A case study in Navi's "go deep" mindset: the popular framing of CAP is incomplete, and a better theorem exists.
The CAP theorem as usually taught
- Framed as "pick two of three": Consistency (data in sync across a distributed system), Availability (servers/databases up and readable), Partition tolerance (how the system behaves when disconnected)
9:20 - The "two out of three" framing "is not super accurate" and the theorem itself is "very incomplete" — it felt hand-wavy to someone who likes tracing things down to the exact line of assembly
9:53,10:18
Validation and the better alternative
- Martin Kleppmann wrote a somewhat controversial blog post criticizing CAP — commenters said he was technically right but nitpicky; Navi felt validated
10:44 - The successor is PACELC: if there's a Partition, choose Availability or Consistency; Else, the trade-off is between Latency and Consistency. More complete, and arguably simpler to understand — Navi sees no reason to teach CAP when PACELC exists
11:09,11:27
The broader point: depth vs. business reality
- Most engineers take the "two out of three" law at face value and move on; only a small subset digs deep enough to notice it's incomplete
11:27 - On the job you can't do "depth-first search on all the document links" — no one person understands a massive codebase, and business problems take priority over full understanding
12:10,12:31 - Agentic coding is forcing a similar transition: you may not even look at the code you produce
12:31