Weave raises $13.5M to build AI prompt router that cuts enterprise coding token costs by up to 80%

Weave, a two-year-old AI infrastructure startup, has raised $13.5M led by Standard Intelligence to build what founder and CEO Adam Cohen calls a prompt router for enterprise engineering teams.

The core problem is cost. Coding accounts for roughly 80% of enterprise AI spend, and most teams are routing every prompt to frontier models regardless of complexity. Weave sits between engineering workflows and the underlying models, reads each prompt, and redirects it to the most cost-efficient option. For teams already using tools like Claude Code or Codex, Cohen says the router drops into existing workflows with minimal latency. The claimed saving is up to 80% on token costs for some customers.

The company hosts open-source models in the US and uses that infrastructure to make the routing decisions in real time. Cohen says Weave also has an early reinforcement learning layer that lets organizations feed routing feedback, with per-org fine-tuning in exploration. The near-term version lets a company signal preferences at the prompt level; a longer-term version would learn those preferences automatically over time.

“Adam Cohen: 'We built the world's best prompt router — it reads your prompt and actually shifts it to the right model at the cheapest cost. Once Opus 4 came out, people realized they didn't need the best models for every task. For coding, this is 80% of AI spend. Companies save up to 80% on their token bills.' The company went through YC at the start of last year and currently has 16 employees.”

Who buys it

The buyer is the CTO, with CFOs and CEOs as natural allies. Cohen is explicit that startups on flat-rate plans don't feel the pain yet. The real target is enterprises where individual engineers are spending $5,000 to $10,000 per month on tokens, and where developer experience or developer productivity teams have taken ownership of the optimization layer. Cohen positions Weave as coding-specific rather than general-purpose, arguing that most competing routers are either app-level or trying to cover too much surface area.

Where it stands

The team is 16 people, all in-person in San Francisco, and hiring aggressively for engineers and account executives. Cohen says their single AE is overloaded, which is the kind of problem that makes a $13.5M round feel well-timed.