Urgent.News

What's breaking now, across thousands of outlets.

AI

Vercel built a feedback loop that treats agent instructions like software

Vercel ran more than 200 agent runs to build design.md, a new public prompt file designed to help agents create The post Vercel built a feedback loop that treats agent instructions like software appeared first on The New Stack .

Vercel built a feedback loop that treats agent instructions like software

Vercel, the company behind the popular serverless platform, has developed a new method to improve the reliability of agent-generated web pages. They created a public prompt file called design.md, which provides guidance to agents on how to design pages that match Vercel's brand. By treating agent instructions like software, Vercel was able to reduce the number of failures in six-page tests by 57%, although they acknowledge that this is not a perfect solution.

The company tested different iterations of the design.md file, using a set of seven evaluation prompts to measure the impact of changes on the generated output. The system now includes a public stylesheet, an evaluation loop that turns human feedback into updated guidance, and deterministic checks for mechanical failures. Vercel's approach demonstrates how teams can manage human judgment as reusable agent guidance, following a development lifecycle similar to software development.

Written by urgent.news from The New Stack's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at thenewstack.io →

More in AI

What is harness engineering and why should I care?

How do you ship a software product with 0 lines of manually-written code? A friend asked me this today, and I realized I didn't have a simple answer. So I dug deeper.

  • Harness engineering designs environment for AI agents to operate safely and reliably.
  • A well-engineered harness ensures AI agent runs within right parameters and prevents disruptions.

Wednesday assorted links

1. Ruxandra on AI and the remaining roles for humans. 2. The buildings of university architecture departments. Most of them I like. 3. Dean Ball on the coming of untethered AI agents. 4.

UK startup Pharosyn raises €2.5 million for pharma competitive intelligence platform

London-based Pharosyn, which builds AI infrastructure for commercial decision-making in pharma, has raised €2.5 million to expand its platform and support wider adoption across pharmaceutical…

  • UK-based AI startup Pharosyn raises €2.5 million investment
  • Moonfire leads funding round with Entrepreneurs First and others
  • Platform aims to aid pharma strategic decisions with AI infrastructure

More from Wednesday 2 September →