AgentBrief: curated daily news for the agentic web. Sign up now →AgentBrief: curated daily news for the agentic web. Sign up now →

From 3,000 to 21,000: what we shipped this week

·Balazs Nemethi
#community#update#scaling#infrastructure
View Source

On Tuesday morning Pacific, Brave posted that they endorse the Agent Community.

By Satuday, we went from around 3,000 members to over 21,000. That is the best kind of problem you can have, and it is also the kind that finds every crack in your system. This is a quick look at what we shipped this week to keep up.

Before the wave

The first half of the week was still peace-time. We had been running three separate sites: the main site, a blog subdomain, and a docs subdomain. Three deployments, duplicated navigation, broken cross-links. We merged everything into one site, one codebase, one search index. Blog, docs, and all content pages now live together with unified search, RSS, and a new toolbar that lets you copy any page as Markdown for LLMs. We also shipped llms.txt so AI systems can discover our public content.

This was a cleanup we had been meaning to do for weeks. Turns out it landed just in time.

Then Brave happened

Once signups started surging, things got real fast. Here is what we worked on across two repos, every single day this week.

The map had to scale

The /map is a periodic-table-style grid of every organization in the community. It was loading around 600-700 profiles. It needed to show over 6,000 without melting your browser or our servers.

What we did: virtual scrolling so only what is on screen is in the DOM, lazy-loaded logos, smarter caching, and pagination past a database limit that was silently cutting off results. The map now loads the full directory, fast, even on mobile.

(While you are there... we hid an easter egg somewhere in the /map. First person to find it gets a special prize.)

Security for the new traffic

A lot more eyeballs means a lot more creative visitors. We ran a full code audit and shipped hardening across the board: tighter content security policies, input validation on every API route, token-based verification on public endpoints, path traversal prevention, and 5,000+ lines of dead code cleaned out. Nothing dramatic happened, we just wanted to make sure nothing could.

Email kept getting throttled

Our email system was built for sending a few hundred emails a day. It was not built for thousands of new signups in a few hours. It got throttled. Multiple times. We rebuilt the sending pipeline to process emails in batches and re-sent endorsement invitations to roughly 8,400 members who had been missed by earlier throttling. If you signed up and your endorsement email was late, that was us catching up.

Auth at scale

When thousands of people sign up in a short window, every edge case in your auth flow surfaces. Rate limiting now shows a clear message instead of a generic error. A race condition with old signup cookies got fixed. Organization signups got guardrails to prevent accidental misclassifications. The document signing flow was redesigned to work inline instead of bouncing to an external page. We ended the week by writing nearly 2,000 lines of tests covering every auth path.

Sterling had to learn to run

Sterling is our agent that handles backend operations for the community. One of its core jobs is enrichment: when an organization joins, Sterling crawls public data to build a profile with a logo, description, website, and taxonomy tags. That profile is what you see on the /map.

The enrichment pipeline was designed for 6 to 10 new members a day. Then we got around 12,000 signups in about 20 hours. Everything broke. The backlog was massive and we are still working through the tail end of it, but over 5,000 organization profiles have already been enriched.

Most of the backend work this week (41 commits, 9,000 lines of code) was rebuilding Sterling's tools to handle this scale: smarter batching, guardrails to skip profiles that do not need enrichment, and a health dashboard so we can see where the pipeline is at any given moment. When the next wave comes, Sterling will be ready for it.

Members can now manage their own data

One thing we shipped on the member side: you can now delete your account entirely from your dashboard if you want to leave. Self-serve, no support ticket needed. Your data, your choice.

The numbers

Across both repositories combined this week:

Commits227
Lines shipped46,000+
Active days7 out of 7
AI-assisted~95%

Almost every commit was co-authored with AI. This is not an AI flex. It is how we ships across six repos, seven workstreams, and every day of the week while also eating and sleeping (mostly).

The Agentic Web Map

The /map is becoming the periodic table for the agentic web. Over 6,000 organizations building agents, agent infrastructure, and adjacent technology, browsable, searchable, and enriched with real data. It is the most comprehensive directory of the agentic ecosystem that exists, and it grows every day as new members join and Sterling enriches their profiles.

If you have not looked at it recently, go take a look. It has changed a lot this week.

What is next

The immediate pressure was handling the surge. The deeper work is making sure the system is ready for the next one. More test coverage, better monitoring, continued performance tuning as the member count keeps growing.

21,000 members. The infrastructure held. The system is stronger now than it was a week ago.

Welcome to everyone who joined this week. We are glad you are here. Back to building.