Continuing with our theme of getting 10x more out of AI, I hosted a webinar on Tuesday for AI-curious leaders.

In it, I shared how to become the kind of confident AI leaders our ​Executive Boot Camp​ graduates transform into over the 3 weeks they spend with us.

Because while other ‘gurus’ are making AI difficult and unapproachable, I want to emphasize how AI is a new, friendly team member, just one that continues to become more and more capable over time.

If you couldn’t attend, here are the five simple steps to take so you too can be an AI leader.

The Five Shifts to Confidently Lead with AI

I agree: the acceleration of AI is the biggest we will see in our lifetimes.

We’ve moved quickly from AI as a simple chat interface (type in, get text back) to multimodal systems (voice, images, video) and are now in the agentic era where AI autonomously plans and acts.

That escalation raises a leadership question: how do we manage a hybrid team of people and AIs?

Two framing points set the stage:

• The “Superworker” (as ​Josh Bersin​ calls them) effect is real: individuals paired with strong AI ​outperform traditional teams​ on knowledge work, research from Ethan Mollick shows.

• Organizations are institutionalizing AI: hiring managers prioritize AI skills, and leaders like Shopify’s CEO have made ​everyday AI use a baseline expectation​.

With that context, I shared five shifts people make to move from AI-Curious to Confident AI Leaders, from mindset to supervision.

1: Treat AI as a colleague and brief it like one

Stop thinking of ChatGPT as a “tool.” Treat it as a capable colleague you need to manage, by setting its role, intent, and standards so it can produce work you would sign off on.

As I’ve repeated, good prompting is simply good delegation.

The ​CODO-Superprompt framework​ gives you a structure to get 10x more out of your prompts (see our free ​prompt generator​ to get started):

• Character: Who should the AI be for this task? (e.g., experienced copywriter, board‑level advisor).
• Objective: The outcome to enable, not the artifact (“convince my board to invest in …,” not “write a summary”).
• Do’s & Don’ts: What good work looks like; frameworks, rules, and known traps to avoid based on your unique experience.
• Output: The exact deliverable (one‑slide summary, risk–reward table, 500‑word memo) and an example if you have one.

Good prompts have become even more important with the introduction of ​ChatGPT-5​.

​Prompting ChatGPT-5​ is different because the model is more steerable and more of a rule‑follower.

It also operates behind a router that chooses between two sub‑modes unless you instruct it otherwise, as ​I shared last week​:

• Fast (chat) for quick responses.
• Thorough (thinking) for deeper, more deliberate work.

Because the router decides, vague instructions can be sent to the wrong sub‑mode. The fix is explicit management within your prompt:

• Mode: “Use thorough reasoning.” (Or, if speed matters, say so.)
• Tools: State tool use allowances/limits (“Web search allowed; stop after first 5 sources for review.”)
• Autonomy & checkpoints: Require a brief plan, interim summaries, and “pause for approval” gates.
• Depth vs. brevity: You can ask for short outputs but deep internal reasoning.
• Scope control: Limit the context you provide; the model is precise and can be tripped by conflicting signals.

2: Start from the work, not the tools

​Tasks, not tools​.

​Tasks, not tools​.

​Tasks, not tools​.

I sometimes feel the urge to let people write this on a chalkboard 100 times over, as many leaders start by shopping for platforms or signing up for tools recommended to them.

The much more reliable path is to inventory your actual work and prioritize where AI is natively strong. I use a simple filter: look for tasks that are ​General, Error‑friendly, and Digital​. That’s where AI compounds quickly.

High‑leverage categories that came up repeatedly:

• Research & analysis: Competitor scans, policy landscapes, pre‑read briefings.
• Writing: Board updates, customer proposals, investor notes.
• Preparation & rehearsal: Interview practice in voice mode, objection‑handling drills, and difficult conversations.

The discipline is to select a few workflows, introduce AI, and get excellent at them, rather than spreading your effort thinly across many tools.

Example: ​Boot Camp​ graduate ​Christine Kasoulis​, a senior UK-based retail executive (ex‑Sainsbury’s) mapped her calendar, identified research and interview preparation as the most impactful use cases, and paired core ChatGPT workflows with NotebookLM for deep research.

3: Use AI to expand capability, not only efficiency

I know it’s tempting to use AI as a productivity booster. And we should, especially in the context of those G.E.D. workflows where AI can be so helpful.

But eventually, capability expansion is the real destination, the place where we become those real AI Leaders managing a team of AIs.

Some low-hanging fruit in this realm are Deep research, commissioning multi‑source, citation‑rich briefings of up to tens of pages, and Presentation design, outsourcing formatting and layout to AI, while keeping your human judgment on message and narrative.

These new capabilities change the quality of your work, decisions, and communication, not just the speed.

Example: ​Kristen Cuneo​, former L&D lead at X, The Moonshot Factory (Alphabet’s innovation arm), refused to accept generic visuals for her new “​Speaking Spanishish​” blog.

She A/B‑tested multiple image generators on an intentionally quirky brief (“chinchillas learning about dust baths”) to pick a style that matched her brand.

Then she built a custom GPT tuned to that style so future images arrive on‑brand by default—an example of moving beyond speed to distinctiveness.

4: Move from off‑the‑shelf use to tailored assistants

Generic assistants create generic outcomes.

Superprompting is a first step, but Confident AI Leaders quickly run into the challenge that they spend a lot of time prompting.

This is where they tap into Custom GPTs and other AI Assistants, like the one from Kristen Cuneo above, to have ‘shortcuts’ to pre-prompted AI assistants tailored to their specific workflows and needs.

In our ​PRO Community​, we recently spoke to Moderna’s Head of AI, ​Brice Challamel​, who guided their team to create over 3,000 of these AI assistants:

One thing that came up during the session was when to use GPTs (​instructions here​), and when to use Projects. Here’s what I advise:

• Projects (ChatGPT): Best for ongoing workstreams. You get custom instructions plus specialized memory across all chats in the project. New conversations start “warm” with context across previous ones.
• Custom GPTs: Best when you want to share your assistant beyond your account (team, company, external). You can keep them private by default, but in that case, a Project may be the better option.

5: Graduate to agentic execution

A step above AI assistants are Agents, and they are becoming more helpful by the day.

Agents can now take goals, plan, browse, click, and deliver outputs, like comparing resumes and building a comparison sheet.

We took a look at ​ChatGPT Agent​ alongside specialist tools like Loveable for ‘vibe coding’ your own websites or apps.

Example: ​Anushka Arellano​ graduated from our ​Certification Program​ last week with a stellar example of taking charge.

An HR consultant in med‑tech, she used Lovable to prototype a lightweight HRIS from a single prompt. As a result, she now has a tailored platform for org chart and other core HR workflows.

The Bottom Line: Becoming a Confident AI Leader

As I said, AI does NOT have to be difficult. Anyone can turn their curiosity into confident AI leadership without any overwhelm.

The only must is diligence and restraint. Not every AI tool needs to be explored, and often learning to prompt your core AI (ChatGPT, Gemini, Copilot) gives you 10x the ROI of adopting yet another platform.

Adopt the colleague mindset, master delegation-quality prompting, layer in ChatGPT‑5 steering, ground adoption in real work, tailor assistants, and then supervise agents.

That’s how leaders move from “experimenting with AI” to running an AI‑powered operating system.

What’s missing? Reply and we’ll discuss.

Until next week,

‍