Perplexity Unveils CobbleDB, Built by 2 Engineers and Hundreds of AI Agents in Two Months

Perplexity has published research on CobbleDB, a self-built key-value database that serves prepared web content to its search engine, replacing AWS DynamoDB. The core infrastructure was built in two months by 2 engineers leading hundreds of proactive, continuously-running AI coding agents. After the migration, production median batch-read latency dropped from 31.4ms to 5.60ms and p99 from 123ms to 24.2ms, with internal cost models estimating at least 20% savings versus DynamoDB (reposts claim actual savings are higher).

Confirmed

• Development model: hundreds of proactive coding agents work under engineer direction with cross-session continuity; engineers handle architecture design, code review, and deployment authorization
• Measured performance: median batch-read latency from 31.4ms to 5.60ms, p99 from 123ms to 24.2ms (roughly 5x reduction); internal cost model estimates at least 20% savings over DynamoDB
• Read architecture: reads prioritize fast batch reads, grouping page keys by partition and using RocksDB MultiGet in parallel; routers prefer same-availability-zone replicas and switch to another replica if one responds slowly
• Write pipeline: to optimize the read path, CobbleDB separates persistent document state, update delivery, and read serving into Pillar, Lorry, and CobbleDB—Pillar stores versioned document state in YTsaurus and enqueues exports, Lorry delivers updates, and replicas consume updates independently in order

Unconfirmed

• Open-source status: only repost titles (m9, m10) claim CobbleDB has been open-sourced; no explicit statement found in Perplexity's official posts
• Actual cost savings: reposts claim savings exceed 20%, while the official posts only state "at least 20%" as a lower bound

Why it matters

CobbleDB demonstrates a new human-AI collaboration paradigm: a tiny number of engineers, by controlling architecture and holding review/deployment authority, can direct a fleet of AI coding agents to build core infrastructure in weeks. The significant latency and cost gains from replacing DynamoDB also offer an engineering reference for read-heavy, high-throughput workloads.

2026-09-16 ~ 2026-09-16 · 10 related posts

• [source] Perplexity: two engineers plus hundreds of AI agents built CobbleDB in two months — perplexity_ai · 2026-09-16
• How Perplexity ran coding agents: continuous cross-session follow-through, engineers gate production — perplexity_ai · 2026-09-16
• Inside CobbleDB: partitioned RocksDB MultiGet reads with same-zone-first routing — perplexity_ai · 2026-09-16
• Why CobbleDB split state, delivery, and serving: optimizing reads required rewiring updates — perplexity_ai · 2026-09-16
• CobbleDB's write path: state, delivery, and serving split across Pillar, Lorry, and CobbleDB — perplexity_ai · 2026-09-16
• [source] CobbleDB vs DynamoDB: median batch-read latency 31.4ms→5.6ms, 20%+ cost savings — perplexity_ai · 2026-09-16
• [source] Perplexity's CobbleDB cuts p99 latency from 123ms to 24.2ms, replacing DynamoDB — perplexity_ai · 2026-09-16
• Perplexity: two engineers and hundreds of AI agents built CobbleDB, replacing DynamoDB to save ~$100M/year — cameronstow · 2026-09-16
• Perplexity built a DynamoDB replacement in 2 months with 2 engineers and hundreds of AI agents — inductionheads · 2026-09-16

1 near-duplicate retellings: inductionheads