The six AI working modes
A practical menu for using AI at work.
AI is not one thing. It is a menu of working modes. The skill is knowing which one to use, how to guide it, and when to switch.
Over the past year, I have used AI, specifically large language models, almost daily in real operational work. Strategy, hiring, writing, system design, client communication, internal training. Not experiments. Actual decisions with consequences.
One thing has become clear. AI is not one thing. It is a menu of working modes.
Most teams are dramatically underusing AI. They treat it like a faster version of Google and then assume they have “adopted AI.” That is useful, but it barely scratches the surface of what these systems can do inside real operations.
Different problems, time horizons, and roles benefit from different ways of interacting with these systems. The goal is not to level up into a single ideal mode. It is to choose deliberately based on the work in front of you. Most of the underperformance I see comes from using the wrong mode for the problem.
Most meaningful work involves switching between modes, often within the same task. As these systems improve at carrying context, switching becomes easier. It also becomes more important to be intentional about how and when you use each mode.
What follows is how I currently think about working with AI in practice. It is not theoretical. It is evolving. And it is meant to be adapted.
Mode 1
Search Engine
When you need fast orientation.
Using AI to quickly retrieve information, definitions, summaries, or explanations. It is often the fastest way to get oriented when you are stepping into unfamiliar territory.
When to use it
- You need fast orientation.
- The answer already exists somewhere.
- Accuracy matters more than originality.
Good for
- Facts and definitions.
- Learning unfamiliar topics.
- Quick context switches between domains.
Watch out for
- Shallow understanding.
- Over-trusting confident answers.
- Treating summaries as truth instead of starting points.
AI is very good at sounding certain. That does not mean it is correct. Treating confident summaries as truth is one of the fastest ways to make avoidable mistakes.
In practice, my prompts sound like this
- “Summarize the key differences between X and Y in plain English.”
- “Give me a quick overview of this topic so I know where the real leverage is.”
- “What are the main schools of thought here and how do they differ?”
- “If I only had ten minutes to understand this, what would I focus on?”
- “What is commonly misunderstood about this topic?”
Mode 2
Content Generator
When you need momentum.
Using AI to produce drafts, copy, or structured text. This is often less about creativity and more about momentum. It helps you get something on the page so you can react to it.
In practice, I rarely say “write this.” I treat the output as a starting point. The goal is to refine, constrain, and sharpen, not to outsource authorship.
When to use it
- You know what good looks like.
- You plan to edit, reshape, or inject perspective into the output.
Good for
- First drafts.
- Getting past blank-page friction.
Watch out for
- Generic voice.
- Over-polished but shallow output.
- Mistaking fluency for quality.
AI is very good at producing plausible content. It is not very good at producing distinctive content. If you are not injecting judgment, you are producing noise at scale.
The role of constraint
The quality of output improves significantly when constraints are clear. Tone, audience, stakes, format, tradeoffs. Vague prompts produce generic answers. Specific constraints sharpen thinking.
In practice, this usually means iterating. Tighten the scope. Clarify the audience. Remove what does not feel authentic. The first draft is rarely (never) the final draft.
In practice, my prompts sound like this
- “Draft a first pass. I’ll refine it.”
- “Tighten this without losing the human tone.”
- “Keep the human in this.”
- “Take the AI out of this.”
- “Sharpen the argument, but do not make it harsh.”
- “Make this clearer and more direct. Remove anything generic.”
- “Grade this. Where is it weak?”
- “This feels over-polished. Bring back some texture.”
- “Keep my voice. Reduce the noise.”
- “Give me brutally honest feedback.”
Mode 3
Productivity Assistant
When the work is messy.
Using AI to reduce administrative and cognitive load by structuring, summarizing, or organizing information. This mode is less about insight and more about clarity. It helps you clear space so you can focus on higher-value thinking.
When to use it
- Decisions are already made.
- The work is mechanical, repetitive, or organizational.
- You want to free up attention for something that actually requires judgment.
Good for
- Meeting summaries.
- Action items and next steps.
- Cleanup and reformatting.
- Turning messy notes into something usable.
Watch out for
- Confusing motion with progress.
- Over-organizing instead of deciding.
- Spending time polishing artifacts that do not change outcomes.
This is one of the easiest modes to overuse. It feels productive because something tangible is produced, even when nothing meaningful has changed.
In practice, my prompts sound like this
- “Summarize this meeting and clearly separate decisions from open questions.”
- “What are the actual next steps here, and who owns them?”
- “What decisions are implied but not explicitly stated?”
- “Turn this into a one-page executive summary I can send.”
- “Clean this up so it is scannable in under two minutes.”
- “What is noise versus signal in this thread?”
Mode 4
Specialist Stand-In
When you need to pressure-test an idea.
Asking AI to simulate a domain expert so you can explore ideas, generate options, or critique an approach. This is not the same as hiring an expert. It is a way to surface patterns, risks, and considerations quickly.
When to use it
- You need a starting point.
- You want to pressure-test an idea before bringing it to others.
- You lack deep expertise, at least for now.
- You want to see how someone in a specific role might react.
Good for
- Early exploration.
- Generating alternative approaches.
- Surfacing obvious gaps or blind spots.
- Stress-testing assumptions before committing publicly.
Watch out for
- Outsourcing judgment.
- Using AI to justify a decision you have already made.
- Mistaking plausibility for correctness.
- Treating simulated expertise as real-world accountability.
Treat this mode as rehearsal, not authority. Confusing simulation with expertise is a quiet but costly mistake.
In practice, my prompts sound like this
- “Act as a skeptical board member and critique this.”
- “If you were an operator who had seen this fail before, where would you push back?”
- “What risks am I likely underestimating?”
- “What would concern a client in this proposal?”
- “If this went wrong six months from now, what would have caused it?”
- “What assumptions would an experienced practitioner question?”
Mode 5
Thinking Partner
When the problem is ambiguous.
Using AI to think out loud, test assumptions, explore tradeoffs, and clarify direction. This mode is not about producing output. It is about sharpening judgment.
I often use it when I am unsure what I think.
When to use it
- The problem is ambiguous.
- Tradeoffs matter.
- Second-order effects are real.
- You are shaping direction, not just producing artifacts.
Good for
- Clarifying what the real question is.
- Stress-testing assumptions.
- Exploring incentives and unintended consequences.
- Mapping how something might break at scale.
Watch out for
- Over-iteration.
- Talking instead of deciding.
- Seeking validation instead of challenge.
- Losing momentum because the conversation feels productive.
This is where AI can quietly become transformative, if you are willing to be challenged. Sometimes I explicitly ask for challenge rather than validation. The quality of the output changes when the intent is clear.
This mode also functions as a mirror. When you articulate your reasoning clearly, AI reflects structure back to you. Patterns become visible. Gaps become harder to ignore. Used intentionally, it can surface inconsistencies in tone, logic, or assumptions before they show up in the room.
The more you work like this, the more AI will default to this stance. It learns that you want to be challenged, and it will challenge you. It will make you better. Many of the people who complain about AI being “too nice,” or that AI “doesn’t help them,” are failing to push it into this mode.
In practice, my prompts sound like this
- “Help me think this through. I don’t want an answer yet.”
- “What assumptions am I making?”
- “Where is my thinking off?”
- “I’m too close to this. What am I missing?”
- “What breaks if this scales?”
- “Argue the opposite side of this decision.”
- “If this fails, what will have caused it?”
- “Is this clear thinking, or am I rationalizing?”
- “Give me brutally honest feedback.”
Mode 6
System / Operating Layer
When something proves repeatable.
Embedding AI into repeatable workflows, tools, or processes so that good judgment scales beyond a single person. This mode is less about individual output and more about institutional leverage.
It is where experimentation turns into an operating model.
When to use it
- A pattern repeats.
- Good versus bad output is clearly distinguishable.
- The cost of inconsistency is high.
- Human judgment is still required, but predictably.
Good for
- Scaling best practices.
- Reducing dependence on individual heroics.
- Creating consistent client or internal experiences.
- Increasing leverage without increasing headcount.
Watch out for
- Freezing bad assumptions into systems.
- Automating too early.
- Over-engineering before the work is stable.
- Trying to automate thinking instead of execution.
Systems amplify whatever they encode. Automation does not fix unclear thinking. It makes it repeatable.
If you cannot do it well manually, you cannot automate it well. Prompts are for learning. Systems are for scaling.
In practice, my prompts sound like this
- “Help me turn this into a repeatable process someone else can execute reliably.”
- “What should be standardized versus left to judgment?”
- “If we scaled this across the team, what would break?”
- “What inputs and constraints need to be explicit for this to work consistently?”
- “How would we document this so a new hire could use it without me?”
- “What metrics would tell us this system is degrading?”
- “What guardrails should I put on this?”
Mode switching matters.
Most real work does not stay in one mode. It is common to:
- Start in search mode to get oriented.
- Move into content mode to generate a draft.
- Shift into thinking partner mode to test assumptions.
- Then systematize what works if the pattern repeats.
The leverage is not in mastering a single mode. It is in knowing when to switch. Staying in the wrong mode is where most AI efforts quietly stall.
The first answer is rarely the final answer. Iteration, with increasing clarity and tighter constraints, is where most of the leverage shows up.
Different problems require different levels of thinking. High-stakes decisions should not be treated like formatting tasks. Early exploration should not be prematurely systematized. Matching the mode to the problem, the stakes, and the time available is what makes this practical rather than theoretical.
A note on originality and voice.
AI is very good at producing plausible content. It is not very good at producing distinctive content.
If the output could apply to almost anyone, it probably needs more human input. If you cannot point to a real experience, decision, or point of view behind it, it is not done yet.
A useful gut check“Could someone else have written this without me being here?”
AI can accelerate writing. Don’t let it replace authorship. That is when the content becomes generic.
Your voice doesn’t erode because you used AI. It erodes because you stopped thinking.
A note on optimization.
AI is very good at optimization. It is not very good at preserving human texture. Using AI to improve something can sometimes optimize the voice out of it.
I have noticed that an A- draft is sometimes stronger than the fully polished version. Not because it is cleaner, but because it feels lived-in and specific.
Optimization is not the same as clarity. And clarity is not the same as character.
How I use these modes.
In practice, I switch modes frequently. I often start by using AI as a search tool to orient myself. If I am writing, I may move into drafting mode to get momentum. When the problem is ambiguous or high-impact, I slow down and use AI as a thinking partner to pressure-test assumptions and explore tradeoffs. If something proves useful and repeatable, I may later systematize it.
I do not treat any one mode as better than the others. The goal is to match the mode to the problem, the stakes, and the time available.
I’m still refining how I use these tools, and I suspect most teams are as well. Some weeks I feel like I’m thinking more clearly because of them. Other weeks I realize I am over-optimizing and have to pull back. That tension is part of the work.
Mode switching is the skill. Not prompting. Not the tool.
We teach the six modes in AI Foundations