Vercel AI Gateway and AWS Kiro Crew Push Composable Multi-Model Agent Stacks

Today’s releases underscore a clear shift toward gateway-mediated access and agent orchestration. Builders now have fresh options for routing inference through unified endpoints while agent frameworks mature toward reliable, multi-step execution. These updates arrive amid broader ecosystem movement on model availability and security hardening. The result is faster iteration cycles for teams shipping coding assistants, retrieval systems, and autonomous workflows.

Top Stories

Gemini 3.7 Flash delivers coding performance gains at 50% discount Practical dev impact: Teams gain faster code completion and generation in IDEs or agents while cutting inference costs through year-end.

AWS open-sources Kiro Crew for multi-agent coding Practical dev impact: Engineers can now run and extend asynchronous agent sessions locally or in the cloud using an Apache-licensed codebase.

Claude Managed Agents integrate with Chat SDK Practical dev impact: Production apps can orchestrate long-running agents directly from the official SDK without custom orchestration layers.

Practical Impact Analysis

AWS’s decision to open-source Kiro Crew accelerates adoption of structured multi-agent patterns beyond single-vendor ecosystems. Combined with Claude’s SDK-level agent support, teams can now prototype reliable, tool-using agents that survive real-world token budgets and context resets. The net effect is a more composable developer experience: choose the model per task, orchestrate via standardized interfaces, and scale without rewriting client code.

Tutorial

Install the Vercel AI SDK (or use the standard OpenAI client pointed at the gateway endpoint). The following snippet demonstrates a simple code-generation call that automatically benefits from the new GLM-5.3 Flash and Gemini 3.7 Flash routes.

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://ai-gateway.vercel.sh/v1",
    api_key=os.environ["AI_GATEWAY_API_KEY"],
)

response = client.chat.completions.create(
    model="zai/glm-5.3-flash",  # or google/gemini-3.7-flash / alibaba/qwen3.8-max
    messages=[
        {"role": "system", "content": "You are an expert Python developer."},
        {
            "role": "user",
            "content": "Write a FastAPI endpoint that accepts a list of integers and returns their sorted unique values.",
        },
    ],
    temperature=0.2,
    max_tokens=400,
    extra_body={
        "providerOptions": {
            "gateway": {
                "order": ["zai", "baseten"],
            }
        }
    },
)
print(response.choices[0].message.content)

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://ai-gateway.vercel.sh/v1",
    api_key=os.environ["AI_GATEWAY_API_KEY"],
)

response = client.chat.completions.create(
    model="zai/glm-5.3-flash",  # or google/gemini-3.7-flash / alibaba/qwen3.8-max
    messages=[
        {"role": "system", "content": "You are an expert Python developer."},
        {
            "role": "user",
            "content": "Write a FastAPI endpoint that accepts a list of integers and returns their sorted unique values.",
        },
    ],
    temperature=0.2,
    max_tokens=400,
    extra_body={
        "providerOptions": {
            "gateway": {
                "order": ["zai", "baseten"],
            }
        }
    },
)
print(response.choices[0].message.content)

Swap `model` to `google/gemini-3.7-flash` or `alibaba/qwen3.8-max`, or keep `zai/glm-5.3-flash` and let `providerOptions.gateway.order` fail over across providers without changing the rest of the agent.

Grok Deep Dive