I once ran an experiment: I asked GPT-4o, Gemini, and Grok to analyze the same business decision, then used Perplexity to help find relevant information.
In that experiment, GPT-4o gave a steady mainstream answer, Gemini pointed out three risks the first answer had not mentioned, and Grok proposed another option. After three rounds of discussion, what I gained was not just a single answer, but a decision map for comparing options.
This does not mean that the models will always display those characteristics, nor does asking more models guarantee a correct conclusion. But it showed me that, instead of completing every step myself, I could also focus on designing the question, arranging the comparison, and deciding what to adopt.
Look at Your Work Before Getting Trapped in Tool Anxiety
Models and tools keep changing, and keeping up can be useful. But if every update forces me to rearrange my direction, I can easily spend my time chasing features without clarifying what the work actually requires.
I would rather begin with three questions: Which steps need to be repeated? Which results require professional review? Which mistakes would carry a high cost?
Seen this way, tool proficiency still has value, but it needs to be assessed within a specific task. Producing a first draft faster and delivering a reliable result are separated by verification, revision, and acceptance. My own role also changes gradually within those steps.
Moving from Execution to Orchestration Still Requires Understanding the Content
My current strategy focuses on three kinds of work. They are not moats that technology can never affect; they are capabilities I find worth continuing to practice.
Problem Definition. AI can help propose questions, add context, and compare options. I still need to establish whether this is the problem worth addressing now. In The Always-On Economy in the AI Era: Work Keeps Running—Who Decides the Direction?, my concern is how people preserve understanding, intervention, and stopping rights while work continues—not a prediction that people inevitably become observers.
Structural Design. I once connected the content workflow for eight social platforms through Apple Notes, Google Sheets, and the OneUp API. That experience made me pay closer attention to handoffs: where a draft is produced, where it is checked, and which content may enter the publishing workflow. Connecting tools does not by itself establish output quality, workload, or time saved.
Value Judgment. Different models may offer contradictory advice, and they may also overlook the same issue together. I need to check the grounds for each claim, consider the costs of the tradeoff, and take responsibility for the decision I adopt. AI can help with comparison; its answer cannot complete that choice for me.
A Role Army Is a Way to Orchestrate Work
I use “role army” to describe a way of assigning tasks: ask one tool to organize the claims, another to look for counterexamples, another to propose a different option, and then return to the relevant sources.
The earlier division of work among GPT-4o, Gemini, Grok, and Perplexity was the configuration and observation from that experiment, not a permanent ranking of model capabilities. Versions, questions, prompts, and accessible information can all change the result. Roles can be reassigned; calling a tool a “critic” does not mean it will always find the flaw.
Finding sources is also only one step in verification. I still need to establish whether a source really supports the conclusion and whether its date and scope fit the claim. Agreement among several models does not mean they have provided independent evidence.
In Building a small AI tool alone: from output counts to acceptance and maintenance, I am more concerned with which work one person can actually take on and how verification, revision, and maintenance should be arranged. That helps me decide the next step more than converting myself into an equivalent number of team members.
Draw a Strategy Map You Can Revise
When facing change, I ask three questions: Which work is suitable for assistance from tools? Which judgments require deeper understanding on my part? Which of my experiences can help address the specific problem in front of me?
For me, ten years of industry experience, theological training, entrepreneurial practice, and technical understanding are materials I can currently bring together when working on a problem. They are not a guarantee against future compression, nor do they mean that someone else cannot build a different advantage. Their value still needs to be checked through actual results and feedback.
A strategy map cannot be drawn only once. I can begin with one recurring task, separate production, verification, and acceptance, record where a tool helps and where I still need to build capability, and then adjust the division of work.
I want more than greater speed. I want to understand more clearly what I am solving, why I am adopting an answer, and which result I am willing to take responsibility for. This cannot guarantee that I will prevail amid change, but it gives me a firmer basis for the next step.
Revised November 2026: The model assignments and tool workflow are retained as personal experience from that time, not as current capability rankings or efficiency comparisons.


💬 Comments
Loading...