Korva is a scroll of the internet’s best ideas.
Why hold things loosely
Technology can change in a month. A setting or interaction that was needed may stop being needed, a new capability may appear, or latency may drop far enough to rethink a whole feature. Until a piece of the product has hardened, the team holds it loosely.
What moves fast
Some work at OpenAI moves at that speed. The ChatGPT composer changes every single day. Codex and its super app get quick tries, and the team builds in public, so feedback arrives instantly from inside and outside the company. Some ideas go from start to ship in four hours.
What gets full treatment
Other features get the full treatment. Some ChatGPT features get a hundred attempts, with ninety-nine thrown out and one shipped, and those get A/B tests and user research. Making ChatGPT feel deeply embedded in a person's personal life is the kind of project that gets sketched on whiteboards and in Figma, then prototyped and used inside the company long before any release.
How does an A/B test judge a feature?
An A/B test shows two versions of a feature to randomly assigned groups of users and compares a chosen measure, such as how many people finish a task. Random assignment makes the groups comparable, so a difference can be credited to the change. Results need enough users to separate a real effect from chance. Products with millions of daily users can reach that point within days.
Why do teams use their own products?
The practice is called dogfooding, and it spread through the software industry in the late 1980s. People who use a product every day hit rough edges that written specifications never reveal. Their reports reach the team without waiting for outside feedback. Microsoft is widely credited with popularizing the term.
Why can lower latency redesign a feature?
Usability researcher Jakob Nielsen wrote in 1993 that a response under 0.1 seconds feels instantaneous, under one second keeps a person's flow of thought, and ten seconds is the limit of attention. Those thresholds trace to earlier human-computer interaction research, including Robert Miller's 1968 paper on response time. When a task drops below one of them, a different style of interface becomes possible, such as results that update as someone types.
More from Lenny's Podcast: Product | Growth | Career
