Korva is a scroll of the internet’s best ideas.
Revenue per spooner
Spooners are the members of the core team, and revenue per spooner is the closest single number to how well the technology is working. Part of the climb is scale economies, and a lot of it is technology, particularly AI. Any such number can be gamed, since hiring fewer people than the business needs would lift it too, so it works as one signal among several.
The 10x target
The target now is 10x revenue with headcount rising 3x at most. AI helps by making documentation far faster. Writing a memo or describing an experiment can be ten times faster than before, and the company is working to automate it fully.
The Diagram design tool
Design shows the pattern. Bending Spoons built an internal design tool called Diagram, deeply embedded in its broader toolkit. Asked to design something, it needs zero context: it already knows the product databases and the guidelines the lead designer on each business has laid out, and it contacts the right people when it needs feedback or approval. Designers work much faster, and more team members can design effectively on their own. Every handoff between people loses information and adds delays and frustration, and the tool removes many of them.
Why can software scale without scaling staff?
Once software is built, each additional customer costs almost nothing to serve, so revenue can grow far faster than the number of people. When Facebook announced it would acquire WhatsApp in February 2014, the messaging service had about 450 million users and 55 employees. WhatsApp's reported figures at the time were widely cited as an example of a tiny team serving a huge user base.
Why do handoffs between people lose information?
Each handoff forces one person to compress what they know into a message and another to rebuild it. Detail is dropped along the way, and the receiver often has to ask questions, which adds delay. Melvin Conway observed in 1967 that organizations produce systems that mirror their own communication structures. Every boundary between people becomes a boundary in the work.
How can an AI tool need zero context?
Retrieval-augmented generation, described by Patrick Lewis and colleagues at Facebook AI Research in 2020, lets a language model look up relevant documents at the moment it answers. A tool connected to product databases and design guidelines can fetch what a person would otherwise have to remember or ask about. The context travels with the tool instead of with the requester.
More from Joys of Compounding
