The $10,000 hour. Why AI prompt engineering is the skill of 2025 and how to get across it quickly.

Today's breakthrough meeting

What a great way to round out the weekend.

I just finished a call with a client who hadn't used AI large language models before, but had access to Microsoft CoPilot at work.

The client has a herculean project that would take many teams at least 12 months to get done. A project that I would normally charge $100k+ to get done. Maybe more.

In 1 hour I taught them prompt engineering, with 15 minutes prep and no set agenda.

Using an interactive live demo in their environment and real life examples on the fly for their project we used their company's AI tool to:

• outline the project
• outline content
• do first drafts of several items

laying a path for energy and excitement that will help them build a fit-for-purpose framework for risk and assurance in their organisation.

Everything that would normally have taken 6-12 months is now attainable in 6 weeks.

Conservatively today we saved the client a month of wheel spin and hard work.

For round numbers, let's say $10k.

More likely it's $50k+ and probably a significant haircut on my future fees.

That's at least $10k of value from 1 hour invested.

That's extraordinary. We both agreed that it was at least that.

The craft of prompt engineering

Using AI is a bit like writing.

You become a better writer by writing. Writing makes you clearer in your thinking and how you communicate.

And you also become better at AI prompt engineering by doing AI interaction.

Like reading, writing, writing policy, code or great product design, clear thought and clear prompting and questions get great results.

But you've got to know what you're doing.

You can only do that by clear thought and practice.

Time to share what I've been doing

I've been using ChatGPT and the rival tools for about an hour a day (usually more), since day 6 of its public launch in November 2022. Around 2/3 of that has been in the risk and governance space to really see what each model can do.

That's getting towards 1,000 hours of experimentation.

I've put in the time to know the usual likely pitfalls and make these tools effortless in my work.

So that leaves me with a conundrum.

Do I teach others what I know and destroy my billable hour model?

At TDA our mission is to help people achieve excellence in risk and assurance.

So if it's on mission then it's in. Even if it's a commercially bad idea in the near term.

So we're going to go for it.

One-time offer - 1 hour webinar/meetup that shows what I've been doing

I'm going to run a small group paid session of what I did with my client today, tweaked for some different use cases.

If you would like to come along to a crash course on how to use large language models to do really useful stuff and watch me blow up my business in real time, you need to do three things.

1. Hit like on this post.
2. Make sure you're subscribed to the Davies Report using this link. I'll send invites from there. https://www.todddavies.com.au/latest-thinking/subscribe/
3. Keep lunchtime Friday March 7 clear to come along (Australian time) - that's Thursday pm March 6 United States).

I'll get invites and placeholders out before the event. Feel free to direct message or email me with anything you'd like me to cover.

Note: This is a session on language in tools like ChatGPT, Grok, CoPilot, Gemini etc, Maybe even NotebookLM. We won't be covering data analytics etc. We'll leave that one to the experts.