Leading in the Age of AI with James Killick

Episode #377 Leading in the Age of AI with James Killick

In this special interview podcast, Marty chats with AI strategist James Killick about how leaders should actually use AI in business, why most AI projects fail, what will still be uniquely human, and how to future proof your own career!

Summary of topics covered:

• How reliable AI really is and why hallucinations happen
• LLM search vs Google search and where each wins
• Why 80–90% of corporate AI initiatives fail
• How leaders should think about job losses and “AI replacing people”
• The concept of the “AI orchestrator” for modern leaders
• Human creativity, trust and what will stay uniquely human
• Tone of voice, writing style and why AI content all sounds the same
• Practical tools and use cases leaders can implement now
• Key mindset shifts for thriving in the AI era

1. How reliable is AI right now and will hallucinations go away?

• The main problem is you never know when it will hallucinate.
• You can get hours of great output then one completely false statement.
• Example: Claude did research on Marty, correctly found he dropped out of law, then “decided” James had also dropped out of law. Total fabrication.
• LLMs are trained on massive, conflicting text. When challenged, they often just agree with you, which makes them feel “people pleaser” like.
• You must keep a human in the loop for anything that really matters.

2. How is LLM search different from Google search?

• Marty’s mental model: Google sends you to websites, AI crawls and summarises many sources into an answer.
• James’ view:
  • Think of AI as an overeager intern that goes deep but can be gamed.
  • “Generative search” can be black-hatted. People can game AI rankings.
  • Traditional Google ranking is still more mature.
  • For the average user, Google is still safer.
  • For someone who prompts well, LLMs can give better results because you can specify context, constraints and go deeper.
• For up to date info, he likes Perplexity with strict recency filters.

3. Why are 80–90% of AI initiatives in companies failing?

• Most are solution first, problem second.
• Leaders get excited about tools then ask “where can we shove this in the business.”
• They copy influencers, like cloning their voice for social, with no real purpose.
• AI should either:
  • Amplify what already works.
  • Remove a clear bottleneck.
• Early wins are usually in sales and marketing because they are more templated. Ops and admin are more complex and variable.

4. What should a middle manager do if executives cut staff assuming AI will replace them?

• The mindset should not be “AI replaces humans.”
• The reality is: humans who use AI will replace humans who do not.
• Start by testing one or two small automations that free your best staff to do more of the high value work.
• Use pilots and prototypes, not big bang cuts.
• Let AI handle repetitive grunt work while your people do deeper customer service, problem solving and advisory work.

5. What stays uniquely human as AI and agentic AI get stronger?

• Differentiation and humanity become more important, not less.
• If everyone uses the same models, everyone’s output starts to look the same.
• The edge will be:
  • Personality and real human connection.
  • In person trust, eye contact, handshake.
  • Truly new creative thinking rather than remixing old patterns.
• AI currently recombines the past. It is not consistently creating genuine “new” yet.

6. How bad could this get and are governments doing anything useful?

• James has some fear about where this could go if corporations and governments push too hard for advantage.
• AI is trained on all of human history including war and the worst aspects of us.
• We need to feed it more of “the best of us” but that raises the issue of who decides what “good” is.
• On regulation, James’ honest view is that governments are already late and will be mainly reactive.

7. How should leaders plan their workforce for the next 2–3 years?

• Think in terms of “AI orchestration.”
• Leaders need to orchestrate:
  • The AI tools.
  • The humans who use them.
• Practical pieces:
  • Give everyone basic AI training.
  • Get people involved through challenges or competitions to come up with use cases.
  • Standardise which tools you use so you do not end up with one person on Claude, another on Gemini, someone else dumping sensitive data into free ChatGPT.
  • Keep humans firmly in the loop for security and quality.

8. Is it sensible to stay sceptical about AI answers, even as a good prompter?

• Yes, completely reasonable.
• You can cross check models: give ChatGPT’s answer to Claude and ask it to critique it “brutally.”
• Interesting quirk: negative framing often makes models more careful and accurate than positive incentives.
• For any topic you do not know well, you must assume “you do not know what you do not know.”

9. Should people stay generalists with AI or go niche?

• Domain expertise is the superpower.
• Use AI to amplify what you already know deeply.
• Do not assume you can be your own lawyer, marketer, or consultant just because you have ChatGPT.
• For serious organisational implementations, talk to an AI specialist who has actually done work in your industry or function.

10. Why does AI writing feel bland and samey and can that be fixed?

• If you just ask it to “write an email” it will produce pattern based, middle of the road copy.
• It loves symmetry and patterns. Lists of three, similar headings, similar paragraph shapes.
• You can push past this by:
  • Having AI analyse a large volume of your writing and extract a detailed tone of voice profile.
  • Explicitly setting rules. For example: Australian English, no American spellings, no EM dashes, no Oxford comma, specific joke style.
  • Using “projects” in ChatGPT: each project has system instructions plus a knowledge base of your stories, tone of voice, examples.
• Once that is in place, your day to day prompts can be much shorter and still on brand.

11. What tools and workflows does James recommend?

• Use ChatGPT projects with:
  • Detailed system instructions.
  • Attached knowledge files: tone of voice, your backstory, examples of good work.
• Meta prompting. Get the LLM to write optimised prompts for your role and tasks.
• Use different models for their strengths:
  • Claude for content and long form writing.
  • ChatGPT as generalist and for building GBTs.
  • Gemini for research that leverages Google’s data.
• Tool examples:
  • Notion with the AI agent for querying your own docs, tasks and content.
  • Gamma for building slide decks from prompts, docs or recordings.
  • Built in AI in Google Workspace or Microsoft 365, especially if you are worried about data security.
  • Meeting tools like Fathom or Fireflies for recording and transcribing calls.

12. Rapid fire questions

One thing every leader should automate this year

• Meeting transcripts.
• Then you can search them, summarise themes, and even have AI critique how your one on ones went.

One leadership skill that becomes more valuable because of AI

• Management.
• You are no longer only managing humans, you are also managing AI agents and the system around them.

Biggest myth leaders believe about AI and tech

• “You have to be technical to use AI effectively.”
• James has never written code. His 84 year old father uses AI and is more productive because of it.

Key mindset shift leaders need in the AI era

• Extreme flexibility and openness to change.
• Things are moving fast and that speed will only increase.
• Test, iterate, and adapt rather than waiting for everything to be stable.

Big opportunity right now

• There is a window where expectations have not caught up with what AI can do.
• If a task used to take seven days and you quietly deliver it in two using AI, you stand out.
• Eventually expectations will reset, so the time to exploit that edge is now.