Postman hands agent workflows to Baseten inference: enterprise deployment gains a division of labor

Deploying enterprise AI agents is now explicitly divided between two vendors: Postman builds and operates the workflows, Baseten runs the models behind them. The two companies offer their capabilities as complementary parts: Postman provides tools for building and operating agent workflows, and Baseten provides high-performance inference and post-training to serve those agents in production. Postman's Agent Stack comprises four products: Passport for scoped credential references, Fabric as an AI-native gateway routing AI, MCP and API traffic, Astropods for runtime, observability and guardrails, and Fern for generating documentation, SDKs and CLIs.

Baseten supports enterprise inference reliability and compliance configuration through pooled multi-cloud GPU capacity and zero data retention by default. Multi-cloud Capacity Management pools GPU capacity across clouds with built-in resilience to hardware failures, and customers can restrict or customize the clouds, regions and clusters they run on, including single-tenant and active-active deployments. Baseten says its model APIs deliver high throughput and low latency, with seven 2026 launches ranking among the top two providers on Artificial Analysis for both throughput and latency.

Post-training is positioned as the way to close the gap between general and specialized models, letting a smaller model reach frontier quality. Baseten states that for the majority of economically valuable tasks, the gap is large between what a general-purpose model does via prompting and what a specialized model does via post-training. Harvey post-trained a 27B open model to match the closed-source frontier on its legal agent benchmark, where top closed models cost roughly $50 per task; Baseten says post-training can match or exceed large frontier model quality at a fraction of the cost.

The report covers the partnership Baseten and Postman announced and their descriptions of each product's capabilities, for researchers and engineers assessing enterprise agent deployment options; no customer list or deployment scale for the partnership is given.

The throughput and latency rankings and the Harvey cost and quality comparison are stated by Baseten without evaluation methodology; the claim that post-training matches frontier quality is qualified to the majority of economically valuable tasks, not all tasks.