Vercel's Sakana Fugu Ultra Says AI Gateways Are Becoming Multi-Agent Routing Layers

Vercel's latest AI Gateway update is interesting because it is not really about one model.

Sakana Fugu Ultra is built on a pool of publicly accessible frontier models rather than running as a single model. It coordinates several models, routing work to one to three agents depending on the problem and combining their results into a single answer.

That is a different kind of product statement. AI Gateway is no longer just a place to send model requests. It is starting to look like a control layer for how multi-model work gets assembled.

The important part is the routing behavior

If the system decides different sub-models should handle different parts of a problem, then the gateway is no longer only a transport pipe. It is participating in task decomposition and synthesis.

That moves Vercel further into the infrastructure business of deciding how model work should be split, not only which endpoint should be called. Teams do not only need model access. They need a way to decide when to fan out, when to combine, and how to keep cost and reliability visible while doing it.

Why this fits the current market

Builders want one place to call models, track usage, configure retries, and control failover. They also want the option to wrap more opinionated logic around the request so they do not have to reinvent orchestration every time. Fugu Ultra sits right in that gap.

That is why this launch belongs next to Vercel's other control-surface pushes like Connect, Passport, and eve. The company is not just shipping features. It is assembling a stack where access, routing, runtime, and project shape all point in the same direction.

AI Gateway is becoming a policy layer, not a dumb proxy

Vercel says AI Gateway provides a unified API for calling models, tracking usage and cost, and configuring retries, failover, and performance optimizations. Add multi-agent routing into the picture and the gateway becomes a policy surface where teams can reason about cost, uptime, model choice, and fallback behavior in one place.

That is the right place for this kind of logic to live. It is closer to the operational truth than scattering routing rules across app code.

Butler's view

The deepest signal in Fugu Ultra is that AI gateways are evolving from request routers into multi-agent orchestration layers.

That is a meaningful shift for teams trying to keep model usage governable. It says the real competition is moving upward: not just which model you call, but which layer decides how the work gets split, combined, and paid for.

Related coverage

• our earlier read on Vercel making production defaults the product
• Vercel's broader control-surface work around deployment access
• how Vercel has been scoping agent access
• the workflow-control side of the same stack

AI Disclosure

This article was researched and drafted with AI assistance, then reviewed and edited for clarity, accuracy, and editorial quality.