Key Takeaways
- Vercel champions practical AI engineering, abstracting complex patterns from internal battle-tested products.
- The Workflow Development Kit enables durable, cost-free serverless functions, streamlining AI agent orchestration.
- AI SDK 6 offers low-level flexibility for evolving AI applications, contrasting with high-level agent frameworks.
- Vercel's integrated agents, including a DevOps agent, automatically investigate production anomalies before human intervention.
- Vercel now natively supports Python backends (Flask, FastAPI) and provides open-source agent templates.
Deep Dive
- Vercel launched a Workflow Development Kit (WDK) at Ship AI 2025.
- The WDK enables durable, streamable, and resumable code execution for AI agent workflows.
- Developers can pause and resume serverless functions indefinitely without cost, with automatic retries for reliability.
- It supports reliable agent orchestration, including human-in-the-loop approvals via ephemeral webhooks.
- Vercel's AI SDK, version 6 beta, is noted for its success despite competition.
- Its popularity is attributed to a low-level approach allowing flexibility as AI application trends evolve.
- The SDK provides foundational capabilities like streaming text and tool selection, built from v0 experience.
- It emphasizes tools developers can configure and control, contrasting with high-level agent abstractions.
- Vercel emphasizes building low-level tools that developers can configure and control.
- This contrasts with frameworks prioritizing high-level agent abstractions from day one.
- The company focuses on developer experience, highlighting streaming for responsiveness in AI applications.
- Vercel's core principle involves never shipping abstractions not battle-tested internally, exemplified by the AI SDK's extraction from v0.
- Vercel launched an 'agent-as-a-service' product, the Vercel Agent, integrated with runtime data and error logs.
- Internally, Vercel developed a code review agent and a new DevOps agent.
- The DevOps agent investigates production anomalies detected by Vercel's system, utilizing observability data and logs.
- This integrated agent approach is compared to onboarding a junior employee, simplifying complex tasks.
- The DevOps agent addresses the recall-precision problem in anomaly detection by allowing aggressive alerting.
- The agent performs investigations before waking human responders, analyzing time series data, IP addresses, and error messages.
- It utilizes query tools and logs for deeper investigation into performance or error spikes.
- While it can investigate, time series analysis and aligning human preferences remain challenges.
- Vercel has open-sourced three custom agents to provide starting points for companies.
- These include a lead qualification agent for sales requests and an abuse analysis agent for pre-processing reports.
- A data analyst agent for querying data warehouses is also among the open-sourced templates.
- The 'Agent on Every Desk' program offers support and consultations for companies building their first production AI agents.
- Vercel has invested in Python support, releasing zero-config deployment for Flask and FastAPI applications.
- A Python SDK for Vercel's API has also been made available.
- Python support is integrated into the fluid compute program, offering active CPU pricing where users pay only for compute.
- Vercel aims to deliver a native developer experience for Python, similar to its TypeScript offerings.
- The role of a CTO has significantly evolved with the rise of AI, requiring company transformation.
- Vercel is developing infrastructure that assumes developer incompetence, aiming for secure applications even with human or AI errors.
- This platform-level security focuses on secure collaboration and data access within applications.
- The company emphasizes maintaining strong individual contributor career paths to adapt to AI's changing nature of software development.