My AI Takeaways After Around 2,000 Hours With Various Models

Practical lessons from a few years of hands-on work with ChatGPT, Claude, Gemini, Codex, Cursor, and other AI tools.

The best way to understand AI is to use it.

Not read about it endlessly. Not wait for the perfect enterprise policy. Not argue from the sidelines about whether it is overhyped, dangerous, transformative, or annoying.

Use it.

Open a free web model like ChatGPT, Claude, or Gemini and point it at a digital task you already do every day. Ask it to help draft something, explain something, summarize something, compare options, organize notes, troubleshoot an error, or pressure-test an idea. If you are more adventurous, pay the roughly $20 entry price for ChatGPT, Claude, Gemini, Cursor, or another serious AI tool and spend a month actually working with it.

After a few years and roughly 2,000 hours with these systems, these are the takeaways that have stuck.

AI Is Not Cheating

Using AI is not cheating. It is a tool.

Was Google cheating? Was spell-check cheating? Was asking a more skilled colleague a question cheating? Were Stack Overflow, Cisco forums, Microsoft forums, documentation sites, and search engines cheating, provided they gave you an actual answer instead of a smug "let me Google that for you"?

AI is another destination where useful information can be present. The difference is that it can also help shape, summarize, generate, reorganize, and iterate on that information with you.

That does not make it magic. It makes it useful.

Newer Models Usually Matter

Every generation of AI model is not just marginally better than the last. The jump is often significant.

If a newer model is available, use it. Test it against work you understand. Ask it the same questions you asked the older model. Give it the same messy task. You will learn quickly where it is better, where it is merely different, and where it still falls apart.

Treat model upgrades like you would treat better hardware, a better search engine, or a better compiler. You do not need to worship the upgrade. You should still take advantage of it.

Treat AI Privacy Like Cloud Privacy

Most privacy concerns around AI should be treated with the same seriousness as privacy concerns around cloud computing.

If your files are in OneDrive, Google Drive, iCloud, Dropbox, a hosted CRM, a SaaS ticketing system, or a managed email platform, your data is already on someone else's computer. That does not mean privacy is irrelevant. It means the question is not unique to AI.

The practical rule is simple: control what you put into the prompt window.

Do not paste secrets, credentials, protected data, private client information, internal strategy, or anything you would not be allowed to put into another third-party system. Use enterprise controls where appropriate. Use local or approved tooling when the data requires it. But do not treat AI as a special category of fear while ignoring the cloud services already holding your work.

Do Not Worry Too Much About Token Limits At First

Yes, models have context windows and token limits. You will hit them eventually.

Do not let that stop you from learning.

At first, the goal is not to write perfect prompts. The goal is to get enough experience to understand what a good prompt even looks like. Over time, you will naturally get better at giving context, asking for structure, setting constraints, and trimming noise.

When you do hit limits, ask the model to summarize the session into a handoff document. Save that file. Use it as seed context in a new chat, another model, or a different tool. Better yet, ask the model to compress your prompt into a shorter version: a bullet list, a project brief, a checklist, or a concise paragraph.

Prompting is a skill, but it is not a sacred art. You get better by doing it.

Move Beyond The Browser

Once you are comfortable with web-based models, try a desktop client, command-line tool, or IDE-integrated AI workflow.

That shift can be dramatic.

Browser chat is good for thinking, writing, summarizing, and general help. Desktop, CLI, and IDE-based tools are where AI starts to feel less like a website and more like part of your working environment.

For development work, this is especially important. A coding agent that can see the project, read files, run tests, inspect errors, and make scoped edits is a very different experience from copying snippets in and out of a chat window.

Once you get used to that workflow, it is hard to go back.

Give AI A Safe Workspace

If you are a power user, create a dedicated folder on your computer that AI tools are allowed to access.

Do not hand an agent your whole machine. Do not casually grant access to sensitive directories. Instead, create a deliberate workspace for AI-assisted work: drafts, experiments, scripts, notes, prototypes, and project files that are safe for the tool to read and edit.

That one habit can unlock a lot of productivity. AI becomes much more useful when it can work with actual files instead of disconnected fragments.

This is especially valuable for application development, automation, documentation, research organization, and general workflow management.

Do Not Be A Pushover

AI will often sound confident. That does not mean it is right.

If something feels off, challenge it. Ask for assumptions. Ask for alternatives. Ask it to explain the tradeoffs. Ask it to test its own answer. Ask it what would make the recommendation wrong.

The model will not always tell you when the premise is flawed. It may follow you down the wrong road if you lead it there. Good results come from scrutiny, iteration, and a willingness to push back.

Use AI like a fast collaborator, not an unquestionable authority.

Creative AI Should Be Embraced

AI in creative work should not be dismissed just because it changes who can make things.

Some people have vivid ideas but were not born with the manual ability to draw, paint, compose, edit video, design interfaces, or write code. Others can code but cannot illustrate. Some can illustrate but cannot build software. Some can imagine entire worlds but struggle to express them through traditional tools.

Should those people be left without creative input? Or should they get a chance to share the thing they can see in their head?

AI does not remove the need for taste, judgment, direction, or craft. It can, however, lower the barrier between an idea and its first visible form. That is worth taking seriously.

Enterprise AI Is Inevitable

AI in the enterprise is not something to fear as a distant possibility. It is already happening, and AI will affect jobs because progress always does. However, some useful questions are how quickly can you learn to work with it, where can it improve your output, and where does your judgment still matter most?

AI will change workflows, change expectations, change what teams can build, how quickly they can build it, and what skills become more valuable.

Adapt early.

Start With Curiosity

These are only a few of the things I have learned in my AI journey so far.

The best advice is still the simplest: if you are curious, satisfy that curiosity. Pick a model. Give it a real task. See what happens. Then try again with a better prompt, a newer model, or a more useful workflow.

You do not need to become an AI expert before using AI.

You become useful with AI by using it.

Happy creating.