ConeheadAI · lessons
Latest lessons
15 of 163 leftover lessons · arena=harness. Start with the takeaway, then do the exercise. Tool-use focused — use AI you already have.
- SNK3-070harnessX-style: build a tiny self-prompt loop todayRun Goal → Draft → Critique → Revise once today. That is a baby system that prompts itself.L0–L2 · Curious–Capable (bootstrap — system rebalances) · ~3m
- SNK3-066harnessX-style: tools save work — pick one patternPick one agent/tool pattern you can teach a coworker — menus of patterns are noise until one sticks.L1–L3 · Hands On–Verifier (bootstrap — system rebalances) · ~3m
- SNK3-062harnessX-style: production guardrails keywordsProduction agents need named guardrails (auth, spend, human send, logging) — not vibes.L2–L5 · Capable–Builder (bootstrap — system rebalances) · ~3m
- SNK3-060harnessX-style: human manages workflow not every keystrokeHumans manage the workflow (gates, send, risk); they do not need to type every keystroke.L1–L3 · Hands On–Verifier (bootstrap — system rebalances) · ~3m
- SNK3-057harnessX-style: routing — hard tickets vs easyRouting: classify hard vs easy work so simple tasks stay cheap and hard ones get humans/tools.L2–L5 · Capable–Builder (bootstrap — system rebalances) · ~3m
- SNK3-056harnessX-style: evaluator-optimizer loop in plain EnglishEvaluator-optimizer: score a draft against criteria, then rewrite once. That is a mini quality loop.L2–L5 · Capable–Builder (bootstrap — system rebalances) · ~3m
- SNK3-049harnessSplit research, draft, and review — even in one chatDo not let one AI pass invent the numbers it later “validates.” Separate research, draft, and review steps.L1–L3 · Hands On–Verifier (bootstrap — system rebalances) · ~3m
- SNK3-048harnessX snack: cost of endless agent loopsEndless agent loops have a cost (tokens, time, error). Budget steps; stop criteria matter.L2–L5 · Capable–Builder (bootstrap — system rebalances) · ~3m
- SNK3-043harnessName the agent pattern — most are not full autonomy“Agent” is not one product: name the pattern (assistant, tool-caller, watcher, etc.) so risk and cost match the job.L1–L3 · Hands On–Verifier (bootstrap — system rebalances) · ~4m
- SNK3-042harnessX snack: prompt vs tool-agent vs multi-step — pick the modeSeparate one-shot chat, a tool-using agent product, and multi-step modes so you pick the lightest tool that still finishes the job.L1–L3 · Hands On–Verifier (bootstrap — system rebalances) · ~3m
- SNK3-041harnessX snack: tool loops beat one-shot chat (Anthropic talk seed)Stop only one-shot prompting: a small tool loop (goal → draft → critique → re-run once) beats a single clever line — use it in chat tools you already have.L1–L3 · Hands On–Verifier (bootstrap — system rebalances) · ~4m
- SNK3-025harnessSimon prompt injection article — high value handoffSimon’s injection write-up: untrusted content can contain instructions. True/false that risk.L1–L3 · Hands On–Verifier (bootstrap — system rebalances) · ~5m
- SNK3-022harnessGoal → draft → done-when (one loop)A tiny loop (goal → draft → done-when) turns AI from chat toy into a checkable work unit.L0–L2 · Curious–Capable (bootstrap — system rebalances) · ~3m
- SNK3-012harnessPrompt injection round 2 — harderHarder injections hide in “notes” and documents; treat untrusted text as untrusted prompts.L1–L3 · Hands On–Verifier (bootstrap — system rebalances) · ~3m
- SNK3-011harnessPrompt injection — spot it (no essay)Prompt injection hijacks the model with hidden or pasted instructions — learn to spot the tell phrases.L0–L2 · Curious–Capable (bootstrap — system rebalances) · ~3m