Five of the most valuable AI lessons I can give you today:
1. Leave Windows.
Use Mac or Linux. If you can go Linux, even better. You’ll learn more, control your environment, understand your tools better, and get hands on experience with local models.
2. Use Hermes Agent.
Even if you can’t afford subscriptions, download the desktop app, sign in to Nous Portal, and start with the free models available there. Learn the environment by actually using it.
3. Stop following viral AI accounts that teach you nothing.
Find people building open source projects, testing strange ideas, sharing real work, and showing you angles you haven’t considered. Some of the smallest accounts here have the highest value.
4. Before building anything, ask yourself:
Would I use this every day?
Does it solve a real problem or improve my productivity?
Can I build and prove it myself, or does it actually require a team and serious money?
If you can’t answer those clearly, don’t build it yet.
5. Never lock yourself into one subscription.
If you have $200 a month, don’t dump it all into one provider.
One model gives you one perspective and often misses its own mistakes. Spread your budget across different models. Use each for what it does best, and let them expose what the others missed.
Those are probably the five most useful pieces of AI advice I can give you today.
I love computer use and rely on it a lot, but it still has two major problems:
It’s painfully slow almost everywhere, with Codex being way ahead of the rest.
And the token consumption is insanely high.
I’m canceling Claude, and honestly, it feels like a relief.
Opus 5 is unbearable. Tight usage, painfully slow models, endless restrictions, and hallucinations on top of it all.
The subscription has lost its value, especially when better alternatives are everywhere.
Grok 4.5 is a huge leap forward, but xAI’s subscriptions, especially SuperGrok, are still among the worst value in the market.
I personally subscribed after Elon told me he would increase the limits and make the plan easier to use. So far, that still hasn’t happened.
The $30 plan shares one usage pool across Grok chat and Grok Build. Using Grok Build seriously, I couldn’t even make the plan last two days. Outside Grok Build, usage seems to burn several times faster. With serious use, you can realistically drain it within hours.
That becomes even harder to justify when you compare it with competitors.
For $20, ChatGPT Plus gives you nearly unlimited everyday chat, very generous image generation, around 40 Deep Research runs per month, voice mode, and Codex with Luna, which is excellent for light and medium coding tasks.
With Grok, everything competes for the same narrow pool.
That’s the core problem. The model is excellent. The subscription isn’t.
Until the limits improve significantly, I can only recommend SuperGrok for light use. For serious daily work, it simply isn’t practical.
Five of the most valuable AI lessons I can give you today:
1. Leave Windows.
Use Mac or Linux. If you can go Linux, even better. You’ll learn more, control your environment, understand your tools better, and get hands on experience with local models.
2. Use Hermes Agent.
Even if you can’t afford subscriptions, download the desktop app, sign in to Nous Portal, and start with the free models available there. Learn the environment by actually using it.
3. Stop following viral AI accounts that teach you nothing.
Find people building open source projects, testing strange ideas, sharing real work, and showing you angles you haven’t considered. Some of the smallest accounts here have the highest value.
4. Before building anything, ask yourself:
Would I use this every day?
Does it solve a real problem or improve my productivity?
Can I build and prove it myself, or does it actually require a team and serious money?
If you can’t answer those clearly, don’t build it yet.
5. Never lock yourself into one subscription.
If you have $200 a month, don’t dump it all into one provider.
One model gives you one perspective and often misses its own mistakes. Spread your budget across different models. Use each for what it does best, and let them expose what the others missed.
Those are probably the five most useful pieces of AI advice I can give you today.