About
Zipf gives teams and AI agents programmable web monitoring with judgment. Consumer tools send notifications. Zipf delivers scored, structured signals to your systems.
The goal is simple: know what changed before the rest of the market does, without forcing your team to spend its day re-checking the same sources.
Mission
Zipf exists for teams that need ongoing awareness, not occasional lookup. The work is to stay current without forcing people to watch the web by hand.
Zipf's Law shows up everywhere on the web. A tiny fraction of updates drive real outcomes, but you cannot know which fraction without watching broadly. That is the paradox the company is named after.
The problem
The old model of search, scan, repeat was built for humans clicking through lists of links. It breaks down when the question is recurring and the surface area keeps expanding.
The work is not finding one answer once. The work is staying current without asking people to live in tabs.
The product
We build monitors that watch the web with judgment. They check the sources your team would normally look at by hand, extract the fields you care about, score every change for information gain, and compare each run against what mattered before.
Search and crawl are primitives. The product is the infrastructure layer on top: programmable monitors with structured output, signal scoring, and delivery to any system — Slack, webhooks, CRM, or AI agents via MCP and API.
The team
The founding team built search and LLM systems at Microsoft Bing, Snowflake, Neeva, Amazon, Walmart, Qualtrics, and Mendel.AI. We have published dozens of papers, hold dozens of patents, and shipped systems handling billions of queries daily.
The company
We are serious about the work and equally serious about building it at a sustainable pace. Zipf should compound over years, which means hiring people with real lives and giving them room to keep them.
Beliefs
These are not marketing claims. They are the operating choices behind the product, the company, and the kind of work we want to compound over time.
Power laws demand watching widely before filtering narrowly. The vital few signals emerge from the many, so you cannot skip the haystack and keep the needle.
Teams and agents need to define sources, extraction targets, schedules, and delivery paths with precision. Monitoring infrastructure should be programmable and transparent, not a black box that decides what matters for you.
A useful system remembers what it is watching, what changed last time, and where the answer should go next. One-shot retrieval is the starting point, not the destination.
If someone still has to open ten tabs to understand what happened, the job is not done. The output should already carry the fields, context, and delta that matter.
We are not in the prediction business. We pull threads from across the web, compare them with discipline, and help teams read the resulting pattern with confidence.
We value deep technical work, but only if it ships. The best idea is the one that keeps working under customer load, messy inputs, and real operating constraints.
Team
Network
Zipf is backed by operators and founders who understand search, AI infrastructure, and the discipline required to turn technical ambition into an enduring company.
Inventor of the Transformer, CEO/CoFounder Essential AI
CTO & Co-founder Baseten
CVP Search and AI Microsoft Bing
Developer Tools and Product Design
CEO Snowflake, ex CEO of Neeva, ex EVP of Google Ads
SVP of AI Snowflake
Contextual AI, Inventor of RAG
CEO/Cofounder Turbo Puffer
Founder and CTO of Mendel AI, Chief AI Officer Ensemble Health Partners
CTO for Neural Networks, Databricks, CoFounder/CTO MosaicML
CTO and CoFounder at Bauplan, former Director of AI at Coveo
Location
We are based in NYC and open to hybrid and remote work. If you are in the neighborhood, come say hello.
Office
368 9th Ave
New York, NY 10001