Essays on the substrate of human connection.
Long-form writing from the team. We publish when we have something to say and a position worth defending.
Apr 12, 2026 · 9 min read
The latency budget for human connection
How much delay can a real-time conversation absorb before it stops feeling real? A breakdown of the perception budget and the engineering decisions that protect it.
Mar 4, 2026 · 12 min read
Why moderation is a recommendation problem
Trust and safety stops working when it's treated as enforcement. The teams that scale treat it as ranking: what to surface, what to delay, what to amplify with care.
Jan 22, 2026 · 10 min read
Identity primitives for relationship-driven products
How to confirm a real human signed up without becoming the company that knows everything about them. A pragmatic framework for proof-of-personhood.
Dec 10, 2025 · 8 min read
The cost of a million open WebSocket connections
What it actually takes to keep a million sockets alive without lighting your bill on fire. Memory, file descriptors, keepalive intervals, and the kernel knobs nobody told you about.
Nov 4, 2025 · 11 min read
How we run the video stack behind our own brands
SFU, MCU, P2P, TURN. The video architecture we built and operate ourselves, and how we chose each piece for the calls our members actually make.
Sep 18, 2025 · 9 min read
The matching loop most teams get wrong
Optimizing for the click is the easiest mistake to make and the most expensive one to undo. A note on the difference between matches that feel right in the moment and matches that retain.
Aug 2, 2025 · 7 min read
TURN servers, the most expensive line item nobody plans for
Every WebRTC deployment quietly subsidizes a percentage of sessions that cannot establish direct peer paths. Here is what that subsidy costs and how to keep it bounded.
Jun 15, 2025 · 10 min read
Designing reviewer tooling that doesn't burn humans out
Trust and safety at scale runs on people. The interfaces those people use determine whether you are running a moderation pipeline or a turnover engine.
Apr 21, 2025 · 13 min read
End-to-end encryption and the moderation paradox
You can protect your users from outsiders or you can protect them from each other. Most products need to do both, and the tradeoffs are harder than the public conversation suggests.
Feb 8, 2025 · 8 min read
Regional payment rails: when local is non-negotiable
Cards are the lazy answer everywhere outside the United States. A pragmatic guide to when regional rails matter, when they don't, and what authorization rates actually look like.
Dec 5, 2024 · 9 min read
The case for boring databases in messaging
Cassandra, ScyllaDB, and DynamoDB are the right answer at scale and a needless complexity tax before that. A note on when boring wins.
Oct 14, 2024 · 11 min read
Designing for the moment a moderator says no
The user experience of being told no is the area where trust and safety either earns its name or loses it. A study of what good looks like.
Building the systems that bring people together.
Questions about our brands, press, or partnerships? Send a note and we'll reply within two business days.