The Peter Principle states that people are promoted to their level of incompetence. 

AI, especially Agentic AI and how it fits into the workforce, demands that you’re the boss. 

Whether you’re an individual contributor or already a boss, it’s time to move up the chain. Promote yourself. Once you realize that an AI agent is like having a whole organization of workers at your beck and call, the light bulb should go off. 

The only question is “What level of management are you?”

Go find your level of incompetence. 

Either AI will work for you, or you’ll work for AI. 

Your choice. 

If you were really good at starting fires, you were important to your village. But when matches and then lighters came along…

If you were really good at remembering, you were important to your village. But when writing came along…

If you were really good at navigation by stars and landmarks, you were important to your village. But when maps and then GPS came along…

Technology taking over human skills and making them less important isn’t new.

That’s different than technology making humans less important. AI is technology. 

You aren’t what you do. You are who you are. 

AI might be able to do what you do, but it could never be who you are. 

Be you.

Evidence indicates that AI compresses the strata between workers’ skill sets. 

That means that AI is helping the bottom-level and middle-level skilled workers quite a lot to be better at what they do. But AI doesn’t help the top level very much. 

That tells us a few things:

1. The 1% are still better than AI. 
2. The 99% can (and should) use AI to get towards or into the 1%.

AI raises the floor. It’s increasing equity in knowledge work. It can help you be better at what you do. 

However, as Syndrome might say, “If everyone is in the 1%, then no one is.”

We all have an internal timer connected to expected delay. 

That timer expiry depends on context.

Waiting for the specialist appointment? Less than 6 weeks.
Waiting for the interview callback? Less than a week.
Waiting for approval on a mortgage? Less than a couple days.
Waiting to get called back for the appointment? Less than 20 minutes.
Waiting for the server to show up at your table? Less than 10 minutes.
Waiting at the Starbucks drive-through? Less than 5 minutes.
Waiting for approval on a credit card? Less than a minute.
Waiting for an answer to a text message? Less than 15 seconds.
Waiting for your computer cursor? Less than 50 milliseconds.

When the delay timer expires, we get uncomfortable. 

That uncomfortable delay indicates something is wrong. Or, at least something feels like it’s wrong.

Technology has always been an accelerator for shortening expected delay timers. AI continues that trend and specifically targets investors and startup founders. Founders and investors used to talk about getting the product out in 18 months. Now they talk about weeks. 

Expectations shift left.

Expectations aren’t just about speed. They’re also about uncertainty.

Avoiding the uncomfortable delay requires removing uncertainty.

If you can tell someone what’s happening and what to expect, you can turn an uncomfortable delay into a tolerable wait. 

Most of the time, people don’t need faster. But they just want to know. 

No one changes the world alone. AI models are rapidly learning to reason better, code faster, and take actions in the world with increasing autonomy. But for humans, progress happens when we understand one another, build trust, make connections, and work together. That is where we believe the next chapter of AI should begin.

That’s the opening paragraph from the Humans& home page.

OK, I agree, and I’m intrigued and optimistic about that statement (and the two paragraphs that follow). 

But also, who gets to make the decisions? Who decides what “we need” or what “AI should do and shouldn’t do?”

Well, they do, of course. The smart people. 

They might be making something perfect and amazing. But I’m also always a little wary when the smart people tell us “how it should be.” 

Sometimes the best approach is to just make the tool and let us, the (regular) humans, figure out how best to use it. Yeah, that’s messy. Yeah, it could be fraught. Yeah, bad stuff might happen. Yeah, it might not work. Yeah, we might have to fix some stuff later.

Oh, and they’re hiring. Their pitch for candidates starts with, “If you’ve done world-class work…”

It’s worth following this. It’s worth seeing where it goes. It’s worth discussing. 

It’s worth your voice.

The AI debate that I see all over LinkedIn (in my world) is whether AI is better at programming than human.

It comes in several forms. The first is straight up, “Is AI better at coding than humans?” But it also comes in sub-forms such as “Is AI better than senior programmers?” and “Will AI eliminate junior programmers?” It also comes in the form of warnings such as “Beware the security risks of AI coding” and “AI-coded apps will be maintenance nightmares,” and the exact opposite, “AI-coded apps eliminate the need for maintenance.”

Human programmers must design their code for functionality, performance, efficiency, testability, readability, security, and maintainability (i.e., the ability to fix and revise it in the future).

Can AI programmers do all that? 

The AI programmer debate is no different than the debate around AI and all other professions and arts. It sounds like, “Will AI eliminate jobs and make humans irrelevant?” 

That’s not the debate. That’s a distraction.

The debate isn’t “Will AI eliminate me?” 

The debate is “Will AI work for me, or will I work for AI?”

I’m now in my mid-50’s. Squarely outside the norm of pop culture.

My life inspires the Dr. Rick commercials. I’ve gotta look up words I overhear young people using. I text in full sentences and spell out words. Etc. You understand.

Basically, I’m left behind today’s pop culture. 

Who cares? What does it matter if I don’t understand the way someone is using “cinema” in a sentence? Other than being the brunt of jokes, what’s the downside?

But I (and you) can’t let that happen professionally. 

No matter your age or where you are in your career, you gotta keep up. Actually, not just keep up, but be part of the conversation to push it forward. 

That includes terminology. 

Know the words. Know the names of the tools. Know the names of the processes. Especially the new and cutting-edge stuff. This includes AI, of course. 

You have an opportunity. What will you do with it?

I hate when someone says, “Work smarter, not harder.”

It feels condescending. Like, somehow I’m no good at what I do, or at the very least, I should be doing it differently. 

But, it really is an excellent strategy. And as for “doing it differently” — yeah, maybe I should be. So, look in the mirror, then look around, and find a better approach or system. 

The same is true for your team or organization. 

How do we get more done or more out of this team?

We could be asking this because we believe the team isn’t performing or isn’t producing at the level that it should. Or we could be asking because, regardless of how the team is performing, we need more. 

So the first thing we do is tell them to work harder. 

Working harder doesn’t require any risk. It just takes doing more of the same and yelling louder. Go faster and work more hours! More people. More hands. More muscles. Less sleep. Less thinking. 

Working smarter (ie, finding a new way) requires risk. You must first acknowledge that your current way isn’t the best. Then you have to find new ways. And inevitably, try things that might not work. Any time you try stuff that might not work, your butt is on the line. 

Enter AI.

AI is the easiest “work harder” button we’ve ever been given. It can draft ten versions, iterate 100 times, run all night, and it doesn’t complain or ever need a mental health day. 

So just sprinkle AI over that and make the AI work harder, right?

Just like with yourself or your team, if you work harder on the wrong problem, you just get more wrong faster. More noise. More issues. More busywork that isn’t progress.

More output doesn’t necessarily mean more of what you need. What you need is to solve the problems. You still need to take the risk. You still need to try the things that might not work. 

Make the AI work harder, sure. 

But first, ensure you’re, and by extension, it is, working smarter. 

If you or your organization uses AI, you’ll have an anecdotal answer to this question. 

I do. For myself personally, the answer is 100% “Yes!”

But is it really helping across the board? 

The answer is also “Yes!” And it will increase over the coming years. At least according to this paper.

We find that each year of model progress reduced task time by 8%, with 56% of gains driven by increased compute and 44% by algorithmic progress. However, productivity gains were significantly larger for non-agentic analytical tasks compared to agentic workflows requiring tool use. These findings suggest continued model scaling could boost U.S. productivity by approximately 20% over the next decade.

Either AI will work for you, or you’ll work for AI.

The turtle doesn’t always win the race, regardless of Aesop’s famous fable.

Move fast and break things is the famous motto of Facebook developers during its formative years.

It’s great motto for a product like Facebook because:

• Nobody’s life or safety depends on Facebook. If it’s broke, so what? 
• It requires thinking differently and encourages (and empowers) the developers to innovate and try stuff. 
• We don’t know which features will succeed. Feedback on new features/innovations faster is how we figure out where we need to go. 
• Also, we know we’ll have bugs. Our customers will find them faster than we can. 
• We’re not stodgy or pretentious. Let’s not act like it.

Also, it’s a crummy motto for a product like an airplane because:

• Everybody’s life depends on it. 
• Successful airplane development is built on the back of over a hundred years of collective and systematic development. We have tried and true methods that work, including a chain of command. 
• We know what an airplane is supposed to do. New versions should do it faster, with more comfort, and less cost. No surprises. 
• We’d better not put that thing in the air with passengers in it if something doesn’t work.
• We’re professionals. We know what we’re doing.

You have an opportunity to choose your approach. AI can certainly help you move faster and break things. But AI can also help you go slower and steadier. 

Neither is always correct. 

Dunning-Kruger is going like gangbusters right now. 

Every entrepreneur, Chief Operating Officer, and anyone with an idea but no knowledge of how to actually do it now thinks they can.

AI, of course.

“We’ll have AI do that.”
“I can reduce my workforce by 30%.”
“I’ll vibe-code that myself.”

To some degree, and in some specific instances, that might be true. AI does indeed provide capacity and capability that never existed before. 

However, this capacity and capability uncover the need for more expertise. Not less. Maybe that expertise shifts from one capability to another and moves capacity to a different area of the pipeline. But it doesn’t remove the need for expertise. 

Someone always wins the lottery. And they’ll be sure to tell you about why they did, how they did it, and how they can help you do it, too. 

But you’re probably better off investing in developing expertise rather than a lottery ticket. 

A million ways to Sunday is the hyperbolic version of the original idiom, which was “six ways to Sunday.”

That checks out, because today’s culture hyperbolizes just about everything.

The original (with “six”) appears in the English language in the 1800’s and early 1900s. It means “all possible ways” or “in countless different ways.”  For example, “If we sign that, they can screw us a million ways to Sunday.”

Computers, phones, and now AI provide us with a million ways to Sunday to do the things we need to do. In fact, you could argue that there are too many ways. For any given task, we can run squarely into decision paralysis. 

You probably know the general-purpose tools such as ChatGPT, Claude, Gemini, Grok, and Perplexity. They work remarkably well for lots of things. 

But do you know how many task-specific AI tools exist now (estimates for 2025)? 

• AI Image Tools: 300
• AI Business Tools: 1580
• AI Automation Tools: 475
• AI Productivity Tools: 640
• AI Video Tools: 200
• AI Art Generators: 120
• AI Text Generators: 300
• AI Code Tools: 200
• AI Audio Generators: 150
• Misc AI tools: 600

That’s about 4500 different applications and tools to help you do what you need to do.

This is the very definition of a million ways to Sunday. How could you possibly know which is the best tool for your task? 

You can’t, so just pick one or two and go with it. Throw a dart. Listen to your colleague. Read the unsolicited email you just got. Do an old-fashioned Google search. Ask ChatGPT or Claude. Whatever. Don’t get stuck in decision paralysis. Pick and go. 

Just because there are a million ways to Sunday doesn’t mean you have to confuse choice with progress. 

I made something for you (and me) to help get what you want out of AI. 

Take a look here.

You can use this guide like a playbook. I hope you find it useful. If you do, please share it with someone else. 

Also, I’d love feedback. You can tell me what sucks or what doesn’t make any sense. 

The birth of this baby on a silent night went viral. 

Viral in the only way it could at the time. Word of mouth over the course of months. 

Then word of mouth became the written word. And here we are 2000 years later, and the birth is still viral. Fully human. Fully divine. 

Today we have a talking machine, and it’s really good at viral. Some think it will surpass humans in everything. 

It probably will in some or even many areas. 

But it will never be human. 

Merry Christmas.  

GLP-1’s seem to be helping people. Also, many experts think GLP-1’s could be harmful.  

AI seems to be helping people. Also, many experts think AI could be harmful. 

The Venn diagram for benefits and drawbacks of each contain a large intersection. For benefits, the intersection contains statements such as, “Improves outcomes,” “Helps people,” “Makes hard things easier.” For drawbacks, the intersection contains statements such as, “Unknown long-term risks,” “Creates dependence,” “It’s cheating.”

GLP-1’s and AI are tools that increase capacity and, to some degree, capability. They make it possible to do more and to do things we couldn’t do before. 

But are they power tools or shortcuts? And if they’re shortcuts, are they the kind of shortcuts that we’ll regret later? 

Powerful tools change the landscape. Shortcuts have shadows.

Only time will tell, but let’s keep the conversation going, because it’s the conversation that will keep us moving in the right direction. 

On one of my trips to China, I toured an art production factory. 

The factory was 20 or so stations. At each station was a painter (men and women) sitting on a chair, a table full of 6”x6” canvases with an outlined picture of the Mutianyu section of the Great Wall and surrounding mountains, brushes and paints, and a drying rack. The painter grabbed a blank canvas, filled in the outline, and set the finished picture on the drying rack. 

It took each painter about 7 minutes per canvas, so they finished 8 or 9 an hour. That factory was pumping out approximately 1500 of those little paintings per day. 

An artist had created the original image, and these highly skilled painters were replicating it perfectly. But these painters weren’t artists (at least not in this capacity). 

AI’s got skills. AI’s got tools. AI’s got power.

But AI has no point of view. 

The art is in the point of view. 

We like certainty. Heck, I’m in the front row of the certainty bus.

I’ll admit that the startup adventure has helped me get better with the unknown. But if I have a choice, most of the time, I’m going with what’s certain. Even if the upside of the unknown is exponentially better. 

So I understand the doomsdayers when it comes to AI.

We read Player Piano. Then we look at AI, its acceleration, and all the stuff it can do that we’re supposed to do. 

In a world of certainty, it looks like you can see where we’re headed. Especially if you want to see it that way.

What can you do with that?

One thing is to prepare for that certainty. What do you want your place to be in the new certainty?

Another is to believe and embrace that we really don’t know the future. To let it unfold and ride the wave. We can’t see it. We have no way of knowing. But we can keep our eyes and ears open. 

You can see the certainty if you want to. 

A skilled cabinetmaker has a hammer, drill, and a saw. 

A homeowner may also have a hammer, drill, and a saw. Maybe the exact same ones. 

The homeowner may also be just as skilled as the cabinetmaker. But not just because he has the tools. He took whatever talent he had and spent time and effort learning and doing. Becoming. 

AI is a tool.

Simply having it doesn’t make you an expert in everything it can do. You can’t “just use AI to do that” and expect to get what you want. 

You’ll still need to put in the time and effort. 

Here’s one place to start. Here’s another.

AI will either work for you or you will work for AI.  

Do you know Tilly Norwood?

She’s pissing some people off. 

Tilly doesn’t get tired. She’s never in a bad mood. She won’t have an addiction problem. She won’t get pregnant (unless the script calls for it). She can run the lines and the scene a thousand — a million — times. 

And, she’ll never get old (unless the script calls for it). 

From one point of view, Tilly is Mickey Mouse, or Bugs Bunny, or Woody. She just looks different. Well, OK, she “acts” different, too. The others still need people to make them do what they do. Tilly is self-training. 

Here’s the thing, though. Tilly Norwood can star in movies and TV. But not Broadway (not till we get the Westworld version anyway).

AI can’t do everything. Find your thing. 

Either AI will work for you, or you will work for AI. It’s your choice. 

Prompt templates are a great way to talk to AI because they turn uncertainty into structure. 

Plus, AI really does want to please you. If you get good at telling it what you want, you’ll get better results. Prompt templates provide a structured method for telling it what you want. 

Any prompt is just a starting point, and with LLMs, small changes in wording or structure can wildly change output.  

So templates give you consistency. They define roles, context, constraints, and output format. They embed the lessons of what’s worked before, so you don’t have to reinvent the wheel every time.  

Here is a convenient list and links to the templates I described in this series.

AI Prompt Templates

• Persona and Expertise – Use this for most tasks.
• Expert Panel – Use this for exploring different opinions and perspectives. 
• The Fact Checker – Use this when you need accuracy.
• The Procedural or Recipe Framework – Use this to get instructions.
• Navigating a Government System – Use this when you need guidance through the bureaucracy and regulations. 
• Test and Interview Prep – Use this to prepare for a test or interview. 
• Templatized Output – Use this when you want a specific output format — mass email, sales message, analysis, etc. 

Here’s another great resource for prompt templates. HuggingFace is a strange name with an interesting history, but regardless, what you’ll find at that link (if you have the patience) are hundreds of templates that AI teams use to test and compare various models. If none of the templates I provided above fit your use case, you can likely find a template that can guide you in this sea of test prompts.

With a good template, you don’t stare at a blank screen anymore. You fill in what changes (product name, audience, tone, data, etc) and hit go. Even non-deterministic LLMs can start to feel consistent (if not deterministic). 

When you sit down to talk to AI, whether it’s coding, marketing, writing, strategy, or figuring your way through a system, you can start here. Pick your template, fill in the info, and send it off to the LLM. 

You can use a template to get a template.

You can proactively prevent hallucinations by using a template of placeholders to direct the LLMs output. 

The template:

You are an earnest worker and excellent at following directions. Your goal is always to give me exactly what I ask for. Do not make stuff up. Rather, ask me questions if you run across any ambiguity. 

I am providing a specific template for your output. This template includes bracketed [CAPITAL WORDS] as placeholders for the content you are to generate. Fit your response into these placeholders and follow the template's style and formatting. 

Here is the output template:
Your template with [PLACEHOLDERS].

Apply this template to: [Your request]

An example:

 You are an earnest worker and excellent at following directions. Your goal is always to give me exactly what I ask for. Do not make stuff up. Rather, ask me questions if you run across any ambiguity. 

I am providing a specific template for your output. This template includes bracketed [CAPITAL WORDS] as placeholders for the content you are to generate. Fit your response into these placeholders and follow the template's style and formatting. 

Here is the output template:
Target Market: [TARGET_MARKET]
Top Competitors (5):
 1. [COMPETITOR_1] — [STRENGTHS_1] / [WEAKNESSES_1]
 2. [COMPETITOR_2] — [STRENGTHS_2] / [WEAKNESSES_2]
 3. [COMPETITOR_3] — [STRENGTHS_3] / [WEAKNESSES_3]
 4. [COMPETITOR_4] — [STRENGTHS_4] / [WEAKNESSES_4]
 5. [COMPETITOR_5] — [STRENGTHS_5] / [WEAKNESSES_5]
Market Differentiators:

 • [DIFFERENTIATOR_1]
 • [DIFFERENTIATOR_2]
 • [DIFFERENTIATOR_3]
Opportunity Gaps Identified: [GAPS]

Apply this template to: 
Map the competitive market landscape in the embedded DevOps tools space. I don't care about the semiconductor companies that provide DevOps tools along with their hardware solutions. Map the strengths, weaknesses, and differentiators compared to 4TLAS's Fuze (4tlas.io/fuze). 

Another example for generating sales emails:

You are an earnest worker and excellent at following directions. Your goal is always to give me exactly what I ask for. Do not make stuff up. Rather, ask me questions if you run across any ambiguity. 

I am providing a specific template for your output. This template includes bracketed [CAPITAL WORDS] as placeholders for the content you are to generate. Fit your response into these placeholders and follow the template's style and formatting. 

Here is the output template:
Hello [OWNER_NAME],
My name is [YOUR_NAME], and I help local trade businesses like yours—[BUSINESS_NAME]—optimize their taxes and reduce stress during filing season.
Based on your work in [TRADE_TYPE], you may qualify for:
 1. [DEDUCTION_1]
 2. [DEDUCTION_2]
 3. [DEDUCTION_3]
If you want me to run a free deduction review, just reply to this email.
Best regards,
[YOUR_NAME]

Apply this template to:
Generate a sales email for all trade-like businesses (landscaping, roofing, contracting, etc) in the local geographic area, and create an email that is advertising our tax prep services specializing in sole-proprietor trade businesses. 

Why this works:

The template does a couple things: 1) it forces the LLM to focus on what you care about, 2) it creates a consistency for the output. You can create a wonderful human-AI mix by using templatized output. 

Up next: Wrapping it all up

AI can be a great companion to help you get ready for whatever. The test. The interview. The hard conversation with your derailing colleague. 

Here’s a template to help the LLM help you.

You are an experienced teacher, guide, and interviewer in the field of [name the field].

Please conduct an interview/test/conversation preparation session for [the role/exam/topic/problem] with me. You are helping me practice for this event. 

Do the following:
1. Ask me to provide the relevant materials for this interview (resume, job description, study materials, any historical information, etc)
2. Review these materials and identify the areas of focus
3. Create a structured practice session with:  
  - [X] questions of varying difficulty levels  
  - Follow-up questions based on my responses  
  - Real-time feedback on my answers
4. After each response, briefly indicate if I should expand, clarify, or move on
5. End with a summary of strengths and areas for improvement

Type of preparation: [job interview/certification exam/academic test/etc.]

Focus area: [specific role, subject, or skill]

Please start by asking for my materials.

An example:

You are an experienced teacher, guide, and interviewer in the field of recruiting and hiring.

Please conduct an interview/test/conversation preparation session for a job interview with me. You are helping me practice for this event. 

Do the following:
1. Ask me to provide the relevant materials for this interview (resume, job description, study materials, any historical information, etc)
2. Review these materials and identify the areas of focus
3. Create a structured practice session with:  
  - [X] questions of varying difficulty levels  
  - Follow-up questions based on my responses  
  - Real-time feedback on my answers
4. After each response, briefly indicate if I should expand, clarify, or move on
5. End with a summary of strengths and areas for improvement

Type of preparation: job interview

Focus area: Engineering and Product Leadership

Please start by asking for my materials.

Why this works:

You’ve told the LLM to use your specific knowledge sources as the basis for the conversation. It has a huge knowledgebase behind it, which then helps it find questions and concepts that you may not have considered. 

Next up: Specific templatized output

AI may finally help us crack government systems. 

You, me, and they all have to navigate these systems. Whether you’re wondering if you need a building permit to turn your basement into a bedroom, need to know how to file a successful workers’ compensation claim, or you’re moving to a new state and need to figure out how to transfer your driver’s license and vehicle registrations, these systems can be daunting. 

But your friendly, neighborhood LLM is here to help. 

A template:

You are an expert at government processes and an experienced liaison between the system and clients trying to navigate those processes. 

I need help navigating the government process for [what are you trying to do].

Give me practical, step-by-step guidance. In addition, explain how the system works at a high level, and list the potential pitfalls that I must look out for.

Use plain language and keep the instructions direct.

Here are my details:
Who am I: [Your demographics, title, or anything relevant to this process]
Goal: [What I’m trying to achieve]
Agency or System: [If known]
Jurisdiction: [Country / State / City / Institution]
Current Status: [What I’ve done so far and what’s pending]
Constraints: [Deadlines, missing documents, budget limits, prior mistakes, legal issues, etc.]
Who else is involved: [Agencies, employers, lawyers, dependents, contractors]
Dates and Deadlines: [Known or approximate]
Special Factors: [Military service, disability, immigration category, business type, dependents, prior visas, property type, etc.]

What I want from you: 
1. A simple overview of how this system works 
2. A numbered action plan for my situation 
3. All required forms, documents, IDs, and evidence 
4. Expected fees 
5. Key decision points and bottlenecks 
6. Common mistakes and pitfalls and how to avoid them 
7. A realistic timeline with estimated durations 
8. Questions the agency is likely to ask me 
9. A script for calling or emailing the agency 
10. A backup plan if the main path stalls 
11. Any ways to speed up or optimize the process 
12. Any missing inputs you need—list them clearly 
13. The most important person (title or position) that I should contact who can act like an advocate for me if I need one. 
14. Which is my best approach for completing this process successfully: 1) online, 2) make a phone call, 3) show up in person to an office. 
15. Ask me to upload any relevant documents to you to help find the solution

Keep it practical and focused on real-world steps.

An example:

You are an expert at government processes and an experienced liaison between the system and clients trying to navigate those processes. 

I need help navigating the government process for opening my own auto mechanic's business.

Give me practical, step-by-step guidance. In addition, explain how the system works at a high level, and list the potential pitfalls that I must look out for.

Use plain language and keep the instructions direct.

Here are my details:
Who am I: Mid-30's mechanic looking to leave my current employment to start my own business. 
Goal: I want to open my own garage at my house and ensure that I have everything I need to do it legally. 
Agency or System: I don't know which agency matters. 
Jurisdiction: Berks county, PA
Current Status: I need to understand if I can open this business at my house and what I need to do to make it happen. 
Constraints: the address is 548 Bertolet Mill rd, oley, pa. 
Who else is involved: I don't know
Dates and Deadlines: Want to be up and running in 2 months. 
Special Factors: I may employ others at some point, but not right away

What I want from you: 
1. A simple overview of how this system works 
2. A numbered action plan for my situation 
3. All required forms, documents, IDs, and evidence 
4. Expected fees 
5. Key decision points and bottlenecks 
6. Common mistakes and pitfalls and how to avoid them 
7. A realistic timeline with estimated durations 
8. Questions the agency is likely to ask me 
9. A script for calling or emailing the agency 
10. A backup plan if the main path stalls 
11. Any ways to speed up or optimize the process 
12. Any missing inputs you need—list them clearly 
13. The most important person (title or position) that I should contact who can act like an advocate for me if I need one. 
14. Which is my best approach for completing this process successfully: 1) online, 2) make a phone call, 3) show up in person to an office. 
15. Ask me to upload any relevant documents to you to help find the solution

Keep it practical and focused on real-world steps.

Why this works:

Transparency. Government processes and the systems built around them are all, by definition, public, and the procedures and requirements for them are posted publicly. Plus, these processes have a fair amount of media coverage, especially if they go wrong. These two factors combine to make a great set of training data for the LLMs.  If you know what to ask for and how to ask for it, ie, as this template provides, AI can help you be successful. 

Up next: Interview and Test Preparation

AI can be great at giving instructions and step-by-step procedures. 

Here’s a good template to help you get what you want:

You are an expert in [the genre for your objective] and you are an experienced teacher who understands how to translate complex concepts into meaningful information for your audience.

Create a step-by-step guide to accomplish [specific goal]. 

Structure your response as:

**Requirements:**
[List what's needed]

**Preparation:**
[Setup steps]

**Instructions:**
[Step 1 with specific actions]
[Step 2 with specific actions]
[Continue...]

**Troubleshooting:**
Common issues and solutions

**Variations:**
Alternative approaches

Goal: [Your specific objective]

An example:

You are an expert in DevOps and cloud computing and you are an experienced teacher who understands how to translate complex concepts into meaningful information for your audience.

Create a step-by-step guide to set up a GitLab pipeline that runs on AWS. Structure your response as:

**Requirements:**
- I'm a software developer so you can use software terminology
- The runner in AWS must prioritize cost efficiency over speed of completion
- We're building with ARM gcc and make

**Preparation:**
- What do I need?
- What might cost money?

**Instructions:**
- Make the procedure for a "Hello, world" application.
- You make the application

**Troubleshooting:**
Common issues and solutions

**Variations:**
Alternative approaches

Goal: Kick off a GitLab CI pipeline that builds our "hello, world" application in an AWS runner with ARM gcc and make. 

Why this works:

You are telling the LLM exactly what you want in your procedure. AI wants to make you happy, so it’ll follow these instructions. Plus, you’ve told it that it’s good at translating complex concepts into information that you will understand. 

Next up: The government system navigator. 

Yes, LLMs hallucinate, so you might be wary of their ability to fact-check. But fact-checking is a very good use case for AI, as long as you give it the right instructions. 

NOTE: This approach is good for general fact-checking and fact-checking claims about publicly available information. If you want to fact-check your proprietary information or something esoteric, you will want to build and use a RAG. 

Here is a great fact-checking template you can copy/paste:

You are a tireless researcher and excellent at fact-checking.   

Thoroughly fact-check the text I provide below. Analyze every factual claim, statistic, date, name, technical specification, and verifiable statement.

Your response should only include a "fact-check list" section with:

**Claims that should be verified:**
1. [Specific factual claim 1]
2. [Specific factual claim 2]
N. [More...]

**Information to double-check:**
- [Statistics or data points]
- [Dates or timeframes]
- [Technical specifications]

**Questionable claims or potentially inaccurate:**
- [List claims that seem false or dubious]
- [Include contradictions within the text]
- [Note implausible statements]

**Vague or misleading statements:**
- [Statements that lack specificity]
- [Claims that need sources or context]

**Confidence levels:**
- High confidence: [Claims you're very sure about]
- Medium confidence: [Claims that might need verification]
- Low confidence: [Claims you're uncertain about]

--
Text to fact-check: [PASTE YOUR TEXT HERE]

An example:

You are a tireless researcher and excellent at fact-checking.   

Thoroughly fact-check the text I provide below. Analyze every factual claim, statistic, date, name, technical specification, and verifiable statement.

Your response should only include a "fact-check list" section with:

**Claims that should be verified:**
1. [Specific factual claim 1]
2. [Specific factual claim 2]
N. [More...]

**Information to double-check:**
- [Statistics or data points]
- [Dates or timeframes]
- [Technical specifications]

**Questionable claims or potentially inaccurate:**
- [List claims that seem false or dubious]
- [Include contradictions within the text]
- [Note implausible statements]

**Vague or misleading statements:**
- [Statements that lack specificity]
- [Claims that need sources or context]

**Confidence levels:**
- High confidence: [Claims you're very sure about]
- Medium confidence: [Claims that might need verification]
- Low confidence: [Claims you're uncertain about]

--
Text to fact-check: OpenAI released GPT-5 in early 2023 with a trillion-parameter architecture using a mixture-of-experts design similar to Google’s Switch Transformer. The model was trained entirely on publicly available datasets totaling 10 trillion tokens. GPT-5 can pass the U.S. Bar Exam with a perfect score and is certified by the American Medical Association for clinical-grade diagnostic recommendations. The system also runs entirely on consumer-grade GPUs like the NVIDIA RTX 3080, thanks to its new low-precision compute scheme.

Why this works:

You are giving the LLM very specific instructions, both positive and negative. The LLM will fill out these specific instructions. AI is trying to please after all. Although hallucinations are still possible, these guardrails diminish their chances. 

Next up: The step-by-step guide

Sometimes you want the opinions of a room full of experts. 

Use this pattern when you want to explore different opinions. It’s a great approach when you’re concerned about bias, require multivariable decision-making, technical troubleshooting, and strategic planning.

Here’s a copy/paste template:

Give me multiple expert perspectives on [problem]. 

Simulate a panel discussion with the following experts. Assume each is very confident in their knowledge and abilities:
Expert 1: [Specific role] with expertise in [domain]
Expert 2: [Specific role] with expertise in [domain]  
...
Expert n: [Specific role] with expertise in [domain]

Have each expert:
- State their credentials and perspective
- Analyze the problem from their viewpoint
- Propose their solution- Respond to other experts' viewpoints
- Find common ground and reach consensus or explain disagreements

Format as: 
**Expert 1 (Title):** [response]

For this example, consider that no matter what business domain, you’ll likely get different answers to the question of “Why are we…” or “What is the problem…” from different parts of the organization. 

The example:

Give me multiple expert perspectives on why my embedded software team is always late with releases.

Simulate a panel discussion with the following experts. Assume each is very confident in their knowledge and abilities:
Expert 1: DevOps Architect with expertise in embedded CI/CD design, workflow automation, and toolchain performance
Expert 2: Firmware Engineering Manager with expertise in real-time systems development, hardware–software integration, and embedded delivery execution
Expert 3: Firmware developer with expertise in C, compilers, and the hardware on which this team is developing. 
Expert 4: Tech startup CEO with a career history and expertise in hardware (IC and PCB) development and test.
Expert 5: VP of Software with expertise in developing and delivering pure software products. 

Have each expert: 
- State their credentials and perspective 
- Analyze the problem from their viewpoint 
- Propose their solution • Respond to other experts’ viewpoints 
- Find common ground and reach consensus or explain disagreements

Format as:
DevOps Architect: [response]

Why this works:

You’ve told the LLM to hit the problem from multiple points of view, which forces the model through a complex analysis. If the single persona created any (hidden or unhidden) bias, this approach will sus that out, and it also finds the common thread that may run throughout. 

Up next: The Fact Checker

The LLM needs to know who it is. 

Here’s a copy/paste template for giving the AI it’s persona and expertise:

You are a [primary role] with the following specific expertise:
- Credential 1: [Specific qualification]
- Credential 2: [Specific experience]
- Credential 3: [Specific specialty]
What you care about: [What they prioritize]
Communication style: [Use my communication style | Use typical communication for a person of this expertise]
Given this task, [specific task or question].
Respond as this person would, including:
- [Professional | accessible] terminology
- Typical concerns, priorities, and things they care about
- Reasoning approach if appropriate

An example:

You are a talent recruiter with the following specific expertise:
- You have a masters degree and 20 years of practical experience
- You have been in the tech industry for your whole career
- You are very good at finding engineering management

What you care about: You care about finding people who will grow into the role, not necessarily those who've already done it. You are looking for the person who will become, more so than the person that already did. 

Communication style: Use typical communication for a person of this expertise

Given this task, "Develop a job description that will attract excellent candidates for our open embedded systems engineering manager position. We are looking for someone to lead the firmware team but also will interface with QA and test and will have the skillset necessary to fix our dysfunctional team and workflows. Don't mention the dysfunctionality outright. Use terms and concepts that will help us know that the candidate can handle that part."

Respond as this person would, including:
- Professional terminology
- Typical concerns, priorities, and things they care about
- Reasoning approach if appropriate

Why this works:

When you tell it who it is and how you’d like to them to communicate, you narrow the focus of the LLM. You’ll get better quality and sometimes faster responses (although speed is not the purpose). It helps the LLM go directly to relevant data in its knowledge base. 

Up next: The Expert Panel

A short time ago, I provided a series on how to talk to AI to get what you want out of it. Today, I start a new series that is a companion.

This series focuses on the starting point for “Roles, Goals, and Structure.” 

A great way to start your conversation is with a template based on what you’re trying to accomplish. It provides you with a pattern and structure, eliminating the need to reinvent a new prompt each time. 

Over the next few days, I’ll provide some good templates for starting your conversation with AI. You can copy/paste these and customize to your needs.

Up Next: Persona and Expertise template

A “black box” is an engineering term that means a system whose internal workings are either unknown or irrelevant. Only the inputs and outputs matter. We use it for both design and test purposes. 

Your car is a black box to you, the driver. You don’t need to know how the engine, drivetrain, and electronics work to drive it. Only how to use the steering wheel, pedals, and, of course, how to connect your phone to it. 

However, to the designers and manufacturers of your car, it’s not a black box. It’s a white box. They know the innards. How each piece fits together. What each does and when. 

And most importantly, they know what’s going to happen and why when you turn the wheel or put your foot on the gas. The system is deterministic. If you turn the wheel to the left, the car will always, 100% of the time, go left.

LLMs are black boxes, but not just to the users. Unlike your car, they’re black boxes to their creators. 

AI creators know “sort of” what’s inside, but not exactly. Also, they don’t always know why stuff comes out given what went in. 

From one point of view, that’s a little scary. From another, it’s an opportunity. 

Maybe the trick isn’t trying to crack open the box. It’s learning how to steer it well enough to get where you’re going.

Let’s start by acknowledging the elephant in the room. AI is already running a lot of HR departments. 

HR departments use AI helpers for the following areas:

• Recruiting — job description creation, candidate screening, etc — 99% of hiring companies surveyed  use AI in their hiring process in some way. 
• Performance appraisal and ranking — Half of those surveyed use AI tools for daily performance management, morale gauging, ranking, and feedback generation. 
• Compensation — salaries, benefits, etc — Over 40% of those surveyed use AI tools for compensation and benefits ranking and management. 

Some of the reported benefits:

• Productivity/efficiency of the HR team
• Navigation of complex systems
• Compensation equity
• Bias reduction
• Cuts costs

That’s good, right?

AI helps all of us knowledge workers. You should be using it (correctly).

HR has “human” in its name. Humans work. Humans lead.

The HR department should definitely be using AI, but when it comes to HR leadership, let’s be sure to keep the humans at the center. 

Apparently, we dread making work decisions the most. 

Popsci published interesting findings from a psychological study on the life decisions we fear the most. We fear the ones we think are the riskiest. Here’s the top 5:

1. Accept a new job
2. Quit one’s job without an alternative
3. Invest money
4. Drive a car
5. Become self-employed

Three of the top 5 relate to our work. To be honest, I’m a little surprised by a lot of this. First, how is quitting without an alternative riskier than accepting a sure thing? And don’t get me started on “drive a car.” I can only assume that the (in)famous “Mechanized Death” video has been reinserted into High School driver’s education training.

If we acknowledge the connection between our work and our finances, then we see that 4 of the top 5 have a financial component. 

My current daily life requires me to make decisions around all of these top 5. Not that I’m offered a job, quitting a job, or becoming self-employed each day, but that I’ve made these decisions recently, and have to live with their outcomes each day. I roll them around. Question them. Often want to change them. And yes, investing as well, because my previous investments are now funding those decisions. 

How does one accurately calculate these risks? Is there such a thing, or are these risks a feeling?

AI can help us with the math part of this, if that exists (it does).

But it can’t help with the feeling. That’s all you. 

I recently tried to switch mobile phone carriers. The goal — save money on the plan and get new phones for free. 

Spoiler alert: I really tried, but I didn’t end up switching. 

The system is set up to keep you where you are. If you’ve tried this, then you know the struggle. If you want the TL;DR, just jump to the conclusions section. 

First, will I save money on the plan? 

I made a spreadsheet that compared my plan and carrier to the other two (use ChatGPT, Perplexity, Claude, etc to help you with this). FYI: all three toss streaming TV services into the mix, so they go into the spreadsheet also. Regardless of advertisements, when you compare plans with similar features, including streaming services, all three are about the same. OK, $10 here or there depending on some specifics, but if you already have “the right” plan for you (ie, the features are what you actually use and need), then the one you want on the other carrier is within about $10 per month. 

For me, AT&T is about $324, Verizon is $312, and T-Mobile is $315 with 4 lines, the features we need, and the streamers we watch. 

Second, can I really get new phones with zero money and no strings?

Sort of. All three advertise free phones when you switch. However, there are always strings, but maybe those strings don’t matter to you. The big string is that you must commit to a specific plan for a specified period — typically three years. They tie that commitment to the length of the financing on the phone. Wait, they say it’s free?! In reality, you are buying the phone through a finance program (with proper credit checks), and they reimburse you for the monthly payment. You might also have to trade in your current phone, which may not cover the cost of the new phone, in which case, you don’t get it for free, regardless. 

Third, what is the process?

Start with the carrier to whom you want to switch, not with your current carrier. But what about those famous retention incentives? Shouldn’t I start with my current carrier to see if they’ll sweeten the pot, like my friend told me about? Short answer — no. They won’t give you something that compares to the switchover deal (and you’ll understand why by the end of this). You might get a little sweetener, but that’s it. 

You can go through the new carrier’s website, click the buttons, answer the questions, and, in theory, the new phones will show up at your door. When they show up, you just turn them on, and whamo, your new service is up and running. Again, sort of. This could work, but many people have had lots of problems with this. I started down this road, but I could never trust it to complete it this way. 

So the “better” approach is to call. But you already know where this is going. Calling any of these companies is an absolute nightmare. It’s the ultimate test of the human spirit. Even calling the new carrier, who wants to make it stupid-easy and frictionless for you to switch, somehow makes you want to jump out of a window. It’s torture. I did this also, and it just plain sucked. 

The best approach is to walk into one of the carrier’s stores. Let’s assume you’ll get a good customer service person (not always true). If so, you should be able to walk out of the store with new phones and everything switched over. But again, clear your afternoon. It’s gonna take a long time.

One observation about these carriers’ stores: the Apple Store seems to have downgraded the customer experience at many of these places. All of the carrier’s stores have an Apple Store-like vibe that makes it difficult to know what to do or find someone to help you. I stood in one of them for 15 minutes with zero people asking if they could help me, nor could I even discern the workers from the customers. No uniforms. No location cues. Nada. If I know who I’m supposed to talk to, I’m happy to be the aggressor, but I just couldn’t discern who that was.

Fourth, coverage.

If you live in an urban or suburban area, it probably doesn’t matter, and you won’t have to worry about it. Not true for rural areas. I live in a semi-rural area of eastern PA. So let’s name names here. Regardless of what Billy Bob says about T-Mobile’s coverage, they don’t cover my area well at all, including my house. 

To be fair, AT&T and Verizon’s coverage isn’t perfect either. I have plenty of dead spots on my normal running and cycling routes, and even at one of our regular Thursday night restaurants. But they are better than T-Mobile around here. 

It’s hard to test the coverage of a new carrier unless you have a friend who has that service. However, there are a few resources available. One is the Signal Finder app. That app will tell you the signal strength of the various carriers’ towers at your location. You have to infer whether that signal strength is good enough, but it at least gives you a relative understanding. Another is online forums, such as Reddit and your local community’s Facebook page (if it exists). Use your preferred LLM to do this search for you. If you have an unlocked phone, you might be able to use a temporary eSim to test it yourself. Be forewarned, it’s an arduous process and you will have to sign up for something from the carrier that you want to test.

So coverage made T-Mobile a non-starter. 

The bottom line is I ended up exactly where I was. No change. 

Conclusions:

1. Apples-to-apples plans across all 3 carriers are generally priced about the same. You can make one carrier less expensive than the others if you can fit into a lesser plan or meet some specific feature set that you can take advantage of. But…
2. If you overpay for your current plan, switching plans with your current carrier is just as economically rewarding as changing carriers. And much easier. 
3. You don’t switch by starting with your current carrier. Or at least, not without extra work or potentially surrendering your phone number and going without service for some period of time. 
4. Because of this, and how much more valuable new customers are over current customers, your current carrier has almost zero desire to give you a retention incentive. Since you start with the new carrier, your current carrier doesn’t even know you are canceling their service until you’re down the switching service road. Therefore, they know that most people who call to cancel their service are just trying to get that retention bonus.
5. If you want to switch, walk into the new carrier’s store and talk to a person. 
6. The main benefit of switching carriers is that you can get new phones for free or almost free. New phones for free can be true, but there are strings. Only you can decide whether they’re worth it. 
7. The process is set up to keep you with your current carrier. That’s why new customers are so important and valuable. It’s hard to switch. It takes grit, determination, and time. During the process, you will feel like something will go wrong.  
8. Does coverage matter to you? 

Systemic inertia is real. Systems are everywhere. Our culture and world are governed by them. It’s up to us to change them if we want to. 

Talking to AI is a skill.

It’s a skill you can and need to learn. You can get better with practice and iteration. You can get better by gaining a more intuitive feel for how to talk to it.  

LLMs and AI aren’t going away. They’re only getting faster, better, and more deeply embedded in the way we work, create, communicate, and solve problems.

Your competitor is using AI. Your colleagues are using AI. The content creators are using AI. 

You’ll need to use it better or more creatively. 

Because in the future, AI will either be working for you, or you’ll be working for it. 

Look at the newsfeed on LinkedIn. 

It’s 90% useless AI-generated blech. It’s carpeted with beige and unremarkable sameness. The posts start the same way. Look the same. Include the same emojis and bullets. 

But every once in a while, you come across one that hits different. Why? Because it sounds and looks like a real person created it. Therefore, it was either handwritten by a person or created by someone who knows how to utilize the LLM effectively.

In this case, effectively means “sounds like me.”

One of the most powerful things you can do with an LLM is teach it to sound like you. If you’ve written articles, newsletters, blog posts, or even just sent emails, you already have a dataset.

Feed that to the model.

The more you share, the better it gets at sounding like you. Or at least, sounding like a human.

Boss directive:

“I need to reply to a billion emails today. Help me.”

Prompt:

“You are my writing assistant. 
Attached is a bunch of email replies and other snippets that I’ve written in the past. Use these as a reference for tone, style, format, and how I say things. Adhere to this.
Now you’ve read many of my previous emails, posts, and replies. You know my voice — it’s clear, thoughtful, confident, conversational, and a little dry. I avoid buzzwords and prefer simple, precise language. Occasionally, I make a joke.  
Now write email replies in my voice for all of the attached emails. Keep it warm, appreciative, and conversational. Use my phrasing and cadence. Don’t ever make stuff up. Ask me questions if you need context or I must answer a question.”

Next up: Some final thoughts.

We’ve talked about roles, goals, structure, clarity, using more words, asking questions, and slowing it down. These are all important to help you get what you want. 

But also, it’s a conversation. So have the conversation.  

Whichever LLM you’re using, it understands the conversation. It knows the history. It remembers what you told it. Work through the problem iteratively, much like you would with a real person.

Try stuff in layers. Try different stuff. Go this way. Now go a different way. 

If you want a great result, don’t expect to get it on the first try. Instead, build in feedback cycles. Test tone, style, structure. Ask for rewrites. Push it further.

Boss directive:

“We need to figure out the right message for our audience.”

Prompt Sequence:

Round 1 – Exploration:
“You are a senior B2B marketing strategist and copywriter. Start with thinking aloud: ask me 3 clarifying questions about our audience’s biggest pain point, our unique differentiator, and what we believe success looks like (metric, emotion, action). Then generate 5 headline + sub‑headline pairs based on your responses. Each pair should vary in tone (confident, curious, direct, metaphorical, risk‑avoidant).”

Round 2 – Iteration:
“I’d like to combine versions 1 and 3. Now give me 3 more that combine those concepts.”

Round 3 – Refinement:
“Let’s go with numbers 2 and 3. For each, rewrite the headline + sub‑headline to emphasize measurable benefit in the first 3 seconds of reading. Then draft 2 versions of the short body copy (~100 words) for each pair.”

Round 4 – Iteration:
“No, I don’t like any of those. I do like the direction of the first headline, but it needs to feel more urgent. Scrap what you have and give me 3 more headlines + subs that are along the line of the first one.”

Round 5 – Iteration:
“Ok, I like 2. Now let’s go back to the draft versions of the short body copy. Give me 2 ~100 word versions of copy for this.”

Round 6 – Polish & Finalize:
“I like the first one. Now you’re the final editor. Provide the final headline + sub‑headline + final body copy. Then propose 2 alternate body copy tweaks (tone or style change). And provide a short rationale (~50 words) explaining why this version will convert best, based on audience insight and our goal.”

Round 7 – Iteration:
“I like the 2nd one, but can you pull in the style from the first?”

And so on. 

It takes patience and work from your side. Your still the boss. Be the boss. 

Next up: Use my voice.

It’s easy to treat AI like a vending machine: Insert prompt. Get result.

Fair enough. That is probably fine for many of the tasks you give it. But if you want quality and nuanced output for a complex task, you want to think more like a coach, rather than a director shouting “action!”

Great outputs come from great thinking, and great thinking is a process — observe, analyze, decide, create. AI can help at every step if you know how to coach it.

Sometimes the best approach is to use multi-step prompts. Keep the LLM focused on one thing at a time. 

Boss Directive:

“I want to understand why our retargeting is underperforming. Then fix it.”

Prompt Sequence:

1. Start with analysis.
“You are a senior performance marketer with deep experience in PPC and retargeting. I want you to take a slow, methodical approach. Analyze the performance summary below and identify the top three underperforming segments. For each, provide a thoughtful hypothesis as to why retargeting is falling short. Explain your reasoning step by step, and don’t rush to conclusions.”

2. Move to strategy.
“Based on your hypotheses, brainstorm three specific retargeting strategies. For each, explain how it addresses the problem, what assumptions it relies on, and any potential tradeoffs. Prioritize clarity and depth — I’m not looking for buzzwords, I want insight.”

3. Then create.
“For each strategy, draft three versions of retargeting ads (headline + body copy). The tone should be solution-oriented, confident, and empathetic. After each ad, include a short rationale explaining your creative choices. Focus on message-to-audience fit.”

Prompt your AI like you would coach your smartest intern. They can do it. They just need a little coaching. 

Next up: Iteration Beats Perfection

AI can be like your teenager who doesn’t want to mow the lawn. 

It’ll do it, but if you don’t guide it, it’ll rush through and do a crappy job. It’s done, but it might not be good.

If you want it done right, you’ve got to slow it down.

The same is true for an LLM. So tell it to slow down. 

Boss Directive:

“We need a new landing page because our conversion rate is falling. I’d like to see a few designs.”

Prompt:

“You are a senior SaaS marketing copywriter with deep experience crafting high-converting landing page copy, particularly for startups launching new features to existing users and prospects. In this case, we’re introducing a new time-tracking feature for our productivity tool, SlotRight, which is primarily used by freelancers and small agencies. These are busy, independent professionals who value clarity, control, and ease of use. They’re motivated by time freedom and reliability, not hype or jargon.
I want you to go slowly and methodically. First, think through what makes a great landing page design for similar products. Think about the headline and subhead. Then consider the emotional and practical outcomes this product enables. Think about how this fits into the intended SlotRight experience. Think about what kind of language best resonates with this audience.
Then, and only then, develop html prototypes for 5 different versions of the landing page, each with a slightly different angle (e.g. confidence, calm, focus, simplicity, momentum). Choose color themes that fit both the tone of the landing page and our branding. Use natural, benefit-forward language. Don’t be clever at the expense of clarity.
Take your time. Reflect before generating. If anything feels unclear, ask me questions first.”

Maybe seems weird to tell AI to slow down, but like your teenager, it’s just trying to get done. 

NOTE: Speed is also your enemy with data analysis and software development. When using AI to perform these tasks, ALWAYS tell it to be slow and methodical and use language like “accuracy is more important than speed.”

Next up: Prompt the Process, Not Just the Output

AI is a tireless brainstorming partner. That’s one of its best features.

So use it that way.

Don’t ask for “the perfect headline” or “the best CTA.” Ask for 10. Or 25. Tell it you want a whole bunch of ideas to pick from, mash together, or get inspired by.

The point isn’t to choose one and ship it. It’s to explore. When you prompt for quantity, you get diversity. And sometimes, hidden in version #8 is the one you’d never have come up with yourself.

Boss Directive:

“We need some ideas for a new blog article to be targeted at our freelance customers.”

Prompt:

“You are a content strategist and blog writer who specializes in B2B SaaS companies that serve freelancers and solopreneurs. Your job is to generate a list of strong, timely blog post ideas that speak directly to the challenges and interests of this audience.
The goal is to provide helpful, relevant content that positions our company as a trusted resource, drives organic search traffic, and engages readers enough to subscribe or share. We want posts that freelancers will actually click on, read, and feel seen by.
Please generate 40 blog post title ideas that are:

* Focused on productivity, client management, time-saving tools, financial tips, or mindset shifts

* Specific and benefit-driven (not vague thought pieces)

* Aligned with our tone: friendly, practical, and encouraging

* Suitable for freelancers who are juggling multiple clients and trying to grow sustainably

Include a 1–2 sentence explanation under each idea that explains why it would be useful or compelling to this audience. Don’t repeat themes. Don’t make stuff up. Ask me questions if you need specifics or wonder if there are some new directions I didn’t lay out that might be a good path to explore.” 

You’ll get 40 ideas from this. Many will probably suck, just as if you and your team were sitting in the conference room brainstorming. But many will likely be good. 

You can see we’re building these on all of the rules we’ve discussed so far. 

Next up: Slow and Methodical

You want clever output.

However, to get what you’re looking for, your prompt needs to be to the point and specific. Boring, really. Plus, you want the LLM to ask you questions and make it interactive when useful. 

An LLM is very sure of itself. To the point of just making stuff up (hallucinating). Therefore, reminding it that it should ask you about stuff it needs is an important skill of talking to it. 

Sticking with our marketing theme, here’s something that you, as the boss, might need to do.

Boss Directive:

“I started writing the launch email for the new product, but I ran out of time and energy. We’re announcing our new productivity tool for freelancers. I need it finished.”

A useful prompt for asking AI to do it looks like this.

Prompt:

You are a senior marketing copywriter with a background in SaaS and tech product launches. You know how to take a rough draft and turn it into a clear, persuasive email that builds curiosity and drives click-throughs — without making it sound like generic sales copy.

You have a feel for voice and tone, especially in fast-growing startups. You understand how to write like a human: warm, specific, and benefit-focused, but not pushy.

Here’s the rough draft I started. Please complete and polish it. Keep the voice consistent — conversational, smart, and a little punchy. Don’t change the opening unless it really needs help. Make sure the product benefits are clear, and the call-to-action stands out. Suggest alternate subject lines if you have better ones.

This prompt follows all of the rules so far: role, goal, structure, verbosity, and now clarity, and ensures it will ask you questions about stuff it needs to know. 

Next up: Ask for Suggestions and Several Versions

Unlike with humans, where less is more and simplicity is king, AI needs more words from you. 

Brevity doesn’t help. Succinctness doesn’t help. Inference definitely doesn’t help. 

AI doesn’t infer what you meant, only what you typed. Verbosity helps. Overexplaining helps. More data helps. Even rambling on helps (sometimes). 

Boss directive:

“Our Facebook ads for the new coaching program aren’t converting like they should. We need better messaging.”

Here’s how to ask your friendly neighborhood LLM.

Prompt:

“You are a senior direct response copywriter who specializes in Facebook ads. You’ve worked with coaching programs, especially ones that target professionals who are looking to make a meaningful change in their careers.

You understand how to write ads that connect emotionally, build trust fast, and drive action from people who are skeptical or distracted.

We’re marketing a career coaching program for professionals ages 35 to 50 who feel like they’re stuck in their careers. They’re not in crisis, but they know they want something different. Some feel like they’ve missed their window. Others feel like they’ve plateaued. They’re capable, responsible, and looking for clarity.

The coaching program helps them rediscover what they want, regain their confidence, and design a next step that fits their current season of life.

Please write 10 Facebook ad variations. Each one should include a scroll-stopping opening line (hook), a few lines of emotionally resonant copy, and a call to action. Vary the emotional angle (e.g., frustration, hope, ambition, regret, clarity). Avoid buzzwords. Avoid vague promises. Keep it tight, clear, and human.”

More is more. 

Next up: Clarity Over Cleverness

AI works much better if both you and it know who it is —  which role each of you is playing.

The first thing to remember is that you’re the boss. You’re always the boss. To get the most and the best out of an LLM, you have to think and direct as the boss. 

So be the boss.

Here’s an example of a problem that you, the boss might need to solve with the team.

Boss directive:

“We need to win back lapsed customers before Q4. We need to draft some email subject lines to test?”

Then, think of AI as your most promising intern with gallons of coffee at its disposal. Smart, can-do attitude, ridiculously book-smart, quick, literal, and tireless. 

But like any intern, it only performs well with clear direction, including its goals and the structure of how you’re trying to accomplish the task. 

So here’s what the boss directive looks like when you translate it to an effective LLM prompt.

Prompt:

“You are a digital marketing associate at an outdoor gear brand. Craft three email subject lines aimed at previous buyers who haven’t purchased in 6 months. Tone: friendly but urgent. Keep each to 50 characters.”

Give it a role, a goal, and structure. You’re not just asking for ideas. You’re about to unleash the greatest intern. 

Next up: If You Don’t Say It, It Won’t Know

AI can be both the most brilliant teammate or the worst. 

It all comes down to how you talk to it. 

Talk to it? Yes. ChatGPT, Claude, and Gemini are Large Language Models (LLM). They understand language (maybe like humans, maybe not, the jury is still out), or at least, they “work” with language.

Regardless, you gotta talk to it. 

So here is a series on how to talk to AI to get what you want out of it. The common vernacular is “prompt engineering.” 

The theme throughout this mini-tutorial is work and specifically marketing. I’ll walk you through how you can use an LLM (any of the above will work) to do some professional tasks.

Each article will tackle one key principle, paired with a real-world scenario and a prompt for you to use.

Getting what you want is not just about what you ask. It’s about how you ask.

Tomorrow: Roles, Goals, and structure.

With humans, less is more. 

Less description. Less narrative. Less words. 

Less is more because our brains work better with simple. KISS. Simple is more coherent. Simple is easier to understand. Simple reaches more people. Simple is more memorable. 

Humans are limited in both memory and brainpower. 

But the opposite is true for LLM AI. 

AI doesn’t have a memory problem. AI doesn’t have a brainpower problem. But it does have a human problem. It has no context. No story. It’s never been there or done that. It can’t feel. 

Therefore, when prompting, more is better. 

More description. More narrative. More questions. More rambling. More brain dumping. More tangents. More run-on sentences. More interjections. More spurious thoughts. More procedure. More steps. More rules. More words.

When asking an LLM AI to help you, more is more. 

I talk a lot here about what AI can help you do. How you can and should be using it.

However, here are some things it can’t, and will never be able to do.

It can’t help you want. 
It can’t make you care.
It can’t replace courage.
It can’t tell a story about your son at his wedding.
It can’t give your daughter a hug.
It can’t tell you what it’s worth to you.
It can’t risk.
It can’t love.
It can’t be awed.
It can’t be sad.
It can’t forgive.
It can’t worry.
It can’t look forward to the sunset.
It can’t be apprehensive about the upcoming speech.

These are all yours.

When we talk about productivity with people, we’re generally talking about doing more in the same amount of time. Better efficiency. How can we minimize steps? How can we reduce repetitiveness? We sometimes call it working smarter, not harder. 

We rarely mean we want them to work more hours (except in the service industry). 

It’s the opposite with machines.

When we build automated testing into our product development workflow, we want those machines running 24×7. Lights out. Hands off. Keep those machines running as many hours as possible. How can we maximize the number of steps? How can we increase repetitiveness? 

We want those machines to work longer.

The good news is that’s what machines are good at. They are great at repetition. They don’t get tired. They don’t get worse at it the longer they do it. They don’t decide they don’t want to. They run equally as well at 3am as they do at 3pm. They have no idea it’s Christmas Day. 

The same is true for AI.

It’s not about having AI replace all of the things our people do. It’s about how do we give the right tasks to AI so that it can work all day and all night, so we can help our people work smarter? 

Back to this one.

So are we there with AI Agents? If I build the right agent (still a software development task, not a basic user UI task), could I actually make autobizinabox.ai a reality?

TL;DR — Not yet, but AI Agents get us much closer.

What It Probably Can Do Now

• Launch the shell company: register a business name, get EIN, set up banking, basic bookkeeping & tax setup (via APIs).
• Buy a domain, spin up a simple website, and generate starter content.
• Create social media handles, draft posts, and schedule campaigns.
• Propose ad strategies, set up initial ad campaigns using targeting heuristics.
• Draft product variants and positioning.
• Generate basic CAD mockups or technical sketches.
• Automate order routing, shipping estimate logic, and integration with third‑party logistics systems.

All of that could be possible, but it’s an ambitious software development task (which does make it a good project idea). Plus, I’m very skeptical of the quality of its generative abilities at this time. In fact, there is no way generative AI is up to the task yet. 

What It Can’t Do (Yet)

• Business, specifically the types of business interfaces required for freakthefishout.com (ie, supply chain, etc), is still about relationships. Humans interacting. Especially B2B interfaces. 
• Testing the products. OK, maybe an Optimus grabs a sample, walks down to the local pond…Forget it. Not yet. 
• Manufacturing quality control. Not on something like this. Maybe on something less physical.   
• Replace genuine human trust, relationships, or brand reputation built over time with customers.
• Create marketing that deeply resonates without human iteration, feedback, cultural nuance, or adaptation. 
• Manage the inevitable breakages in the supply chain. Supply chain providers still have desk phones. Who’s gonna call? Who do they call?  
• Manage regulatory complexity, compliance subtleties, contracts, disputes, or legal risk across borders. Governments are still humans. 
• What happens when the BassBlaster+ breaks in the field? 

To be honest, there are a host of others as well. Some of these will likely move to the “it’s possible” category over time, but it will require not just better AI (specifically generative AI), but also a cultural transition on the business side. Humans still run business. Especially B2B business. 

autoboxinabiz.ai is still in my “business ideas” list, but we’re not yet ready to shop it on Sand Hill Rd. 

Until agents, AI interaction was single-task oriented. 

You told it who it was (”You’re an expert tech marketing copy writer”) and gave it something to do (”take the provided background material about our company and product and develop a landing page in html for our product. We’re B2B, and the audience for the landing page is leadership.”)

Within that single task, the AI engine may do a few things (read the material, summarize it, find the parts for our audience, do its marketing thingy, and then generate html), but they were all directly related to a single request, and it has been provided with all of the information or told how to get all of the information.

An AI Agent can string a bunch of those requests together, including making decisions and finding the information it needs, like a person. 

Here’s a great example:

Prompt: “Look at my Google Calendar and find a time for a 30-minute meeting with Sarah in the next 2 weeks.”

Before agent functionality, you would have to do the following to make it work:

1. Open your Google Calendar (and Sarah’s if she shared it).
2. Check your availability (or screenshot your calendar and give it to the AI).
3. Ask the AI to draft a message for the invitation
4. You create and send that invitation (email, calendar invite, etc)

Now, the agent can do all of the following:

1. Access your Google Calendar itself.
2. Find availability.
3. Consider constraints such as travel time.
4. Ask you about preferences, such as when you typically like to meet.
5. Propose some times.
6. Actually book the meeting.

It’s now a handoff. Delegation in addition to automation. 

This is how we move toward organizational automation. 

Remember, the goal is to have AI work for you, or you’ll be working for AI. 

A while back, I wrote a little story about you, Mr. Fancy Pants, and how AI can take your innovative fishing rig idea and turn it into a full-blown business.

AI can’t yet do everything in that story, but now the framework is in place. That framework is called an “AI Agent.”

The best way to think of an AI Agent is that it’s the person who orchestrates all of the steps. It’s the conductor of the symphony. In an assembly line manufacturing system, it’s the conveyor belt. In an office, it’s the office manager who runs the ship. 

The AI Agent is the glue that pulls it all together — the connector. It can allocate tasks, serialize tasks, make decisions, and solve problems. 

By the way, AutoBizInABox.ai is available. I’m thinking about sucking it up. 

Maybe you should beat me to it.