I keep running into posts that make AI sound like a black belt discipline you’ll need years to master. So it was refreshing to find a LinkedIn creator who says the opposite. The models are now too good for that, and the original poster admits he could have written “AI is so hard to master” just to grab clicks, but chose not to. Instead, he laid out a short list of habits that actually move the needle. I read it twice, and it’s one of the most practical AI workflows I’ve seen this year.
Below is the author’s four-step system, plus my take on why each step works, so you can put it to use today.
Step 1: Use the app, not the browser, and pay for it
The first step sounds almost too simple. Download the desktop app. The author says every model runs better there than in a browser tab. Then open ChatGPT Work or Claude Cowork, which this contributor calls unmatched right now, with a small “(yet)” attached.
Why it matters: the apps tend to get the good stuff first. File access, connectors, background tasks, and the agent-style features all land in the desktop clients before they show up on the web. If you’re still copy-pasting into a browser tab, you’re using a stripped-down version of the tool.
The second half of this step is blunter. Always pay for AI. The expert is serious about this one: the free plan won’t get you anywhere. I agree. Free tiers give you the weakest models, tight usage caps, and none of the integrations that make the rest of this list possible. Twenty bucks a month is the cheapest productivity upgrade you’ll buy.
Step 2: Start with the smartest model, then switch mid-task
This is where the post got interesting for me. Most people pick one model and stick with it for the whole session. The original poster does the opposite.
- Write a long first prompt that gives the model a lot of context.
- Use the top-tier model for that opening move. The author names “GPT-6 Astra-Medium” in ChatGPT or “Fable-5.1 High” in Claude.
- Once the plan is built, switch to a cheaper model (Sol or Opus) to execute it.
Why it matters: the expensive models are brilliant at understanding messy context and shaping it into a plan. But once the plan exists, the heavy lifting is mostly follow-through, and a cheaper model handles that just fine. You get top-shelf thinking where it counts and save your usage limits for the parts that actually need them. Think first with the smart model, then work with the fast one.
Step 3: One AI can’t do everything, so bring in specialists
Even the best general model has weak spots. The post’s author keeps a short bench of tools, each covering one job:
- Slides: Gamma
- Voice to text: Wispr Flow
- Automation: Grok Bot
- Meetings: Granola
- Learning: NotebookLM
- Video: Seedance
- Excel: ShortcutAI
Why it matters: each of these does one thing very well, and trying to force a chat model to produce a polished deck or a clean spreadsheet usually ends in frustration. The trick is not to live in these tools. They’re plug-ins for your main workflow, which brings us to the step that ties everything together.
Step 4: Spend most of your time in Claude or ChatGPT, connected to everything
Here’s where the author’s setup clicks into place. Despite the long tool list above, this savvy professional spends most of his time in just one place: Claude or ChatGPT. The difference is that he connects them to everything else.
- Go to Settings, then Connectors (Claude) or Plugins (ChatGPT).
- Add Gamma, Granola, Notion, Drive, Gmail, Outlook. There are 1,000+ connectors to pick from.
- Use Wispr Flow to record your prompt instead of typing it.
That third point surprised me. The author says talking forces you to ramble for longer, and longer prompts mean better results. I tried dictating a prompt after reading this and ended up giving three times more context than I would have typed. The output was noticeably sharper.
Once the connectors are in place, a real workflow looks like this. Claude reads your meeting notes straight from Granola. You ask it to draft the email follow-ups, and it already knows who said what. Then you hand the summary to Gamma to turn it into a deck, because, as the creator puts it, Claude and ChatGPT are “pretty bad at slides.” No copy-pasting between five tabs. One conversation, several apps working behind it.
The lesson I took away: the winning move isn’t finding the perfect AI tool. It’s picking one home base and wiring everything else into it.
Bonus: the power-user path
The author teases a deeper level for anyone who insists on going further, though he jokes the post was already too long to explain it fully. The outline is still worth having:
- Upgrade to Claude Code or ChatGPT Codex.
- Connect data platforms to it. The author’s favorite is Apify.
- Prompt it with a mix of /plan and /goal.
This is the agentic tier. Instead of chatting, you’re giving an AI a goal, access to live data, and room to plan its own steps. It’s a bigger jump, but the first four steps set you up for it nicely.
Honestly, what I love most about this post is how little it asks of you. No secret prompts, no 40-step frameworks. Just the app, a paid plan, smart model switching, a few specialist tools, and connectors doing the glue work.
If you want the full breakdown in the author’s own words, go check out the original LinkedIn post. It’s a fast read, and the comments are full of people sharing their own connector setups too.