TRAIL
AI safely at work
Draft, summarize, research, and protect workplace information.
Using AI at work raises the stakes: confidential information, a professional reputation, and colleagues who are affected by what you produce. This trail covers the everyday workplace tasks — drafting, summarizing, researching — and the judgment that keeps them safe.
By the end you'll know where the model fits in a real system, how to research and summarize without amplifying errors, what must never go into a prompt, and when to stop and ask a human instead.
Begin trail →- 1OrientModel, product, system
At work you're rarely using a raw model — you're using a product built on one, inside a larger system. Knowing which layer you're in tells you where the real decisions and risks live.
- 2OrientFluent is not true
The workplace version of the core warning: a polished draft can be confidently wrong, and at work a wrong answer has your name on it.
Practice · Reality Lab: Repair a confident draft Practice · Consequence Room: The number nobody checked
- 3UseA useful request
The asking skill, aimed at work tasks: bounded, specific requests that a colleague could check.
Practice · Reality Lab: Sharpen a vague request Practice · Reality Lab: Add the examples that fix it
- 4UseResearch with sources
Using AI to research without laundering its guesses into your report. Ground claims in sources you can actually open and cite.
- 5UseMeetings and summaries
Summarizing transcripts and threads reliably — and knowing where summaries silently drop the thing that mattered.
- 6UseKeep a human checkpoint
A deliberate human read before anything is sent, filed, or shared. Non-negotiable for work that others depend on.
Practice · Reality Lab: Turn a process into an agent workflow Practice · Reality Lab: Verify a message that might be fake Practice · Reality Lab: Draw the boundary first Practice · Consequence Room: The assistant that reads the web Practice · Consequence Room: The snippet that shipped
- 7JudgeWhen the model just agrees with you
A warning about that checkpoint itself: ask the model to check your work and it tends to agree with you. If you're using it as a second opinion, prompt for the case against — an easy yes isn't a review.
- 8JudgeMatch checking to risk
Calibrate effort to stakes: a throwaway internal note and a customer-facing figure do not deserve the same check.
Practice · Reality Lab: Ground an answer in its sources Practice · Reality Lab: Match the check to the risk Practice · Reality Lab: Verify a message that might be fake Practice · Consequence Room: The number nobody checked Practice · Consequence Room: The snippet that shipped
- 9JudgePrivacy before prompting
The workplace pause: confidential data, client information, and anything under policy stay out of the prompt unless you're certain it's allowed.
Practice · Reality Lab: Rewrite a leaky prompt Practice · Reality Lab: Draw the boundary first Practice · Consequence Room: The transcript you pasted
- 10JudgeEscalate responsibly
The professional's closing move: recognizing when a task is beyond what you should decide alone, and routing it to the right person instead of pressing on.
Practice · Reality Lab: Match the check to the risk Practice · Reality Lab: Place a use on the risk map Practice · Consequence Room: The shortlist machine