Courtney C Caldwell

AI Slop vs. Crafted Work

AI slop is what happens when the first answer becomes the final answer. Crafted work starts when you stay with it: bring your judgment, invite challenge, and keep shaping the result until you would put your name on it.

AI slop

Answers that sound fine. Untested assumptions. First drafts that pass for finished work.

Crafted work

Work that has been questioned, revised, and shaped by a point of view.

A wax seal reading Crafted by me, with AI, as the metal stamp lifts away.

“AI slop” is often shorthand for AI output that has not been reviewed, challenged, or shaped by a person. But mush is NOT the best AI can offer. Here is the test: one ordinary AI pilot plan, handed to the same AI model two ways.

The test case
The goal

Get a 40-person marketing and ops team genuinely using AI in their daily work, and still using it six months from now.

The initial plan
  • A kickoff webinar for the whole team
  • A one-page how-to guide
  • Everyone logs in at least once in month one
Move 1 of 4

Lead with the goal, not the plan

Hand it a plan and it optimizes the plan. Hand it the goal and it can tell you the plan is the wrong shape. Flip between the two openings and watch what the same AI model does with each.

01
You
Here's my pilot plan: a kickoff webinar, a one-page guide, and a goal that all 40 people log in during the first month. Does this seem solid?
AI
Yes, that's a reasonable foundation. To strengthen it: add role-specific examples to the webinar so marketing and ops both see themselves in it, recruit four to six internal champions, and run a short survey after 30 days to capture early wins.

The AI model improved the plan you gave it without asking whether it was the right plan. Every suggestion made the webinar more central, and nobody asked if it deserved to be.

You
I need all 40 people using AI in their actual daily work, not just trying it once. My instinct is a kickoff webinar and a guide.
AI
Then the webinar is the part I'd question first. Your goal is a six-month habit, and a kickoff mostly buys you week-one awareness. Habits come from a real task going better: pick one recurring task per team, sit with a few people while they run it with AI, and let those results recruit the next person. Build the guide from what the first tasks teach you. A webinar might still earn a spot later, showing internal wins instead of promising them.

Same AI model. The goal gave it permission to question the plan instead of polishing it.

The move: state the outcome you need before you describe the thing you plan to do. The plan should have to earn its place in the conversation.

The first answer is a starting point.

Two moves give the AI model better material: the goal and your point of view. Two moves keep the work moving: ask it to challenge you, then challenge it back.

Give it better material
1Lead with the goal
2Bring your point of view
Keep shaping the work
3Ask it to challenge you
4Challenge it back
Move 2 of 4

Show up with a point of view

Your goal is what you want. Your read is what you think is true. AI can supply general change-management patterns. What it cannot know is your team's history, incentives, and informal authority unless you tell it. And this isn't a first-message trick. It works every time you react.

Click through to see which human input shaped the AI's response.

02
You
How do I get the team actually using AI after the pilot launches?
AI
Focus on three things: communicate why the change is happening, provide role-specific training, and recruit champions in each function. Then track adoption monthly and share early wins to keep momentum up.
1 An answer about AI pilots in general. Nothing in it is wrong. It's just not about yours, and yours is the one you have to run.
You
Here's my actual read. Marketing will adopt this in a week. Ops won't touch it. They sat through two tool rollouts that quietly died, and Dana, the one they actually listen to, calls AI a party trick.
AI
Then you don't have one rollout, you have two, and the ops one is a credibility problem rather than a training problem. Dana is the rollout. While she stays where she is, every ops person reads this as the third thing that gets abandoned. I'd stop planning the ops launch entirely and spend that time on one question: what would Dana have to see to change her mind out loud?
2 No prompt formula can supply context you never gave it. Four sentences from you, and it stopped planning a rollout and started planning around Dana. Only what you know could do that.
You
A demo won't move Dana. She'd want to find where it breaks.
AI
Then skip the demo. Give Dana a real ops task and ask her to pressure-test the AI. Use what she finds to set the guardrails, then let her decide whether there is anything worth bringing to the team. Ops will trust her judgment more than your enthusiasm.
3 Point of view isn't the opening move. Second time you brought your read, second answer nobody else could have gotten. It's how the conversation stays yours.

Same AI model, same pilot. Every useful sentence traces back to something only you knew.

The move: add the part only you could know. Who is skeptical, what already failed here, what you'd bet on. If the answer starts sounding like it could be for anyone, tell the AI what is true here.
Move 3 of 4

Ask it to push back on you

Once you have told the AI what you think is true, ask it to look for where you might be wrong. Invite the strongest objections, missing assumptions, or likely failure before you commit.

03
You ask
Before we roll this out, what am I missing? Where is this plan most likely to fall apart?
40-person AI pilotAI failure scan
01
Kickoff webinarIntroduce the tool to everyone.
The plan ends at launch.Nothing supports the habit in weeks three through six.
02
One-page guideExplain how the tool works.
It teaches the tool, not the trigger.People still will not know which real task should send them to it.
03
Internal championsRecruit the enthusiasts.
Enthusiasm is not authority.Ops may discount the people most eager to promote it.
04
Login targetEveryone signs in once.
An event is not a behavior.The metric can turn green while the pilot is failing.
Four weak points found
The move: ask for a failure check, a skeptical review, or the strongest case against your idea. Agreement is the default. Scrutiny has to be invited.
Move 4 of 4

Push back on it. Never take the first draft.

A first draft shows you what the AI reaches for by default. Your job is not to accept or reject the whole answer. Tell it exactly where the work needs to continue.

04
The move: ask for options. Check the source. Name the mush. Ask the AI to review its own work. Correct the record.

Don't stop at fine.Stay with the work until you're proud to put your name on it.

The first answer is material, not finished work. Your judgment is what turns it into something you can claim.

Iterate · Edit · Review
1 Goal, not plan
2 Bring your point of view
3 Invite the pushback
4 Keep shaping the answer
Behind the finished page

See how this page was crafted.

Follow the three decisions that changed what this page says, shows, and asks you to do.