To try

Investing — investing

• Don’t use leverage (margin, loans, leveraged funds) for long-term investing — leverage · lazy-investing—108-the-leverage-trap
• Avoid leveraged ETFs on any index — leverage · lazy-investing—108-the-leverage-trap
• Before investing, define volatility tolerance, emergency fund, index, equity exposure, fees, horizon, goals — index-investing · lazy-investing—108-the-leverage-trap
• Check top holdings and weights of any index fund you buy — concentration-risk · lazy-investing—108-the-leverage-trap
• Start investing young in broad indexes and let compounding work — compound-interest · lazy-investing—108-the-leverage-trap
• Get AI exposure through broad indexes, not AI-themed or leveraged AI ETFs — market-bubbles · lazy-investing—111-the-ai-bubble
• Don’t sell or delay investing over bubble talk; decide by horizon and volatility tolerance — market-timing · lazy-investing—111-the-ai-bubble
• Compare 5.6% vs 11% over 30 years in a compound-interest calculator — compound-interest · lazy-investing—111-the-ai-bubble

AI Engineering — ai-engineering

• Never let the agent install external skills; have it rewrite them in its own words — ai-guardrails · osim-tochna—second-brain-and-llm-wiki

• Enforce critical rules with hooks/checks, not instructions alone — ai-guardrails · osim-tochna—second-brain-and-llm-wiki

• Debrief every unexpected agent behavior and add a preventing mechanism — ai-guardrails · osim-tochna—second-brain-and-llm-wiki

• Fact-check by opening and quoting each source — ai-verification · osim-tochna—second-brain-and-llm-wiki

• Before agent-proposed deletions, demand proof of backup — ai-verification · osim-tochna—second-brain-and-llm-wiki

• Use 3 query guesses + rank 3–4 candidates for wiki retrieval — wiki-retrieval · osim-tochna—second-brain-and-llm-wiki

• Diagram what your agent loads per request; keep CLAUDE.md small — context-engineering · osim-tochna—second-brain-and-llm-wiki

• When an agent fails, ask which capability it lacks to verify itself, and build it — harness-engineering · langtalks—73-harness-engineering

• Track productivity per token over time — harness-engineering · langtalks—73-harness-engineering

• Keep product docs as files in the repo with an index prompt — agent-ready-codebase · langtalks—73-harness-engineering

• Add “when to update” to every skill description — agent-ready-codebase · langtalks—73-harness-engineering

• Let agents launch isolated per-branch environments via CLI — agent-ready-codebase · langtalks—73-harness-engineering

• Have agents write and freeze failing tests before implementing — test-driven-development · langtalks—73-harness-engineering

• Require proof-of-work (e.g. Playwright video) for agent-built features — ai-verification · langtalks—73-harness-engineering

• Git hook blocking commits below 80% coverage — ai-guardrails · langtalks—73-harness-engineering

• When correcting an agent, state the reason so it persists — context-engineering · langtalks—73-harness-engineering

• Write custom skills for your org’s deploy/design/test/review norms — agent-ready-codebase · langtalks—73-harness-engineering

• Use batch/async inference for latency-insensitive bulk workloads — llm-inference · ai-engineering-podcast—why-your-llm-costs-so-much

• Narrow and rephrase the task before upgrading the model — model-selection · ai-engineering-podcast—why-your-llm-costs-so-much

• Keep each small-model call to ≤2–3 logic steps — model-selection · ai-engineering-podcast—why-your-llm-costs-so-much

• Set reasoning-token budgets where available — model-selection · ai-engineering-podcast—why-your-llm-costs-so-much

• Benchmark an open-weight model on your workload — open-weight-models · ai-engineering-podcast—why-your-llm-costs-so-much

• Wait a month before adopting a new AI tool — ai-hype · ai-engineering-podcast—why-your-llm-costs-so-much

• Check product ideas: could a prompt recreate this in 10 minutes? — ai-hype · ai-engineering-podcast—why-your-llm-costs-so-much

• When a session degrades after compaction, dump state to MD and restart — context-engineering · ai-engineering-podcast—why-your-llm-costs-so-much

• Least-privilege access for agents (no destructive DB rights) — ai-guardrails · ai-engineering-podcast—why-your-llm-costs-so-much

• Isolate parallel agents; no unsupervised agent-to-agent messaging — ai-guardrails · ai-engineering-podcast—why-your-llm-costs-so-much

• Start with (agentic) RAG + expert-graded evals before anything fancier — rag · ai-engineering-podcast—company-brain

• Route all LLM calls via a gateway with per-feature spend limits + alerts — ai-gateway · ai-engineering-podcast—ai-infra-at-scale

• Strip PII before prompts leave your system — ai-gateway · ai-engineering-podcast—ai-infra-at-scale

• Keep a pre-tested fallback model/vendor per AI feature — ai-gateway · ai-engineering-podcast—ai-infra-at-scale

• Small offline eval set in CI on every PR — llm-evals · ai-engineering-podcast—ai-infra-at-scale

• Code checks for deterministic criteria; LLM-as-judge for subjective — llm-evals · ai-engineering-podcast—ai-infra-at-scale

• Continuously score sampled production sessions — llm-evals · ai-engineering-podcast—ai-infra-at-scale

• A/B test new models via gradual rollout (quality + cost) — llm-evals · ai-engineering-podcast—ai-infra-at-scale

• Stable-first prompt order for caching; no leading timestamps — llm-cost-optimization · ai-engineering-podcast—ai-infra-at-scale

• Constrain output length/structure — llm-cost-optimization · ai-engineering-podcast—ai-infra-at-scale

• Plan with a strong model, execute with a cheaper one — llm-cost-optimization · ai-engineering-podcast—ai-infra-at-scale

• Give each new AI capability a visible entry point — proactive-ai · ai-engineering-podcast—the-ai-ux-paradox

• Deliver AI answers where users already ask, privately first — proactive-ai · ai-engineering-podcast—the-ai-ux-paradox

• Cluster real user intents and turn known ones into eval sets — llm-evals · ai-engineering-podcast—the-ai-ux-paradox

• Pre-filter large candidate sets with a cheap classifier before agent reasoning — decision-classifiers · langtalks—74-jev

• Don’t decompose conversational agents into classifier decision trees — decision-classifiers · langtalks—74-jev

• Let your assistant learn nightly from your corrections — personal-ai-assistants · langtalks—72-personal-assistant-agent

• Give agents their own identities in shared tools — personal-ai-assistants · langtalks—72-personal-assistant-agent

• Route internal-agent complaints into structured issues an agent can fix — personal-ai-assistants · langtalks—72-personal-assistant-agent

• Composable skills + per-user workflow layer — skill-engineering · langtalks—72-personal-assistant-agent

• One versioned shared skills repo with a contribution norm — skill-engineering · langtalks—72-personal-assistant-agent

• One container per agent; only mounted dirs persistent — agent-security · langtalks—71-claw-architectures

• Keep API keys out of agent environments via an outbound credential proxy — agent-security · langtalks—71-claw-architectures

• Human approval for outbound actions at the network layer — agent-security · langtalks—71-claw-architectures

• Make projects agent-operable: CLAUDE.md + CLI — skill-engineering · langtalks—71-claw-architectures

• Harness-agnostic workflow plugin (Superpowers) — coding-agent-workflow · langtalks—70-our-claude-code-tips

• High-level technical plans, rendered as HTML — coding-agent-workflow · langtalks—70-our-claude-code-tips

• “Context recap” output style — coding-agent-workflow · langtalks—70-our-claude-code-tips

• Raise skill-listing budget; extend session cleanup — coding-agent-workflow · langtalks—70-our-claude-code-tips

• Nightly routines for heavy reviews — coding-agent-workflow · langtalks—70-our-claude-code-tips

• Preview environment per agent PR — harness-engineering · langtalks—70-our-claude-code-tips

• Explicit self-checkable definition of done — ai-verification · langtalks—70-our-claude-code-tips

• Dedicated browser profile for the agent — agent-security · langtalks—70-our-claude-code-tips

• Publish a skill + MCP for your product — agent-ready-products · langtalks—69-marketing-for-agents

• Submit to awesome lists, ClawHub, ChatGPT/Claude marketplaces — agent-ready-products · langtalks—69-marketing-for-agents

• CLI with OAuth login so tokens stay out of agent context — agent-ready-products · langtalks—69-marketing-for-agents

• Define SDLC source of truth + human feedback points first — ai-sdlc · langtalks—68-ai-sdlc

• Start AI-SDLC with bug triage and on-call investigation — ai-sdlc · langtalks—68-ai-sdlc

• Route agent work to human review by blast radius — ai-sdlc · langtalks—68-ai-sdlc

• Map and eliminate copy-paste steps in the dev loop — harness-engineering · langtalks—68-ai-sdlc

• Require use-case metadata on every LLM call (proxy/gateway) — llm-cost-optimization · langtalks—67-finops-for-ai

• Review heaviest/lightest AI spenders’ sessions — llm-cost-optimization · langtalks—67-finops-for-ai

• Full unit economics per AI feature — llm-cost-optimization · langtalks—67-finops-for-ai

• Slim tool responses before they hit the model — llm-cost-optimization · langtalks—67-finops-for-ai

• Define KPIs before the first AI experiment — llm-pipelines · langtalks—66-scaling-llmops

• Pipeline + reasoning + confidence + ground-truth check for high-volume features — llm-pipelines · langtalks—66-scaling-llmops

• Version prompts with their context-building code in a shared SDK — llm-pipelines · langtalks—66-scaling-llmops

• Cross-region inference to raise rate limits — llm-cost-optimization · langtalks—66-scaling-llmops

• Normalized data + purpose-built tools for investigation agents — ai-sre · langtalks—65-ai-sre

• Inject a validated topology map into agent context — ai-sre · langtalks—65-ai-sre

• Shadow-run new versions beside production — llm-evals · langtalks—65-ai-sre

• Baseline engineering productivity before AI attribution — ai-engineering-metrics · langtalks—64-ai-coding-metrics

• Track code survival of AI vs human code — ai-engineering-metrics · langtalks—64-ai-coding-metrics

• Rising success baseline per custom-agent version — llm-evals · langtalks—64-ai-coding-metrics

• Start a GenAI guild of early adopters — ai-rd-rollout · langtalks—62-ai-rd-rollout

• Run a “no hand-written code” sprint — ai-rd-rollout · langtalks—62-ai-rd-rollout

• Give PMs AI read access to the codebase — ai-rd-rollout · langtalks—62-ai-rd-rollout

• Separate AI validation step per generated artifact — ai-verification · langtalks—62-ai-rd-rollout

• State machine around realtime voice models — voice-agents · langtalks—61-voice-agents

• Route complex voice intents to an existing text engine via one tool — voice-agents · langtalks—61-voice-agents

• Judge voice calls from audio, not transcripts — voice-agents · langtalks—61-voice-agents

• Benchmark a fine-tuned small model vs your LLM for segmentation/classification steps — small-language-models · langtalks—59-slms

• Compose real documents into synthetic labeled training data — small-language-models · langtalks—59-slms

• Check licenses before LLM-labeling training data — small-language-models · langtalks—59-slms

• Auto-generate sub-agent capability descriptions from their actual tools — multi-agent-orchestration · langtalks—58-reinvent-predictions

• Track rule usage/acceptance; prune low-value rules — agent-memory · langtalks—57-memory

• Per-source-type memory ingestion with links to originals — agent-memory · langtalks—57-memory

• Backfill memory from historical PR discussions — agent-memory · langtalks—57-memory

• Build automations one reliable step at a time — workflow-automation · langtalks—56-n8n

• Assign an automation owner for non-technical teams — workflow-automation · langtalks—56-n8n

• Sub-agents read large files, return only answers/snippets — context-engineering · langtalks—55-context-engineering

• Specific, actionable API errors for agents; keep errors in context — context-engineering · langtalks—55-context-engineering

• Trace audit: trim context you scroll past — context-engineering · langtalks—55-context-engineering

• Files as agent working memory before vector memory — agent-memory · langtalks—55-context-engineering

• Stable tool list + logit masking instead of tool swapping — llm-cost-optimization · langtalks—55-context-engineering

• Give an autonomous agent an outcome + tools; ask what it learned — ai-growth-marketing · ignore-instructions—28-ai-marketing-enso

• Feed post performance back into AI copywriting — ai-growth-marketing · ignore-instructions—28-ai-marketing-enso

• Weigh one horizontal AI platform vs per-function point solutions — enterprise-ai-adoption · ignore-instructions—24-wonderful-road-to-100m

• Ship side projects through widening feedback circles (10 → 100 → wider) — solo-builders · ignore-instructions—20-idan-benyon-build-ship-grow

• Open a WhatsApp beta-tester group for anything you build — solo-builders · ignore-instructions—20-idan-benyon-build-ship-grow

• Finish one side project end to end, even a symbolic launch — solo-builders · ignore-instructions—20-idan-benyon-build-ship-grow

• Build in public: post weekly about what you build — ai-growth-marketing · ignore-instructions—20-idan-benyon-build-ship-grow

• Before fine-tuning, split knowledge into memorize (needed for planning) vs retrieve — continual-learning · explainable—164-gilad-levi-continual-learning

• For any AI product, write down what value survives a competitor swapping in a better model — ai-moats · explainable—164-gilad-levi-continual-learning

• Audit which skills, hooks, and plugins load into every agent session; drop the unused ones — llm-cost-optimization · explainable—163-hidden-cost-of-agents

• Write a fallback plan for when your primary model vendor is down, degraded, or blocks you — ai-gateway · explainable—163-hidden-cost-of-agents

• Review AI-written tests for meaning, not just green status — ai-verification · explainable—163-hidden-cost-of-agents

• Adapting a model to a new language: compare tokenizer tokens-per-word first — hebrew-llms · explainable—157-training-hebatron

• Use loss and benchmark drops as alarms, but pick releases by user preference — llm-evals · explainable—157-training-hebatron

• Add automated security testing to the release process for agent-written code — ai-cybersecurity · ignore-instructions—26-ai-cyber-pavel-gurevich

• Raise pentest frequency to quarterly or monthly and test integrated staging — ai-cybersecurity · ignore-instructions—26-ai-cyber-pavel-gurevich

• Check whether bug-prone code came from spec-driven sessions and used internal tools — ai-engineering-metrics · langtalks—64-ai-coding-metrics

• If adopting a new model provider, keep a fallback to your existing LLM path — decision-classifiers · langtalks—74-jev

• List workflow decision points (triage, lead quality, routing) and try a decision model on one — decision-classifiers · startup-ideas—jev-is-here

• Test whether an agent can complete your product’s top 3 user journeys — agent-ready-products · startup-ideas—webmcp-clearly-explained

• Add a five-line daily brief (done / blocked / needs me) to each recurring agent — multi-agent-orchestration · startup-ideas—grok-bot-one-person-company

• Run one project with an agent team for a month: build, execute, hire/fire, automate — solo-builders · startup-ideas—grok-bot-one-person-company

• When grilling an idea, have the agent log every “probably” as an assumption and rank which to verify first — coding-agent-workflow · startup-ideas—8-github-repos-viral-money

• Run a design-critique skill (e.g. Impeccable’s critique) on a page you already built before asking for fixes — coding-agent-workflow · startup-ideas—8-github-repos-viral-money

• Before installing a third-party skill or repo, read which agent settings it changes and test it in a scratch project — skill-engineering · startup-ideas—8-github-repos-viral-money

• Make a 20-second launch video for something you built (e.g. with Brag) and share it — ai-growth-marketing · startup-ideas—8-github-repos-viral-money

• Build one free calculator or tool for your audience; show the result before asking for an email — ai-growth-marketing · startup-ideas—8-github-repos-viral-money

• Try a builder + checker agent pair on a small job (e.g. a quote calculator checked against real pricing) in a test project — multi-agent-orchestration · startup-ideas—8-github-repos-viral-money

• Have an agent search Reddit/X/YouTube for “I export it” / “workaround” posts about a paid tool and bring back originals — customer-pain-research · startup-ideas—8-github-repos-viral-money

• Before building, contact one poster and ask them to show their workaround — customer-pain-research · startup-ideas—8-github-repos-viral-money

• Before building an agent, classify the job and solve only that problem type — agent-architectures · startup-for-startup—362-agent-architecture-fit

• Give agents the goal and review their plan instead of dictating micro-tasks — ai-rd-rollout · startup-for-startup—362-agent-architecture-fit

• Rewrite “don’t” rules in agent prompts as positive instructions — context-engineering · startup-for-startup—354-reliable-agents-lean-context

• For vision features, test empty/black and adversarial-text images before launch — multimodal-llms · osim-tochna—ai-in-production-reality-vs-imagination

• Try visual hints (marking the region of interest) before upgrading to a pricier model — multimodal-llms · osim-tochna—ai-in-production-reality-vs-imagination

• Verify any personal agent (OpenClaw-style) isn’t internet-reachable and has its own phone number — agent-security · osim-tochna—ai-in-production-reality-vs-imagination

• Name an owner for AI adoption and run hands-on sessions on real code — ai-rd-rollout · osim-tochna—gong-ai-for-developers

• Write an agent policy before launch (what it may say/do, competitors, regulated advice) — ai-red-teaming · hidden-layers—alice-avi-golan

• Map each agent against the lethal trifecta and remove a leg where possible — agent-security · hidden-layers—alice-avi-golan

• Audit permissions before rolling out a workspace AI assistant — agent-security · hidden-layers—cyera-shiran-bareli

• For write-capable agents, approve a deterministic plan and keep audit logs plus undo — ai-guardrails · hidden-layers—twine-nadav-erez

• Route privacy-sensitive subtasks to a local open model via a skill — open-weight-models · hidden-layers—nexar-roni-goldshmidt

• Learn inference by running a small model locally (Ollama, then vLLM) and comparing throughput with and without continuous batching — llm-inference · osim-tochna—running-llms-at-scale

• Keep fixed system prompts identical across requests so their KV cache can be shared — llm-inference · osim-tochna—running-llms-at-scale

• Match cloud commitments to actual usage and steer the autoscaler’s spot/on-demand mix by commitment utilization — llm-cost-optimization · osim-tochna—running-llms-at-scale

• Start agent adoption with one concrete problem (e.g. bug duty) and its guardrails before expanding — ai-sdlc · ai-engineering-podcast—atlas-ai-teammate

• Encode team standards as machine-checkable rules and enforce them at spec, pre-commit, and PR — ai-guardrails · ai-engineering-podcast—atlas-ai-teammate

• Replay each production failure as a test; if the judge doesn’t fail it, fix the judge — llm-evals · startup-for-startup—366-qa-for-agents

• Run each agent test many times and track pass rate per category over time — llm-evals · startup-for-startup—366-qa-for-agents

• Model margin per service and the expected usage mix before choosing a pricing model — ai-pricing · startup-for-startup—369-ai-pricing

• Consider committed per-user credit buckets with expiring credits to fund deep discounts, plus list-price pools for production use — ai-pricing · startup-for-startup—369-ai-pricing

• Before building an agentic workflow, write down the human process step by step, then hand each step to the agent with explicit bounds — agentic-vulnerability-response · hidden-layers—cato-networks-avishai-zvinsky

• Ship new automated detections in silent mode and validate them against historical data before they block anything — agentic-vulnerability-response · hidden-layers—cato-networks-avishai-zvinsky

• Add behavioral detections on pace and sequence (e.g. who runs PsExec) on top of signatures — ai-cybersecurity · hidden-layers—cato-networks-avishai-zvinsky

• After a session that taught the agent something new, ask it to record the lessons as skills or rules and commit them — agent-ready-codebase · hidden-layers—cato-networks-avishai-zvinsky

• Before building, have the agent grill you with questions to sharpen the spec — coding-agent-workflow · developers-outside-the-box—185-lets-talk-about-our-tools

• Give coding agents current library docs (e.g. Context7 MCP) when a library changed after the model’s cutoff — context-engineering · developers-outside-the-box—185-lets-talk-about-our-tools

• Move a business rule from the prompt into an enforced policy whose block message explains the reason, then delete it from the prompt — ai-guardrails · startup-for-startup—370-agentos-month-to-day

• If business people queue agent tweaks through engineers, give them a declarative agent definition (prompt, skills, tools, policies) they can edit and ship themselves — harness-engineering · startup-for-startup—370-agentos-month-to-day

• When a new tool makes you feel behind, spend an hour on a tiny POC to learn its core idea — ai-hype · startup-for-startup—370-agentos-month-to-day

Knowledge Management — knowledge-management

• Ask the AI to interview you to personalize your second brain — second-brain · osim-tochna—second-brain-and-llm-wiki
• Sketch your life domains on paper before deciding what goes into the system — second-brain · osim-tochna—second-brain-and-llm-wiki
• Use Obsidian’s graph view to find orphan notes — second-brain · osim-tochna—second-brain-and-llm-wiki
• Add a routine checking the index is current and raw inbox is processed — knowledge-rot · osim-tochna—second-brain-and-llm-wiki
• Periodically review whether each scheduled routine is still needed — knowledge-rot · osim-tochna—second-brain-and-llm-wiki
• After restructuring, ask the agent to verify links and routines still work — llm-wiki · osim-tochna—second-brain-and-llm-wiki
• Turn each newly learned practice into a reusable skill or rule — second-brain · osim-tochna—second-brain-and-llm-wiki
• Map data sources per concrete use case before building a company brain — second-brain · ai-engineering-podcast—company-brain
• Build team-level context before company-wide — second-brain · ai-engineering-podcast—company-brain
• Record decisions with their reasons (the “missing why”) — second-brain · ai-engineering-podcast—company-brain
• Weight sources by type/authority — second-brain · ai-engineering-podcast—company-brain
• Add a nightly memory-consolidation job (merge, update, prune) — memory-consolidation · ai-engineering-podcast—company-brain
• Feed agent task outcomes (CI, rollbacks) back as context — memory-consolidation · ai-engineering-podcast—company-brain
• Task-aware memory extraction — human-vs-ai-memory · langtalks—60-brain-memory
• Resolve memory conflicts by frequency, not just recency — human-vs-ai-memory · langtalks—60-brain-memory
• Prototype a team brain locally (note-taker + Slack MCPs into a small git wiki) before automating — second-brain · startup-for-startup—353-self-updating-team-brain
• Run a retro on each finished agent session and write the lessons back to memory — memory-consolidation · ai-engineering-podcast—atlas-ai-teammate

Unsorted

• Don’t act on investment success stories from Instagram/TikTok — fomo · lazy-investing—108-the-leverage-trap
• Try a CGM or ketone meter for a few weeks to see real responses to meals or fasting — biohacking · explainable—152-biohacking-like-a-data-scientist
• Self-experiments: alternate conditions A-B-A-B to reduce time effects — biohacking · explainable—152-biohacking-like-a-data-scientist
• Consider whole-genome sequencing and check pharmacogenomic variants before you need the drugs — personal-genomics · explainable—152-biohacking-like-a-data-scientist
• Ask whether a medical risk number accounts for evidence you already have — bayesian-updating · explainable—152-biohacking-like-a-data-scientist