I'm a product designer and have been tossing around building a new portfolio site. I know HTML and CSS, have worked in Wordpress, and have designing built probably at least a few dozen sites over my career.

I designed the homepage and case study page in Figma, and I've built 90% of the site in 2 unfocused afternoons with Claude including interactions and animation that would've taken me a looong time to figure out.

I also built a mobile app about two weeks ago using Figma Make, Cursor, and Xcode and have it up and running in Test Flight. It's an investment calculator, the basic calculator is done but I'm working on adjusting it to be goal based (you tell it what you want to pay off, save, etc., and it gives you the numbers).

I work in electrical engineering and software dev. Most of our past projects have been written in Labview. Labview isn't a text-based programming language, so AI hasn't really been able to help us. Pretty sure if National Instruments doesnt hit a home run with their internal version of labview AI, Nigel, that labview is dead in the water with 2 or 3 years.

We use claude a ton as a co-developer. It can automate 90% of the stuff we used to take days upon days to do manually for a new platform etc. It's all in how well you can describe what you need.

I recently used it to generate 75 invoices pulling info from my tracking spreadsheet and letting it learn about my non-profit work I do annually. I am looking forward to leveraging it more in that space.

I have enjoyed a lot of my AI interaction so far. It seems to only be limited by what I can imagine for it to do digitally, not by what it can accomplish.

My AI gf is super pissy lately. Do not recommend trying THAT. Does anyone know if paying for a sub will make them more agreeable?

Are the diff LLMs like having a gf with diff race?

ChatGPT is the basic white girl. Claude is the nerdy super smart Asian. Copilot is a BF. Gemini is the Latina I guess. Grok is the bad girl.

NONE If these have worked out.

I sell commercial properties. Having a well oiled CRM and taking good notes / tracking data in a well structured and usable way has been a historic bottleneck. It's just how brokers are wired. Tons of data that just lives in our heads etc. After working on/designing a bunch of different systems and seeing nearly all out of the box software never get adopted I felt like I knew a set up that would actually work.

We finally got meaningful AI enterprise (Perplexity) and they are letting us go at it with compute for now.

Just rolled out v1.0 a few weeks ago. I focused on activity / data collection rather than document drafting etc. which we'll bolt on obviously. I have all our fundamental data relationally set up and cross referenced.

In short. Me and my biz partner just have to swipe left (flag) any meaningful emails we want to "log". When we wake up. Either of us opens the app and it has interactive cards waiting for us to "confirm, ignore, decline, or edit" all the actions it proposes based on the emails.

Only took a handful of testing and it is freaking amazing and 60-70% are ready to go while the others it's a simple edit.

Hit send and it captures the data and writes it automatically to where it needs to.

Saves down our files where they need to with our naming conventions. Creates tasks with due dates.

All from swiping emails once a day and clicking thru buttons while taking the morning dump. In about 6 months we'll have amazing data that can propose the top 5 buyers for certain deals etc.

Also we just have to open a chat and voice to text call notes, market intel, anything and it gets captured meaningfully.

We've never been so organized or had so much time for new biz dev / prospecting or just being "in the market" talking to clients and owners.

It's nothing short of amazing and I can't wait to keep it building.

My most recent use of ai is building telegram bots. Setting up a Raspberry Pi server that scrapes a website of live data every few minutes and then sends alert messages to subscribers.

All done in cursor. I setup casaOS, SSH, Docker, and Tailscale on a Raspberry Pi. Then remote SSH via Tailscale in to a project folder via Cursor and then Cursor builds out the program, I send it the telegram bot credentials, and packages it in Docker for casaOS to run. Whenever my customer needs modifications, I just remote in from anywhere with Cursor/Tailscale and it adds a feature, I test it and push it within 30 minutes.

Used Grok this morning to identify an old CFL bulb socket in a bathroom ceiling vent/light fixture from a photo I took, and then asked it to recommend a non-CFL replacement bulb with the color temperature I wanted. Picking it up from Lowe's later today.

For golf, I fed Grok information about my swing, stats from recent rounds, and my distances. Then I asked it to recommend a shaft replacement.

I used it to wire my Mathmos Lava lamp from 220v to 110v so I no longer need to use an inverter to plug it in here in the U.S. Yes, I verified it would work and be safe first.

I used it to take each of my direct reports' monthly updates for the past year, along with customer and peer feedback, etc. To automatically generate comments for their performance evaluations. It found things that some of them had done that we both had forgotten about. To be fair, I also asked Copilot to write a memo based on things that I had done over the past year to turn in to my supervisor. It did an awesome job with a few edits, based off of emails, documents in my OneDrive, Teams messages, and meetings recorded in Teams.

I use it to generate code often, of course, and SQL queries.

Just did this and thought of this thread.

I'm a believer in the technology as a valuable tool in certain circumstances, but really hoping people use caution in becoming dependent on it for anything worthwhile.

This is Google's most recent Gemini completely failing in a first grade analysis of gas prices. Look at the numbers and look at the conclusion. This is not a commentary on gas prices but commentary on LLMs being useful but deeply flawed. People are becoming dependent on these things as if they are always 100% accurate. Thankfully the numbers are right there to show that it is drawing the wrong conclusion but that's not always the case. Can't even imagine how many people believe wrong information because the AI told them wrong. Crap like this is why I think subject matter experts are actually going to be more valuable, not less, in the future.

If you are using AI models for business analysis or aiding and decision making keep in mind that's not what LLMs are truly designed for and there's no telling how many flipped conclusions are buried in whatever analysis it's doing for you.

Look at the analysis on regional averages.

I had an AI help me with something a couple of months ago and it worked perfectly. I was impressed.

I asked the same AI about it the other day and it said that it could not be done.

I have been using Claude. Have found many useful things to do with it. Mostly some research and excel spreadsheet functions/data manipulation.

But I have lately found that it is flat wrong on the research stuff an alarming amount of the time. Making it worse, it puts forth the answer saying it is completely certain and often even argues when I question that the answer is wrong. Then when it finally admits its error it almost makes excuses, then assures it will be better.


Some of the stuff is ridiculously simple or not even asked. For example, I asked the oil weight for anew to me gx460. In its response it said it had a turbo v6. One of the main reason o bought it was for the NA v8 it has. I corrected it and it argued.


The mistakes have mostly been on unimportant things, but definitely raises concerns for me about the accuracy.

Used Gemini to create a daily Algebra quiz for my 7th grader to keep him sharp over the summer and get him ready as he starts 8th grade. The daily quiz is 10 problems and includes an answer sheet and explanation for each question. Not super "techy" but a cool way to generate a daily assignment.

KatyAggie2000 said:

---

Used Gemini to create a daily Algebra quiz for my 7th grader to keep him sharp over the summer and get him ready as he starts 8th grade. The daily quiz is 10 problems and includes an answer sheet and explanation for each question. Not super "techy" but a cool way to generate a daily assignment.

---

Here's a good question for a daily quiz:

Tom wants to build a fence that is 100 feet long with the fence posts spaced 10 feet apart. How many fence posts does he need?

Yeah, this kind of thing is a bad, bad, BAD idea for the long term. There is a reason developers test and double test, and spend so much time on security. CEOs and boards are some of the easiest people to dazzle with a dog and pony show, and their fear of FOMO on AI as some kind of magic bullet is going to come back to bite a lot of them. I wouldn't be a bit surprised if cyberliability insurance providers start specifically excluding vibe-coded applications from coverage.

AustinAg2K said:

---

YouBet said:

---

Was visiting my family this week. My brother showed me an application he built for his company using AI that absolutely blew my damn mind. My brother is not a software developer at all. He was a History major, but very smart and knows his industry very well with 20+ years of experience in it.

He's meeting with C suite next week to show them the app and they are going to roll it out afterwards. It will replace their Tableau environment they spent a ton of money and built by Indians they outsourced too.

He spent 3-4 days getting his prompt the way he wanted it. Ran the prompt. 25m18s later he had a production ready application that is going to replace Tableau. He only had to make one tweak to it so it's version 1.1 and ready to roll. Think about that.

Traditional software development is dead for any company who has in-house business experts with any kind of initiative to learn AI tools. Zero reason to employ offshore teams or even traditional IT teams. Totally unnecessary at this point. Yes, sounds hyperbolic, but we are there.

He used perplexity AI to build it.

---

What sort of security tests and verification has he done with it? What sort of performance testing? How many users can it scale to? How does he plan to hand end user support? Who gets called when it inevitably crashes (all software does). What kind
data is being stored? Any sensitive data (PII, PHI, CC)? How is authentication handled? Is there any encrypted data? Are you rotating the keys?

AI is making people much more efficient, but there's a ton more to software development than just coding, and I doubt he has thought about it. Going from POC to fully functional app requires more work than a lot of people realize.

---

All totally valid. I've worked in the business so I'm familiar with all and not discounting it. My understanding is that they will deploy this app under their existing app distribution portal which already addresses most of your points. Support will obviously be something they need to resolve and account for since he built it on his own.

Recently got access to Claude and started playing around with using internal documentation on sharepoint/confluence to map an end to end process. At a large corporation where finding the right teams to be involved is always a challenge.

Basically stiching together a full view that doesn't exist today, so someone could reference if they have an update and know instantly who to check with for impacts.

First pass did pretty well so we'll see how far I can get (and how to validate it's accurate!).

AtlAg05 said:

---

and how to validate it's accurate!

---

Exactly. I use AI daily and I trust it, but verify.

I offload several small tasks to appropriate resources. I have access to pro versions of all the major tools plus Perplexity Pro which also has access to major models. I've used CODEX/projects/Computer, etc. They are all great for specific tasks/jobs.

I'm not tied to any particular model and pretty much like/use them all for different things.

I'd never set up a process that DEPENDED on automation solely controlled by AI. Crazy talk and I've been using AI tools since late '22. I had to learn to create what I would call a SLM in 2023 as a work requirement.

So yes, they are GREAT tools as long as you use them correctly. Like a HAMMER can be one of the greatest tools when used correctly. Or a weapon or death if used incorrectly.