Tool Use - AI Conversations AI Coding Agents Can Do So Much More (ft Kiran)
17 snips
Oct 7, 2025 Discover how AI coding agents are changing software development with Kiran's insights on optimizing your development environment. Learn about the importance of feedback loops, clear pipelines, and organized code to enhance AI performance. Kiran dives into Slate’s unique context management strategies that prevent common pitfalls and help agents excel in complex tasks. Explore why traditional software engineering skills are still crucial, and gain practical tips for structuring projects that maximize AI coding capabilities.
AI Snips
Chapters
Transcript
Episode notes
Define Verification And Review Pipelines
- Give agents explicit rules for verification and clear acceptance pipelines for code review.
- Build parallel workflows (implementer agent + reviewer agent) and a clear pass/fail pipeline.
Origin Story Of Slate
- Kiran traced interest in coding agents back to ChatGPT and Stable Diffusion, exploring why models struggle with codebases.
- He built early agents inspired by Voyager and open-sourced a CLI in 2024 that evolved into Slate.
Slate Targets Hard, Long Tasks
- Slate targets long-running, production-level tasks and edge cases where other agents degrade.
- It encodes reasoning experiments and state-tracking learnings into tooling and context management.