AuthorAgent — AI Cover Letter System
Overview
Most AI-generated cover letters read like form letters. They are generic, structurally flat, and easy for hiring managers to dismiss. The problem is not the language model. It is the absence of a decision architecture that forces the model to do the strategic thinking before it writes prose. AuthorAgent is an open-source Claude AI agent that scores existing cover letters against a 100-point rubric and builds new ones through a structured generation pipeline. It is free to use, requires no coding, and runs inside a standard Claude account.
Check it out here.
My Role
I designed, built, tested, and shipped AuthorAgent end to end. This included rubric design, agent instruction architecture, prompt engineering, iterative calibration, the product website, and open-source release.
What I Delivered
100-Point Scoring Rubric
I designed a ten-category evaluation framework covering format, argument integrity, opening discipline, strategic orientation, JD coverage, evidence depth, organizational integration, voice and language, transition argument, and closing discipline. Every deduction is tied to a specific quote from the letter and a concrete fix, replacing subjective feedback with structural diagnosis.
Structured Generation Architecture
I built an agent instruction set that completes eight pre-writing decisions before any prose is generated. This prevents the model from discovering its thesis while writing, which is the root cause of most AI cover letter failures. A five-paragraph construction sequence assigns each paragraph a specific structural role, and seven empirically discovered failure modes are encoded as guardrails throughout the generation pipeline.
Iterative Calibration System
I developed and validated the system across over 140 letter generations and six agent model iterations. Each round followed a generation, scoring, diagnosis, and fix cycle that surfaced specific failure patterns. This process produced an eight-point pre-submission audit that catches scoring losses before the letter is presented to the user.
Zero-Friction Deployment
I designed the product to require no API keys, no installation, and no coding. Users copy a single instruction set into a Claude project and begin chatting. The agent presents three workflow options and guides the user step by step. Setup takes two minutes.
Impact
Produces cover letters scoring 85 to 92 on the rubric, compared to typical AI-generated letters scoring below 60
Reduces full cover letter generation time to ten to fifteen minutes with guided question flow
Eliminates seven empirically identified failure modes through encoded guardrails
Validated across over 140 letter generations and six agent iterations prior to release
Deployed as a free, open-source tool requiring zero technical setup