Building the products was the easier part.
Over the last few months, I’ve been building two AI products on nights and weekends.
One is MayaGuru, an image workflow platform. The other is ApplyBench, a job-search decision tool.
Both are live. Both work. And both ended up teaching me different lessons.
MayaGuru pushed me toward a subscription mindset.
It is not enough for something to be useful once. A subscription product has to earn a habit.
That changed how I think about the product. The question stopped being “can this workflow work?” and became “who actually needs this often enough to come back next month?”
ApplyBench taught me almost the opposite lesson.
A lot of AI job-search tools are built around speed. I was more interested in decision quality.
The candidate should still be in the loop, because no model knows your experience as well as you do.
So ApplyBench starts one step earlier: is this job real, current, and worth your time?
If it is, then AI can help with the next steps — tailoring the resume, drafting a cover letter, checking ATS fit, and preparing for interviews.
That changed the product shape and the business model too.
Not everything needs to be a subscription. Sometimes people just need help at the right moment.
Looking back, I also value both products as engineering labs.
ApplyBench pushed me into grounded job inputs, source confidence, authentication, credits and AI budgets, rate limits, persisted results, recovery paths, and retention controls. MayaGuru pushed me into image-generation pipelines, shared identity across multiple apps, usage controls, streaming chat, conversation persistence, and model cost management.
I did very little distribution for either product, so I do not treat them as traction stories. Their value to me has been building and operating the whole system around the AI feature — not just getting a model call to work.
Working on both at the same time has been a good reminder that a product can be technically real before the business is real.
Getting the software working matters. But distribution, positioning, repeat usage, and pricing logic are where the harder learning starts.
That gap between building something real and building something people return to has taught me more than the shipping part itself.