesthetic—less “believe in yourself” than “automate your own administrative humiliation.”

It also serves as transition management. The displaced worker is encouraged to become an optimizer of the displacement process: search faster, tailor more, compete harder, and accept the resulting increase in applicant throughput as progress.

The Verdict

Useful engineering write-up, strategically delusional at the labor-market level.

SearchWork is a servitor credential at best, not a sovereign asset. It may buy the author interview probability and demonstrate real implementation skill in the short term. It does not reverse P1, P2, or P3. It automates job-seeking inside a system that is automating the need for junior cognitive labor.

The article’s most revealing fact is not that the AI was easy. It is that the author had to build an AI system to compete for permission to perform work that the system is steadily making unnecessary. That is not escape from the discontinuity. It is a well-engineered position inside its blast radius.