I've been riding the IT train for over 30 years, and I've watched many signposts approach quickly and then quietly fade into the distance. Every single time, the people who got left behind weren't the ones who lacked technical skill; they were the ones who played it too safe, too slow, and too small.

AI is no different. And right now, I'm watching the cycle happen again.

Most folks are using AI the way you'd use a cautious intern: ask it to clean up an email, punch up a paragraph, maybe summarize a meeting. Careful. Incremental. Safe. I get it, my procedural/governance/security mindset often gets in my way. Well, bless their hearts; they think they're innovating. Yeah, not so much.

The people extracting real value from frontier models aren't the careful ones; they’re the ones bold enough to walk in with a moonshot prompt!

The “timidjority”

With an entire industry built around "prompt engineering," courses, certifications, libraries, etc., with so many people focused on crafting a clear/perfect prompt, and with the conversation so fixated on precision and safety rails, we trained ourselves to think small.

Ask AI to help you write a tricky email? Fine.

But ask it to architect, scaffold, and build a working tool from scratch, including the deployment config, documentation, and the tests? Now you're talkin'.

That's the gap. Not skill. Not knowledge. Audacity.

Why the harness changes everything

Think about it: a moonshot prompt in a chat window only leads to frustration, but a moonshot prompt inside an agentic harness like Claude Code or OpenAI Codex might be transformative.

The horse harness appeared around 4,000 years ago, and its design improved incrementally at a cadence measured in hundreds of years. Today’s AI harnesses are iterating at a pace measured in weeks or days!

The difference is in the execution context. In a chat interface, the model might provide great advice, but you still end up doing all the work. In an agentic environment, the model reads your codebase, writes files, runs commands, iterates on errors, and keeps going until the job is done or it hits a wall worth talking about.

That's not autocomplete. That's a collaborator with the patience of Job.

What audacity actually looks like

I'm not suggesting recklessness; I'm recommending proportional ambition. Prompts that match what these models are actually capable of, not what we assume their limits are.

To be clear, the difference between timid and audacious isn't the length of the prompt. It's the scope of what you're willing to let go of. "Write me a function" is timid. "Here's my architecture, here's my goal, build me the whole thing, flag anything that needs a human decision, and tell me what you'd do differently." That's audacious.

Start with the outcome you actually want. Let the model tell you where it needs to stop. More often than not, it won't stop where you expected.

The real skill gap in your organization

As IT leaders, we talk a lot about upskilling our teams, and not just for AI, but for all technologies we implement. But the curriculum I keep seeing is backward; it's all about controlling AI rather than trusting it. We're training people to be better supervisors of a tool when we should be training them to be bolder partners.

The teams that'll pull away over the next 18 months won't just have access to AI. They'll have a culture of audacious prompting, meaning the team has internalized that the model can handle more than they think and learned to write requests that prove it. Now, that's a culture worth building.

One thing to do this week

Open Claude Code, Codex, or whatever agentic environment you have access to. Write the prompt you've been too scared to write, the one for the thing that would take you a week to do manually. Don't hedge. Don't allow caveats to hold you back. Just tell it what you want and let it do the work.

You might be surprised how far it gets before it needs you.

That surprise? That's the skill. And right now, most people are leaving it on the table.

What's the most audacious prompt you've ever run? Drop it in the comments. I'd genuinely love to see what you are attempting. And if you're a leader trying to figure out how to build this muscle on your team, let's chat.


#AI #Leadership #FutureOfWork