Escaping the prompt debt trap : Drew Breunig live with Tim O'Reilly.

Format:

Video

Contributor:

O'Reilly, Tim, presenter.

Breunig, Drew, presenter.

O'Reilly (Firm), publisher.

Language:

English

Subjects (All):

Artificial intelligence.

Physical Description:

1 online resource (1 video file (1 hr., 1 min.)) : sound, color.

Edition:

[First edition].

Place of Publication:

[Sebastopol, California] : O'Reilly Media, Inc., 2026.

Summary:

The clearest sign you're overinvesting in a prompt is "fighting the weights." This happens when the model won't do what you ask because you're working against its training. It's a common failure mode Drew Breunig saw everywhere when he and coauthor Srihari Sriraman examined how different labs write system prompts for their applications. And as labs train their harnesses into their models ( which they already appear to be doing), this problem will only become more pronounced, as the systems applied AI engineers are building no longer behave like Claude Code or ChatGPT do. Drew, writer, technology leader, and CEO and cofounder of cmpnd.ai, has been exploring the challenges of AI and context engineering on his blog and soon to come, in The Context Engineering Handbook from O'Reilly. He'll join Tim O'Reilly to discuss the prompt debt that developers accumulate and share actionable strategies that will help you shed it, drawn from the work of agentic software developers, who are, as Drew notes, "likely years ahead of everyone else when it comes to experimenting with and evolving their culture and practices to accommodate AI tools." As always, the conversation will be guided by your questions and comments, but Drew and Tim will likely touch on how to specify system behavior through rigorous measurements, evaluations, and tests, the merits of automated systems and tooling such as DSPy for optimization, and maybe even the coding agent "golf pro tips" Drew has been collecting from top-tier agentic developers. Recommended prep or follow-up: Read The Problem is Prompt Debt (blog post).

Notes:

OCLC-licensed vendor bibliographic record.

OCLC:

1611141182

Publisher Number:

0642572421823