Vercel's Andrew Qu on why agents are a new kind of software

2024-12-27 · Source: Latent.Space - Www.latent.space · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, medium

Summary

Vercel's Chief of Software, Andrew Qu, explains why AI agents represent a new form of software, distinct from traditional web applications due to their dynamic interaction and outputs. Vercel's experience building its "v0" vibe-coding product led to the development of "eve", a prescriptive framework addressing common agent development "paper cuts" like model switching, fallbacks, and resumable runs. Qu highlights that agents excel at repetitive tasks requiring reasoning, such as legal contract redlining or data store queries. He also notes a significant rise in bot traffic, advocating for an "agent-readable web" where structured, machine-readable content is served to agents, separate from human-facing visual sites. Vercel itself is evolving into an agent, integrating agent capabilities across its platform.

Key takeaway

For AI Engineers building dynamic applications, recognize that agents demand a different development paradigm than traditional web apps. Focus on primitives for context, tools, and resumability, and consider frameworks like Vercel's "eve" to streamline development. You should also prepare for an "agent-readable web" by providing structured, machine-readable content to bots, ensuring your products remain discoverable and usable as bot traffic increases.

Key insights

Agents are a new software type requiring distinct primitives for dynamic, reasoning-based tasks.

Principles

• Agents need primitives for context, tools, and resumability.
• Feedback cycles should fit the task's autonomy needs.
• Skills provide portable, on-demand knowledge for models.

Method

Vercel developed "eve" by accumulating solutions for agent development challenges like model switching and resumability, then assembling them into a cohesive, prescriptive framework.

In practice

• Use agents for repetitive tasks requiring reasoning.
• Publish skills for current product versions.
• Serve Markdown to agents instead of HTML.

Topics

• AI Agents
• Vercel eve
• Agent Frameworks
• Agent-Readable Web
• Software Development
• Bot Traffic

Best for: AI Architect, CTO, VP of Engineering/Data, AI Engineer, Software Engineer, Director of AI/ML

Related on AIssential

• Vercel Built a Programming Language for AI Agents — AI agent development is driving new infrastructure, including agent-first programming languages and knowledge systems.
• What's Easy Now? What's Hard Now? — AI coding agents' long-term capabilities are primarily determined by the availability and effectiveness of feedback loops, not just open-loop model behavior.
• How I code with agents, without being 'technical' — AI agents enable non-technical users to build complex software by abstracting traditional coding into a guided, iterative process.
• The Sequence Opinion #840: The Agent-Native Rewrite: Why Every Piece of Software Infrastructure Needs to be Reimagined for AI Agents — AI agents necessitate a fundamental rewrite of software infrastructure designed for human-driven explicit instructions.
• How to Use Agent Skills — Agent skills enable reusable, dynamic AI capabilities, moving beyond static prompts for improved performance and reliability.
• OpenAI Codex Can Now Ship Live Shareable Websites — AI tools are rapidly evolving to empower non-developers with direct, integrated, and automated capabilities for complex tasks.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Latent.Space - Www.latent.space.