Before you build, name one thing that is new and yours and put it on the roadmap — and avoid spending two years on the same output as everyone else
I Spent Two Years Building the Median — Hariprasad Sudharshan — AI Engineer · Read the original ↗
This page is a translation of the Korean original. Read the Korean original →
What you get
- You treat an agent's default output as the starting line for human work, not the finish
- Three demo questions confirm differentiation and distribution before you commit investment
- You separate output that is good enough for internal use from output meant to compete in the market, and allocate effort accordingly
When to use it
If you build a quick demo with a coding agent for internal use or ship something as a learning exercise, the median result is fine. You need this standard when drawing up the roadmap for a product that competes in the open market, or when, as a leader, you evaluate the stream of weekly demos to decide where to invest.
How to do it
- Before starting to build, first search for 'who has already solved this' to suppress the build-first reflex.
- Before starting to build, name and define one thing that is new; if there is none, use the median result as it is or admit it is a learning exercise.
- Put that novelty plan into the roadmap and project plan, and review it regularly the way you review spending.
- Treat the point where the AI says 'done' not as the end but as the signal that human work begins.
- When evaluating a demo, ask 'what do only we have?' and 'where will this app be used?' to check differentiation and distribution first.
What to borrow
For agent and skill projects already under way, write down 'what only we have' in one sentence; if you cannot, declare the project a learning exercise instead. This is worth adopting directly. The operating rule of treating the moment a coding agent reports completion as the point where human judgment starts can also be applied to a team.
Links
Background — what the original covers
Hariprasad Sudharshan, a product lead with 20 years of experience, describes two years of shipping app after app on his own. Right after a soft launch, people told him they had 'seen a similar calculator'; on checking, nobody had copied anyone, yet the results were almost identical. The talk is about where to put your differentiation when everyone borrows the same models and harnesses, so the same problem yields the same answer.