Five principles

Do not wait until the tools are perfect, or until you feel ready. Bring them into real work—the report that matters, the lesson that is not landing, the draft you are afraid of—where you can watch what they do to problems you actually own.

A seat at the table means real context: the draft, the constraints, the audience, the doubt. It also means real questions. Ask to be challenged, not comforted. Ask what to cut. Ask what not to add.

And remember that a table has more than one seat. Yours does not disappear because the tools sat down.

On this siteThe brief that accompanied a prototype of these pages to Claude for critique ended with a request: tell us what not to add.

The machines are fast, fluent, and indifferent. They will produce the options; they will not care which one ships. Caring is your job, and it cannot be delegated.

Being in the loop is not proofreading. It is judgment: knowing what the work is for, whom it serves, and what would betray it. Every option that reaches the world should pass through a person who can say why this one.

Sign your choices. If you cannot explain a decision, it was not yours.

On this siteEvery palette, sentence, and interaction here was chosen by a person—and the records keep the options that lost.

The tools produce sentences. Expertise decides which one is true—and notices which one is almost true, which is worse.

Everything you know—the standards of your field, the taste built by years of mistakes—did not retire when the machines arrived. It became more valuable. The person who knows what good looks like gets better work from the same tool, by asking for the right thing and refusing the wrong one.

Bring your whole training to the table. A career of judgment can fit in something as small as a comma.

On this siteOne Spanish comma was removed from these pages in review. A test now guards its absence.

Working with AI is not a trick you learn once. It is a practice, and practices change you as they change.

Keep notes on what worked. Watch your own instructions evolve: the elaborate ones that taught you what mattered, then the short ones you earned the right to write. Let the machine do more only where you have learned to check it—and never hand it your own learning.

A few months into the practice, your first prompts will embarrass you. That is the evidence you want.

On this siteThe instructions for one photograph on this site shrank from twenty-five art-directed prompts to four words.

It is easy to decide in advance what the machines cannot do: write with feeling, read with care, surprise you. Every one of those sentences now carries an expiration date, and none of them prints it on the label.

So run the experiment. Hand the tools the task you are certain they will fail, and watch closely. Keep your skepticism—it is what makes the wonder trustworthy. But leave the door open.

Wonder is not naïveté. It is information about a moving frontier. It is also, simply, fun.

On this siteDuring this site’s quality reviews, an illustration was corrected for contradicting the short story it accompanied. The reading was that close.