The 1.0 Hangover: Learning & Iterating

The fun part of building product for me has always been post 1.0. Getting something out the door is important, and in service of that timebound mission, we often have to make tradeoffs and work under stressful constraints that point to "good enough" design decisions rather than "really hits the spot". These constraints and the paths they shape are all too real even when you're working for yourself.

Right before launching autobiographer.io, I had an epiphany that resulted in a pivot from a freemium model to a completely free community based model that rewarded participation and sharing.

I pushed out launch by a few days in order to accommodate the metering changes that constituted MVP and breathed a sigh of relief when all the tests passed and v1.0 was done.

When "done", we effectively had a really good resume and cover letter writer built on Claude Opus that we were giving away for free. We weren't pinching pennies in an effort to maximize margin, and using the smartest and most thoughtful frontier model around resulted in an obvious quality difference that user feedback backed up.

However, I had the nagging feeling something was still missing. "Really good resume and cover letter writer" isn't the same thing as "really good job coach". We were fundamentally still facilitating the same goal of quickly cranking out resumes and cover letters. We were still feeding the same ATS black hole that we wanted to destroy.

Our credits structure reflected as much. The number of credits it costs a user to generate an AI output is graded in terms of actual token costs for the model used for the intelligence task.

Our most expensive task was still the resume generation task. For all the lip service about the job analysis being so crucial, our spending patterns didn't reflect this priority. We were using Sonnet to do some basic inference based on the job description, which sounded useful, especially when delivered with some of Claude's trademark sass, but there was clearly much more we could do to help (intel on company culture, recent newsworthy business moves etc.).

The list went on - we were offering basic insight in the gameplan about the delta between a job's requirements and the candidate's qualifications, but we weren't giving them anything actionable. What if you're a long way from your dream job and you don't want to give up? Surely, one of the greatest hive minds in the world can point you in the right direction and give you a path to remediating skill or experience deficits? Again, this was because in the original design of the product, the gameplan was in service of creating a really good resume and was bundled in with the cost of resume generation.

Effectively, we had given our patient a change of clothes when what they needed was a brain transplant. If our goal is fewer, more thoughtful, better researched and targeted job applications that are more likely to land the candidate their dream job, the product needs to be fundamentally different.

-Kal