AI Productivity

Build the mental model

AI tools are genuinely useful for productivity, but the mental model matters: AI is an assistant, not a replacement for thinking.

It is good at drafting, transforming, summarizing, and structuring text quickly. It is not good at deciding the actual goal, what context matters, or whether a claim is accurate.

Those decisions still belong entirely to the human using the tool, every single time.

• Summarizing long documents
• Brainstorming options
• Preparing for research
• Drafting first versions of writing
• Planning a project's rough shape
• Breaking a large task into smaller ones
• Summarizing meetings
• Organizing scattered notes
• Assisting with learning new material
• Improving the clarity of your own writing
• Hallucination - confidently stated but false information
• Privacy exposure from pasting sensitive data
• Overreliance that lets your own judgment atrophy
• Summaries that quietly distort the original source
• Citations that sound real but do not exist

Hallucination

AI output that states false or fabricated information with confident, plausible-sounding language.

AI ASSISTANCE LOOP
------------------
HUMAN INTENT --> AI ASSISTANCE --> HUMAN REVIEW --> FINAL OUTPUT
                      ^                  |
                      |                  |  (problems found)
                      +------------------+

Connect it to a real scenario

Before asking an AI tool for anything, spend thirty seconds writing down what you actually need: the task, the audience, the format, and any constraints.

Give the AI real context, such as pasted source material or prior notes, instead of asking it to invent facts from nothing.

After you get a draft back, read it critically: does it actually answer the question, are any numbers checkable, do any named sources actually exist.

For anything factual, verify it against a real source before trusting it, especially statistics, quotes, and citations.

Using AI faster does not mean using it worse; the review step is what actually makes the whole workflow trustworthy enough to rely on.

• AI is an assistant for drafting, transforming, and summarizing - not a replacement for human judgment.
• The intent -> assistance -> review -> output loop keeps AI use trustworthy.
• Hallucination, privacy exposure, and overreliance are real, concrete risks worth checking for every time.

Before Trusting an AI Output

• Facts, dates, and numbers have been verified
• Sources and citations have been checked to actually exist
• No sensitive data was pasted into the tool
• The output still matches the original goal
• A human made the final judgment call

Try the working example

function isSafeToUse(draft, checks) {
  const failed = Object.entries(checks).filter(([, verified]) => !verified).map(([name]) => name);
  return {
    title: draft.title,
    safe: failed.length === 0,
    failedChecks: failed
  };
}

const verifiedDraft = {
  title: "Q3 planning summary",
  checks: {
    factsVerified: true,
    sourcesReal: true,
    noSensitiveDataPasted: true,
    matchesGoal: true,
    humanReviewed: true
  }
};

const unverifiedDraft = {
  title: "Competitor research summary",
  checks: {
    factsVerified: false,
    sourcesReal: false,
    noSensitiveDataPasted: true,
    matchesGoal: true,
    humanReviewed: false
  }
};

console.log(isSafeToUse(verifiedDraft, verifiedDraft.checks));
console.log(isSafeToUse(unverifiedDraft, unverifiedDraft.checks));

{ title: 'Q3 planning summary', safe: true, failedChecks: [] }
{
  title: 'Competitor research summary',
  safe: false,
  failedChecks: [ 'factsVerified', 'sourcesReal', 'humanReviewed' ]
}

NIST AI Risk Management Framework — Productivity Systems