From Prompt to Production

So… how’d you do?

If you beat the clock — congrats, you’re in the raffle and you’ve already learned more about AI than most people who use it daily.

If you didn’t — welcome to the club. Most people don’t beat it. The reason matters more than the result.

The quick story

You asked an AI for an analog clock at a specific time. It almost certainly drew you something close to 10:10.

That’s not random. Watch and clock manufacturers have photographed their products at ~10:10 for decades — it’s symmetric, it frames the brand logo, and it vaguely looks like a smiling face. Nearly every analog clock image on the internet shows ~10:10.

When an image model trains on billions of internet images, “analog clock” becomes statistically synonymous with “10:10.” The model isn’t drawing a clock — it’s drawing the platonic clock-photograph it has memorized.

This is training-data bias, and it’s the most important thing to understand about AI in production.

Why you should care (even if you’re not in AI)

The 10:10 problem is funny. The version that shows up in your actual work is not.

Ask AI for “a good API design” and you get the 2019–2022 industry average, not your team’s standards.
Ask for “a professional email about an outage” and you get the median internet outage email — bland, slightly corporate, emotionally generic. Your company’s actual voice is invisible to the model.
Ask “should we use microservices?” and you get whatever blog-post consensus dominated tech writing during the model’s training window.

The model’s confident output looks like an answer to your question. It’s actually a regression to the mean of its training data. Recognizing this is the difference between using AI and being used by it.

The cheat codes

The four prompting moves the booth host walked you through. Try them on the clock challenge at home — or, more usefully, on whatever you’re stuck on at work.

1. Decomposition

Break the impossible request into atomic steps the model has seen.

Draw an analog clock face with no hands.
Now add an hour hand pointing exactly at 2, slightly toward 3.
Now add a minute hand pointing exactly at 7 (35 minutes past).

Where this helps at work: any time the model gives you a “shaped right but wrong in the details” answer — drafting docs, writing code, summarizing meetings.

2. Negative constraints

Name the wrong answer and explicitly forbid it.

The clock hands MUST NOT show 10:10.
The hour hand is between 2 and 3, closer to 3.
The minute hand is on the 7 (35 minutes past).

Where this helps at work: any time you can predict the bad default the model will give you. “Do NOT use Python’s set() — order must be preserved.” “Do NOT include corporate filler phrases like ‘circle back’ or ‘synergy’.”

3. Reframing with an anchor

Add a second representation of the same information so the output has to match it.

Generate an analog clock face showing the same time as a digital readout
displaying "2:37" right next to it. Both must show the same time.

Where this helps at work: structured outputs (give it the JSON schema), tone control (give it 3 example emails in your voice), formatting (give it the template).

4. Knowing when to give up

Sometimes no prompt works, and the production-grade move is a different tool.

The clock-at-2:37 problem is genuinely solvable in ~10 lines of code. Wrestling with prompts for 20 minutes when the answer is matplotlib is the enthusiast move, not the practitioner move.

Where this helps at work: if you’ve tried decomposition, negative constraints, and anchoring and you’re still not getting reliable output — stop. The next prompt is rarely the answer. The next tool often is.

Want to go deeper?

Internal: Watch the AI Brownbag — Demystifying AI and Learning How to Prompt — covers prompting fundamentals in more depth. (link TBD)
The 10:10 phenomenon: “Why do analog watches always show 10:10 in ads?” — the original cultural artifact behind the bias.
Applied prompting: Anthropic’s Prompt Engineering guide · OpenAI’s Best Practices for Prompt Engineering

Some links above point to external sites. Internal SAP links will be added before the event.

One last thing

The model’s output is biased toward its training data. Your job as a practitioner is to steer past that bias deliberately.

That’s it. That’s prompting tradecraft. Everything else is just patterns layered on top of that one core idea.

Thanks for playing. Now go beat the clock at home — and send us a screenshot if you win.

— The d-com “From Prompt to Production” booth team
The Support Business Network Team · Pittsburgh Office