Jared Zoneraich¶
Founder & CEO of promptlayer, a New York–based AI-engineering platform. Active prompt / agent practitioner, dog-foods his own product daily.
Links¶
- X: @imjaredz
- Site: https://imjaredz.com/
Appearances in the wiki¶
- Closing workshop at AI Engineer 2025, "How Claude Code Works" — reverse-engineered architectural tour of claude-code (see jared-zoneraich-claude-code-works-2026).
Positions¶
- Less scaffolding, more model — the claude-code-master-loop wins because it stops trying to engineer around model flaws.
- Bash is all you need — see bash-as-universal-tool; he's on the "one mega tool call" side of the future-of-tool-use debate.
- Rigorous tools, flexible loops — happy middle ground: keep the outer loop model-driven, but make individual tools eval-able like functions. Coined agent-smell as a cheap sanity metric.
- DAGs are over (mostly) — rebuilt PromptLayer's own engineering around Claude Code with an "if it takes <1h, just do it" rule.
- AI therapist problem — no global maximum in agent design; different philosophies (Claude Code vs Codex vs Cursor Composer vs Amp) win different use cases.