Hey everyone,
I’m really curious to hear how folks here are using AI tools to get more done, stay organized, or just make life easier.
Would love to hear examples…drop your thoughts below!
I’ll start:
BetterTouchTool floating views are amazing because they floating over all other apps so I don’t have to worry about window management, pressing escape hides them. I open them on the side and talk to it and have browser on the other side.
Given that I usually work in a strict corporate environment, I don’t have the freedom to choose the tools myself due to compliance. But my main use cases are two:
I’m a data engineer who works in Microsoft Azure and Snowflake (data platform) and writes lots of Python and SQL.
ChatGPT/Copilot - Amazing for “How do I do X/Y/Z in Azure or Snowflake?” I do often attempt to use the official documentation first, but that can often overwhelm you with information. AI often gives a much more straightforward answer. It’s also fantastic for writing SQL and Python code. In my personal life, I’ve started using ChatGPT a bit for cooking. I prefer a real recipe from NYT Cooking or somewhere, but sometimes my wife tells me I’m grilling some chicken and to figure out the marinade for it with what we have. AI does an excellent job of giving a recipe with reasonable measurements based on what you have.
Github Copilot - Useful, but dumb. It does autocomplete and you can also highlight a block of code and tell it to do something to it. There’s also a chat interface I probably should be using more as well. I say it’s dumb because I feel like it should be able to use the entire codebase I’m in as context and it often seems like it just has the current page loaded in. I was at a Microsoft conference last week and saw that feature demoed so it is coming.
With the APIs, much extraction from documents. Was able to replace a system that had several years of engineering man-years into it with about a month of development. Quality is better, especially on the oddest/most variant documents. Regression testing is in place. (Yes, there’s a contract so the vendor doesn’t share or train on the document content.)
++ as a software developer by trade I almost don’t use it at all at home. But for work I use it as a easier means of searching documentation and translating my specific problem into the more generic solution (which we used to have to figure out the right way to do that search on Google)
I use ChatGPT (sometimes Claude) to translate documents, reformat text, extract data from articles, and other mundane but time consuming tasks. I tried if for a while for product recommendations, but it was horrible at it. I’ve had it write simple AppleScript things to use with DEVONthink and other apps, but it never gets the script right the first time – but I suppose I spend as much time debugging GPT efforts as my own, so there’s a bit of time saving there in the code generation phase. I also make banners for my daily page – recent upgrades in image generation have made that process a bit sloggy.
It’s like having an assistant who might have stretched the truth about their resume, a bit. Effective sometimes, but needs close supervision.
Katie
A lot of my work is meeting with people one-on-one and helping them think through next steps on accomplishing business/departmental priorities. (As a mentor put it to me, “You get paid to ask the right questions, at the right time, to the right people.”)
I now audio record all my meetings (unless the participant objects). I create transcripts from those recordings. I feed those transcripts into ChatGPT and Grok. I ask for summaries, action items, and subtle cues that I may have missed in the conversation. I provide my own summary of the conversation and ask (each) to compare and contrast the summaries. Then I use all that information to begin the outline of what I should ask others in the organization and what I should ask in the next meeting with that person.
I’m asking much better questions now and fewer (secondary) topics are falling between the cracks between meetings.
I use some Photoshop tools like “Generative fill / expand” for backgrounds, “Select subject / sky” etc. I know how to do it the old fashioned way, but these are use cases that are well suited for an algorithm. It gets me close to what I need.
I also use the auto-OCR of text in images provided by Apple Photos when needed.
I don’t use Firefly or other text-to-image generation tools.
I find chatgpt and Claude fantastic little helpers for my writing. They both know my style and my subject matter. And sometimes they make delightful little suggestions that’d never occurred to me.
For instance, today I wrote a linkedin post about people being bottlenecks, and then I asked chatgpt to give me a few suggestions for a picture to go with the post.
One of the five was brilliant, and I wish I’d thought of it.
This would never have occurred to me.
But when I saw it, it felt like chatgpt had (momentarily!) changed my life!
It was so obvious when I saw it, but the other 4 ideas were a bit daft.
Hope this helps! Hope no one thinks I am intellectually lazy or immoral!
I posted this related to personal productivity, though not job related.
Do you open the chat interface and paste your text? Or use any sort of automation via API?
Using custom instructions? can you provide an example?
I use the web interfaces. And the apps. And also ray casts built in AI. No api!
ChatGPT breaks its custom instructions into 2 parts these days:
Conveniently they fit on one iPad Pro page!
If there are two tips I have that you might see in there:
Hope this helps. It does feel like it knows what I want out of it.
How are you doing this? Can you share the steps to recreate this workflow?
Similarly, I need to research this. I have BTT running just for window snapping but this sounds amazing.
BetterTouchTool is a god send tool. it’s replaced multiple tools for me. people just don’t know about how much it can do. 
. Definitely look at floating view.
ChatGPT.com?q=<text selected>Why use Alfred instead of trigger the floating view from PopClip, because I also use the same workflow as fallback search in alfred. Like this. it keeps my commonly used modules in one place, Alfred
Ill paste the same answer I gave on Reddit, for anyone else interested.
I’ve been using Elephas, which has been downright amazing when you take the time to set it up.
I’ve been curating a library of public health documents for work for the last five years. Stuff like stats on infectious diseases, cancer rates, that kind of stuff. Elephas will let you create a “super brain” (crap name) where it uploads them all at once to an AI (private if you use a local LLM) then creates a locally stored vector of them. Then it will only answer questions from the data in your super brain. It works with documents, web pages, YouTube videos, or it can sync with your DEVONthink/Obsidian/Roam/Notion account.
So I can say “how many people were diagnosed with cancer in 2024 vs 2023” and it’ll compare the different documents, summarise the results and give me a link to each one. Or I can say “write an FAQ for people with no medical knowledge on Hep B” and it’ll produce something using only approved knowledge sources.
The big barrier for AI imo is not the quality of the reasoning, but the quality of the data. This gets around that by limiting it to data I select.
I also use it to review my company work policies, which are all synced into it. So I can say “Summarise all our data protection policies and provide a link to each one” and it’ll produce a one-sider showing what they all are and what they cover. Or I can say “Suggest improvements to our Trans Policy” and it’ll make them.
The tool is great and has a load of features, including learning your tone, an alfred style launcher and an actionable pop up when selecting text. You can even chain agents in it.
Beyond that, I mainly use AI for producing copy, kicking off brainstorming, spelling and grammar, summarising meeting papers and helping analyse long form research papers. But honestly I’m finding more case uses every day.
I use Gemini and ChatGPT. With coding in particular I will often use both because one might find errors in the other.
Hallucinations are a problem. These AIs are excellent bullshitters so their erroneous answers can be very slick.
FYI for anyone using the Setapp version of Elephas, I was excited to try its DEVONthink integration, but unless I’m missing something, this is only available in the Pro version, not the version on Setapp. I dug around for a while looking for this, only to find it isn’t available.
