# Hornet blog: Insights and product updates

Insights on retrieval, runtime architecture, and building search infrastructure for agents. Each post links to its Markdown version. Token counts are approximate.

## Posts

- [The agent traffic explosion, measured](https://hornet.dev/blog/the-agent-traffic-explosion-measured.md): Bots became the majority of the web traffic Cloudflare measures, ahead of forecast. Projections for agent query volume run to 1000x. Here is what the data actually supports (2026-07-29, ~1.9k tokens)
- [Code mode vs tool mode: the agentic retrieval harness matters](https://hornet.dev/blog/same-retriever-fewer-tokens-better-recall.md): Same retriever and model, different harness: 65% higher recall, 51% fewer prompt tokens, 44% lower cost (2026-07-14, ~2.0k tokens)
- [Code mode for agentic retrieval](https://hornet.dev/blog/code-mode-for-agentic-retrieval.md): Agents can write retrieval code. The hard part is exposing the primitives that decide what comes back, how it is scored, and what survives into context (2026-06-18, ~1.4k tokens)
- [Agent queries cost more to serve](https://hornet.dev/blog/agentic-query-workloads-change-retrieval-cost.md): Agent queries are longer and carry more operators than human queries. On the same engine and index, that distribution shift cuts serving capacity 8.4x, and agents are now the workload to optimize for (2026-06-03, ~2.1k tokens)
- [This is what agentic retrieval looks like](https://hornet.dev/blog/this-is-what-agentic-retrieval-looks-like.md): GPT-5 searches like a power user. Phrase quotes in 98% of sessions. Median query length past the human 99th percentile (AOL query log). The new user of search (2026-05-20, ~2.6k tokens)
- [What we learned building a 100M-document search engine, part 3: Hybrid search depends on query type](https://hornet.dev/blog/100m-doc-search-part-3-hybrid-search.md): We tested hybrid search across document-grounded query regimes at 100 million documents. The best merge strategy changed with query type, and the evaluation data had to be cleaned before the results meant anything (2026-05-13, ~2.8k tokens)
- [What we learned building a 100M-document search engine, part 2: ANN Tuning](https://hornet.dev/blog/100m-doc-search-part-2-ann-tuning.md): We tuned ANN indexing, query embeddings, and quantization across 100 million documents. The components interact in ways benchmarks do not predict (2026-04-20, ~2.1k tokens)
- [Deep research is a retrieval problem](https://hornet.dev/blog/deep-research-is-a-retrieval-problem.md): Why retrieval is a dominant bottleneck in BrowseComp-Plus (2026-03-24, ~1.9k tokens)
- [What we learned building a 100M-document search engine](https://hornet.dev/blog/100m-doc-search-part-1-what-we-learned.md): Every experiment at 100 million documents costs a full day. We built a retrieval system at this scale and learned that iteration speed matters more than any single parameter (2026-03-12, ~2.0k tokens)
- [Context Engineering for Coding Agents](https://hornet.dev/blog/the-context-window-is-not-your-database.md): Why giving agents more context often makes things worse (2026-02-28, ~1.4k tokens)
- [The scaling dimensions of keyword search](https://hornet.dev/blog/the-scaling-dimensions-of-keyword-search.md): Document count is only one scaling dimension. Query length and query structure matter just as much, and agents are pushing all three (2026-02-06, ~2.5k tokens)
- [Mutually assured distraction](https://hornet.dev/blog/mutually-assured-distraction.md): Why better retrieval and better reasoning can make agentic systems less reliable (2026-01-30, ~3.0k tokens)
- [How we build a retrieval engine for agents](https://hornet.dev/blog/how-we-build-a-retrieval-engine-for-agents.md): Hornet's verifiable API surface enables agents to learn, configure, and optimize retrieval on their own (2026-01-16, ~1.3k tokens)
- [The case for a new retrieval engine for agents](https://hornet.dev/blog/the-case-for-a-new-retrieval-engine-for-agents.md): Our motivation for building Hornet, a retrieval engine built for agents (2025-10-08, ~1.1k tokens)