Most People Never Get Past Chatting With AI

Are you using the right AI for the job in front of you?

I split mine by the job. ChatGPT handles daily driving. Claude handles technical coding and Excel analysis. Gemini handles image and video creation. NotebookLM handles research. Perplexity handles real-time search and sentiment analysis. Each one stays in the stack because it has a specific strength.

The most effective AI users curate a toolkit, with each tool on the job it handles well. Does your current setup match the work? Sort the tasks you ran this week: daily driving, code, Excel, images, video, research, real-time search, sentiment. If one tool is sitting on a job outside its strength, move that job.

Where you actually are

So where are you?

Level 1 is AI-assisted. You chat back and forth for ideas. Useful, and it’s where most people get stuck. They never bridge the gap from conversation to execution.

Does this sound familiar? Lots of chatting. The task still sits with you. That’s Level 1.

Level 2 is AI workflows. The AI executes a task because you connected it to a tool you already use. Email is the one I would move first. Connect the AI to your inbox so drafting and sending happen through that integration.

Level 3 is AI agents. Picture autonomous digital employees that monitor an inbox and pre-draft responses based on your context. You still approve the response.

Start at Level 1 and get genuinely comfortable prompting. Pick one repetitive task, email drafting, and move it to Level 2 by connecting the tool to your inbox. Let comfort and competence show up before you attempt agents.

Two text files beat a longer prompt

How many times did you explain yourself to a fresh chat this week? Identity. Values. The way you make decisions. The way you write. Then the session ended and the next one needed the speech again.

Two text files hold that context.

Who I Am. Call it About Me if that name fits better. Put your identity, your values, and your decision-making framework in it. Load the file at the start of a session. You get consistent context immediately, without re-explaining who you are or what you do.

How I Write. Fill it with samples of your LinkedIn posts, your emails, and your blog intros. The model can then mimic your voice across those mediums without starting from scratch. That file is what solves the blank page.

These files are also the foundation for personalized skills or plugins, whenever you want them. When you make this your tweak for the week, create both and load them before you ask for anything.

Close the loop on your own voice

Generic output has a simple cause. You take the draft into your email client, you fix it there, and the model never sees the fix. Editing outside the app breaks the feedback loop.

Edit inside the chat. Refine the tone and the word choice in the thread. When the draft is done, paste the final version back into the chat. Tell it, “This is the final version. Learn from this style.”

After you refine the output, the note matters. You’re telling the model that this version is the standard. Repeat that 4 or 5 times. The model internalizes your voice, and later one-shot prompts come out closer to the way you write. When the writing sounds generic, run the loop on your next 3 emails.

300 leads, one prompt

Some tasks are a chain of lookups. Lead enrichment is the one I can walk you through. Talent sourcing works the same way.

Perplexity Computer runs autonomous agents for this. You feed it a list of 300 leads. An autonomous agent browses the web and returns a fully enriched dataset, with company size, location, and the other qualifying data. Minimal human intervention. One prompt can process hundreds of leads the same way and pull qualification status along with the rest.

When research is the task you usually do manually, run it through Perplexity Computer and read the dataset before you add more process around it.

Draft the email out loud

Voice cuts the friction of starting. Wispr Flow is a dictation app with an AI layer that understands intent. It can correct you in real time. Say “I want four… no, five hamburgers” and you get “five hamburgers.”

I know. A hamburger. The correction is the whole point.

Prompting feels like a natural conversation, and the gap between thought and output gets smaller. When you want less friction at the start, install Wispr Flow, or another dictation tool that understands corrections, and draft one email with your voice. Notice whether it feels faster.

One tweak a week

Implementing every tool in the same week is a fast way to overwhelm yourself. One new AI habit or one new tool per week. Pick the tweak on purpose. A new prompt habit counts. A new tool counts. Basic prompting, then workflows, then agents, as comfort and competence catch up.

I’m no security engineer. The practical rule is still simple. Dedicated AI browsers are a poor fit for sensitive tasks like banking, because of the security risk. Use the Claude Chrome extension on a standard browser like Chrome. Enable AI assistance only when you need it, and disable it afterward. Check your extensions and disable any AI tool you are not actively using. That check can be its own week.

Here is the move. Open the AI you already use for writing. Connect it to your inbox. Let it draft and send the next email through that workflow. One task. This week.