Skip to content

Podcast

How to Use /Goal to Do More With AI

The AI Daily Brief: Artificial Intelligence News and Analysis

Source ↗ ← All highlights
  • Slash Goal Enables Looping Agent Workflows
    • Slash Goal shifts interaction from turn-based prompts to a looping, autonomous process that can run without constant human steering.
    • Nathaniel links this to earlier ideas like side panels, voice input, and heartbeats enabling parallel agent workflows. (Time 0:02:39)
  • Slash Goal Is A Finish Line Contract
    • Slash Goal is a finish-line contract that defines what should be true, how success is checked, and what must stay intact during work.
    • Nathaniel Whittemore explains the model continuously loops, self-evaluates, and stops when evidence meets the defined finish line. (Time 0:06:32)
  • Pick Tasks With Durable Objectives And Evidence
    • Choose tasks with durable objectives, uncertain paths, and clear finish-line evidence as ideal candidates for Slash Goal.
    • Prefer tasks where completion is verifiable by tests, citations, logs, or artifacts rather than subjective ‘vibes.’ (Time 0:08:18)
  • Define Six Elements When Writing Goals
    • Write goals that specify six elements: outcome, verification surface, constraints, boundaries, iteration policy, and block/stop conditions.
    • Use tests, artifacts, and clear stop rules so the harness can self-judge progress and know when to pause or stop. (Time 0:13:52)
  • Keep Goal Scope Just Right
    • Keep goal scope in the Goldilocks zone: not too narrow (lacks flexibility) and not too broad (hard to verify).
    • Define concrete output artifacts (e.g., a docs page that builds locally) to provide an inspectable verification surface. (Time 0:14:18)
  • Goals Fit Audit‑Style Knowledge Work
    • For non-coding work, Slash Goal works best when the output is an auditable ledger or rubric-driven audit rather than a single answer.
    • Success often comes from user-provided rubrics like hiring criteria, vendor scorecards, or editorial standards the AI can test against. (Time 0:15:38)
  • Run Claim Audits With Verifiable Tables
    • Use Slash Goal to run claim audits by asking the model to verify each claim against provided and external sources and output a labeled table with citations.
    • Structure the output as supported/contradicted/partially supported/unverified plus uncertainty notes for traceability. (Time 0:18:20)
  • Make Market Landscapes Auditable Goals
    • Turn market research into a Slash Goal by specifying verification sources and a comparison table with confidence levels and evidence gaps.
    • Require cited company pages, filings, analyst reports, pricing pages, and product docs to make findings auditable. (Time 0:18:53)
  • Use Goals Selectively Along An Autonomy Spectrum
    • Slash Goal is not always the right tool; many tasks still suit traditional turn-based prompts or human-in-the-loop patterns.
    • Nathaniel emphasizes a spectrum of autonomy—choose goal when process autonomy and verifiability matter, otherwise use prompts. (Time 0:20:36)