A startup just raised $200M to deploy 10,000 AI agents at once
Blitzy's platform reverse-engineers your entire codebase into a knowledge graph, then orchestrates thousands of agents in parallel for days. No human writes the code.
On May 5, Blitzy announced a $200 million funding round at a $1.4 billion valuation. The round was led by Northzone, with participation from Battery Ventures, PSG, and Liberty Mutual. But the number that matters more than the valuation is this: thousands of agents running in parallel for days to weeks, shipping 80% of what would otherwise be multi-month engineering projects.
Blitzy doesn’t help developers write code faster. It replaces months of human engineering work with autonomous agent swarms.
What it actually does
Blitzy reverse-engineers your existing enterprise codebase — 1 million to 100 million+ lines — and builds a dynamic knowledge graph of the entire estate. It maps every service, every dependency, every API contract, every database schema, and every deployment target. It maintains a persistent understanding of the environment, updated as the codebase changes.
Grounded in that knowledge graph, its orchestration layer deploys thousands of agents in parallel, running for days to weeks of uninterrupted inference. Each agent handles a slice of the work — one refactors the database layer, another rewrites the API endpoints, a third handles the frontend, a fourth runs tests and validates output.
The results get shipped. Humans review and polish the remaining ~20%.
The company claims a record-breaking 66.5% on SWE-Bench Pro, ahead of any publicly known score. More importantly, they’re already deployed across dozens of Global 2000 enterprises in 10 industries.
Why this matters more than another funding round
We’ve seen a lot of money go into AI coding tools. Cursor raised $105M. Codex is part of OpenAI’s core product. GitHub Copilot is in every IDE. But almost all of these tools share the same bottleneck: they work one session at a time, with a human in the loop.
Blitzy’s approach is fundamentally different. It doesn’t augment a developer — it replaces the entire development cycle. The human is no longer writing code or reviewing every PR. The human defines the objective, the platform executes. This is the difference between a better hammer and a machine that builds the whole house.
The key insight is the knowledge graph. My last post covered how agents spend 81% of their time searching for the right file — reading code, grepping, trying to understand the codebase. Blitzy’s answer is to do the understanding once, upfront, and persist it. The agents don’t wander. They navigate a pre-built map of the entire system.
The team
Founded by Brian Elliott, a former Army Ranger and serial entrepreneur, and Sid Pardeshi, an NVIDIA Master Inventor with 27+ patents in neural networks and AI-driven systems. They met building advanced software at Harvard. The combination is telling — this isn’t a team of AI researchers trying to make a better model. It’s a team that thinks about mission execution at scale.
Who this is for (and who it isn’t)
Blitzy is for the Fortune 500. Companies with 50-million-line monorepos, regulatory compliance requirements, and engineering orgs that measure projects in quarters, not sprints. They’re already in financial services, insurance, and government — industries with the most to gain from automation and the most to lose from sloppy code.
Blitzy handles that by orchestrating multiple models (Google, Anthropic, OpenAI — whoever scores best per task), running 100,000+ model calls per project. Quality gets validated at scale, not by a single model’s judgment.
The signal
Three months ago, Blitzy was a company few had heard of. Now it’s worth $1.4 billion and enterprise customers trust it with their core codebases.
The timeline for “AI replaces software engineers” keeps getting pulled forward. Not because the models got smarter — they were already smart enough. But because the infrastructure around them — knowledge graphs, agent orchestration, persistent execution — finally caught up.
The question isn’t whether enterprises will adopt autonomous coding. They already have. The question is who builds the platform that earns their trust.