Are you actually using AI on your PC?

[Julio Franco]

AI is everywhere now, or at least that is what the industry keeps telling us. It is in browsers, search engines, image editors, office suites, developer tools, Windows, phones, and soon enough, probably your toaster. But there is a difference between AI being available and AI becoming part of your daily workflow.

At TechSpot, we do not use generative AI to write content. That remains a human job. But like many power users, we do use AI-adjacent tools where they make sense: proofreading text, summarizing long documents, reformatting HTML or messy tables, or using Photoshop's generative expand when an image needs to fit a different crop or aspect ratio. Those are not magic tricks, but they are useful shortcuts.

The more interesting question is whether AI has become something you actively rely on, or just another feature you occasionally poke at. Have you used any of the many models (GPT, Gemini, Claude, etc.), coding tools, image generation, translation, transcription, or Photoshop's AI features in a way that genuinely saved time?

What about agents? I have had one useful experience with an AI agent automating a painfully repetitive government form, clicking through the same broken workflow dozens of times. It worked, slowly, and it saved me the job. But I certainly would not trust an agent with broad control over my PC or personal data.

So, how about you? Is AI now part of your everyday PC use, a niche tool for specific tasks, or mostly hype you can live without?

Permalink to story:

Are you actually using AI on your PC?

 

[yRaz]

I do lots of home lab stuff.

>It's great for making scripts I'm too lazy to make myself.
>I use it for troubleshooting almost daily. Even if it doesn't fix it for me, it changes my thought process enough to figure it out on my own
>I made an agent to help me learn Spanish because I was a tawtwaffle and choose to learn German in highschool and Latin in college
>We started a DnD game at work and are using Gemini as the dungeon Master. We play in discord. We are often many states away so having a couple turns a couple turns a day, every day, is kinda an awesome way to play.

 

[Inthenstus]

It helps with my coding, and my health anxiety (insanity). Also, random questions about science and how to fix things.

 

[Alpine7995]

I've had good experiences reformatting data, asking GPT instead of Googling stuff works really well most of the time, and coming up with email drafts of stuff I don't care about too much (but want them to be well written anyway).

 

[GeoffreyA]

Back in 2024 and 2025, I used to chat with Copilot now and then, and Bard/Gemini to a lesser extent, probing them about their nature and with philosophical, existential questions. Copilot could talk in a poetic and metaphorical way I appreciated; Gemini felt too flat and serious. I also generated pictures for fun. Since the end of 2025, however, I have refrained from chatting, do not have the app installed any more, and am trying not to look at Google's AI summary, which I admit is hard.

 

[Kirby1]

Use it for coding, info gathering, whatever it can do faster/better basically.

 

[TooQik]

I'm not knowingly using it for anything.

 

[nnguy2]

I used it to write my employees annual reviews

Which is funny because I can tell theirs was written by AI due to the overly abundant redundancy. Like the same sentence written five different ways.

 

[toooooot]

I love it for simple questions. It is like the entire internet archive within the hand's reach.
I do not trust it when it comes to politics, I know how many of "THOSE" people
trained it.
It is amazing for people who are curious. I have asked gpt so many questions regarding science and technology.

But I still wish it left alone my RAM and SSDs.
It is year 2026, and I can no longer afford a 4tb SSD.

 

[Theinsanegamer]

I've used it precisely once, to troubleshoot a google chrome font flag failure that google search was utterly useless in finding.

For that, it was useful, although one must wonder if google search from 5 years ago couldn't have found the problem too, before it was enshittified.

Now image generating AI I have used, quite frequently, it is very useful for shitposting. Same with the AI voice features.

 

[Megalomaniac]

I enjoy watching AI videos.

#2 I use AI on my iPhone 17 to identify plant species.

#3 I use AI to scan photos and emails.

#4 AI screens my phone calls. I haven't had to talk to a single scammer in months.

#5 I use RTX on high settings in games that allow it. Nvidia also has noise cancellation AI. My 5090 has DLSS and I use it if the default mode is on for maximum settings "ultra".

 

[Axeia]

Useful as a Google alternative because Google itself has gotten terrible. Useful as a coding aid although mostly in a new language or to give some additional polish.
If I already know the language well trying to get it to generate actually good code might take such a big prompt and so many revisions it takes less time to just do it myself.

Definitely feels like it's heavily overvalued and I'd happily trade in the minor convenience it gives if it meant PC component prices went back to those of years ago.

Also sucks to see how much of the web has turned into AI generated garbage that'll be headed to train models on to make even more garbage.

 

[loki1944]

Nope.

 

[Nobodysfool]

I use generative AI only when I feel I need a quick, nuanced response to an inquiry (Let's face it, not everything can be answered by search engines). That being said, this is no substitute for research and critical thinking, as AI is notorious for "hallucinating" information.

 

[Burty117]

I use Copilot for helping me create powershell scripts or finding EntraID / Graph commands quickly. Copilot for Teams is also really good, when I miss a meeting, it transcodes, summarises, notarises and points out where I was mentioned and why, super useful in Teams.

Cowork (Frontier) is Claude underneath, This has been great for doing tenant-to-tenant migrations and I need to mix and match hundreds of users and shared mailboxes and what not from different .csv files. It's also great at creating a presentation for me to work from, and really good at researching as well.

Gemini is generally okay at helping me research an issue, it can give quick easy answers to simple questions, but it's not up-to-date enough that the latest and greatest stuff will show up, and still needs a Google search to find the latest articles on a subject.

Plus the other bits of AI I don't even think about that I use all the time, Nvidia DLSS, Apple's Siri for call screening, Photo recognition in Immich, Notification summaries etc...

There's also plenty of things I've found AI to be relatively useless at still, and I do think AI's inability to remember much history to be quite an Achilles heel for AI.

 

[MarkHughes]

At work yes, I use it like a very clever search engine or to write unit tests, scripts and sometimes small internal apps, AI code never goes into our code base though, there is always a human in the loop.

At home, No, not really, I simply don't have a use for it in my every day life.

 

[TechSam]

I've been exploring some heavy duty ideas. How would you decompile an old game and recreate the source code. Codex seems to have a great grasp on how to do it. Interested to see how close it can get to it, or if its just bluffing.

 

[JaredTheDragon]

Absolutely not.

 

[Alfonso Maruccia]

Only for translations in different languages I don't know, or for which I can't even understand the alphabet because it doesn't use latin-based charsets.

Apart from that, I don't trust AI for anything. The few times I try to use a chatbot for something a bit more complex - like translating a latin motto in Hieroglyphs, Hieratic and ancient Greek, or explaining a piece of code from an x86 emulator - I'm usually forced to discard the results because I fear they have been hallucinated.

AI agents feel like autoeroticism, only that you are not taking part in any sort of fun :-D

 

[kaosken]

The only AI that makes sense, but will never happen, is fixing traffic lights at high-noon everywhere.
And, the only AI I have found that now works is StarLink's tech support. (it is fantastic)

 

[human7]

I use it all the time at work. I used to write code, now I tell it to write code, and the platforms we use are agentic (it can create and edit files, execute code, run jobs, create dashboards, etc.). I'm a data scientist so a lot of the code is boilerplate stuff, but I've had to think about where I personally provide value, since it isn't in writing code anymore. It's definitely been a shift towards code review, higher level thinking, presentations, interpretation, etc, but I feel like AI isn't so far behind from being able to reliably do some of that. Most of the work we do is at scale (billions of records by thousands of columns, so we frequently stand up ephemeral 1000+ CPU core clusters), and the AI still isn't entirely reliable when it comes to debugging failures related to processing data at scale. It can say "you ran out of memory" or whatever but it hasn't (yet) been able to solve the problem in the most efficient way unless we give it our wisdom on the subject.

We also use AI to write documentation, but let's face it, developers never liked writing documentation in the first place, so that's been a real win (since it's AI slop documentation versus what used to be little to nothing). That said, AI code and documentation can get unnecessarily verbose and unwieldy.

On my personal, I do have an older rig with a 2080 Ti that I use to run local models (do all my gaming on a 5090 rig these days, haven't bothered to setup AI on that one). Those 11 Gigs of VRAM on the 2080 Ti are actually used completely but they are enough to get medium sized models running just fine with decent amounts of context. When I'm not using a local model I often use Anthropic for web searches or asking it questions (everything from philosophy to physics to what the impact of a specific change in the tax code or some law would affect). But I have found myself overly reliant on it, so I've been forcing myself to do traditional web search first.

One caution to those who frequently use AI for quick answers: it can be like candy. I got to the point where I asked it a question, and it said "I already answered this question in an earlier chat, but here's the answer again", and I had completely forgotten I had asked it the exact same question a week earlier. It was at this point when I realized I had been relying on AI a bit too much for too long and was using AI to get a quick dopamine fix every time I had a question rather than actually thinking about it.

 

[human7]

I've been exploring some heavy duty ideas. How would you decompile an old game and recreate the source code. Codex seems to have a great grasp on how to do it. Interested to see how close it can get to it, or if its just bluffing.

I asked it to do that once, about a year and a half ago. I found some old retro game that was small because I needed it to fit in context. It worked okay, but I never tried to see if the generated code would actually run.

It would be a lot easier today with better maturity around tool calling, orchestration, and the models are just better. One problem is that AI loves to fill in things that aren't real. It will tell you it implemented something from the game or whatever but in reality it might have made up its own implementation in places to fill in the gaps, rather than doing what the original did. At work we discovered this problem simply in converting some Java code to Python, where at least there we had the benefit of source code in both places. It will be harder to detect that in a decompiling or disassembly regime.

 

[bloomsun]

I use AI every day. Kimi AI specifically. I use it to learn stuff, research stuff, create visual slides for fun, fact-check stuff. There is so much you can do where AI opens the doors to your own curiosity way more than anything ever could before. It also makes learning so much easier as you don't have to search manually with some ancient search engine, but just ask AI and see what sources it picks up, and maybe you then check the source and find more info. It's a rabbit hole. AI is the greatest tool we now have and we can't really imagine what the world will be in 20 years. Look what happened in the last 20 years.

The only AI that makes sense, but will never happen, is fixing traffic lights at high-noon everywhere.
And, the only AI I have found that now works is StarLink's tech support. (it is fantastic)

Already happening in china.

AI-powered traffic management eases congestion in Shanghai

Shanghai has implemented an AI-powered traffic management system designed to improve traffic flow and reduce congestion.

english.shanghai.gov.cn

 

[learninmypc]

Wish I could get it off my computers.

 

[TechSam]

It worked okay, but I never tried to see if the generated code would actually run.
One problem is that AI loves to fill in things that aren't real.

In my experience from transitioning to AI Chat -> AI desktop: AI hasn't been really capable of coding until a few months ago with desktop agents. It would always hallucinate, make bad code, so I was never able to be a "good programmer" (I'm a technical designer). Now I'm Using Codex Desktop / Claude Code, it has been life changing. For my decompile, its got every tool in the box to run itself, verify everything, make evidence for what does and doesn't work. Evidence and context driven AI development seems to be the way and its providing great results.

Its not just "put a prompt" in and you're done, far much complicated, a lot of workspace maintainance, questions, project direction.

Can highly recommend giving your project another try if you were interested in it. The PDB files, .exe decompile and many other things you can do to find bread crumbs to craft a game back together.