Yes. It's mindblowing. We are using AI and LLMs to write the code for our new operational platform.

We put a 1.5 page product spec on paper today written in English with very little tech jargon. Ran it through ChatGPT and basically got about 80% of our code out of that we will need. Total of about 1 hour of work. Dev will tweak the last 20% to finalize it. Whole process shaves weeks to months off the traditional process.

Anything agile is now snail's pace comparatively.

This stuff is world changing for the tech space.

The key here is learning how to use ChatGPT (and peers) to effectively get what you want out of it. Instead of asking it questions you should write a persona or use case and then tell it what you want it to do.

This will reduce the number of people I have to hire in the Product space because Product and IT effectively merge now. Hell, anyone in the business can now go write a Product spec and generate 80% of the code you will need without having to know jack **** about coding. All you have to know how to do is how to best talk to the AI prompt.

Seriously, if you aren't leaning into this you are already extinct.

I am wildly skeptical of the level of productivity you are claiming. FAANG would be shedding developer jobs en masse if it was that good and we're not seeing anything close to that.

We've been working with various AI/LLM systems doing everything from anomaly detection models for equipment reliability to patent/IP screening searches. It's had some nice use cases but we're talking like 5-10% net productivity improvements (when you back out the training costs, vCPU cost, etc) - not 80%.

I stare at PLTR charts all day to calculate how soon I can quit working

Best way I can describe what I've found helpful:

- Early days of computers they were useless if you didn't know how to write code.

- Once they came up with software, you didn't have to know how to write code. You just had to know how to use the software which was 100x more intuitive than writing code. But you still had to be know how to use the software.

- The AI stuff is basically doing the same thing again. Now you don't need to know how to use the software, you just have to know how to talk to the AI which is 100x easier than learning how to use the software.


For now you still have to be the intermediary between the AI and the software. The AI still won't do it FOR you (yet), but it will tell you what to do.


Example: I recently started using Notion. I knew almost nothing about Notion when I started but any time I would run into any trouble I would just ask Perplexity or ChatGPT what I wanted to do, and it would tell me how.

I got the brilliant idea that I wanted a dashboard that would include a summary of my upcoming schedule. In order to make that work, I needed to sync my Outlook Calendar to a database in Notion. Any time an event was created, updated, or deleted in Outlook it would update this database.

Turns out the best way to do this was via Microsoft Power Automate. It involved using HTTP actions in Power Automate to interact with Notion's API. It used post and patch requests to create and update the data, and JSON formatting to structure the data for Notion. End result was that my Notion database stays perfectly synced with my Outlook.


So I don't know what most of that stuff in the preceding paragraph even means. I don't know jack about Notion or how to use Power Automate. I'd heard people talk about APIs and I still don't know what it is, but apparently I successfully connected to one? And Power Automate isn't like Zapier or IFTTT. I don't know wtf most of the fields were even asking me for. I don't know what JSON is. The list goes on.

It would tell me what to do step by step. It would give me code and tell me to copy and paste it into certain fields. If I ever got stuck I would literally take a screenshot, paste it into my ChatGPT conversation, and say "what do I do next?" And it would tell me.


Whatever tools you have, you might be in this situation. Don't know how to do something in your ERP? Excel? Mail client? Used to be that you would have to Google how to do it and learn. Now you just have to tell an AI tool what you want to do, and a lot of the time it'll just tell you what to do.

pocketrockets06 said:

---

I am wildly skeptical of the level of productivity you are claiming. FAANG would be shedding developer jobs en masse if it was that good and we're not seeing anything close to that.

We've been working with various AI/LLM systems doing everything from anomaly detection models for equipment reliability to patent/IP screening searches. It's had some nice use cases but we're talking like 5-10% net productivity improvements (when you back out the training costs, vCPU cost, etc) - not 80%.

---

This is partially a function of organizational maturity. We are a startup with almost zero tech debt. We largely don't have to worry about supporting or changing out legacy tech (we have a little) and are just building.

So, to be fair, we don't have the friction or bureaucracy most organizations have. We can be as nimble as we want to be.

We have something like this that's set up to access all of our technical documentation. A couple ways to do it:

If you use Microsoft, you can use Copilot to point to directories within your SharePoint to use that as the knowledge base it pulls from. And then you just have a Teams chat where you can ask it questions and it uses the knowledge base to answer.

I don't use the Google suite but I assume they have something similar that pulls from Google Drive.

You can also do the same thing with a custom GPT in ChatGPT. Example here is before the latest Dallas election I loaded the text of all the different ballot propositions into a Custom GPT and then I talked to the GPT to ask about them to help me make my decision.

Take that above example and swap out ballot propositions for your employee HR handbook and now when someone wants to know about how much vacation they accrue, who to talk to about the 401(k), or the service animal policy they can ask the bot instead of knocking on your door.

This is a pretty common "first thing" people do. It's pretty easy and low risk, but a good exercise in showing you what's possible.


Hopefully this addresses what you were asking about.