I thought it might be helpful for us to share ways we are using AI to save time and increase productivity without having AI do our writing.
I’ll get us started. This morning alone, AI saved me an estimated four to five hours by completing two tasks:
Converting an accreditation template. I had AI take a PDF template for an accreditation reaction report and convert it to a Google Doc so I could share it with the appropriate staff. The template included fill-in forms that needed to carry over properly. AI completed the conversion in a few minutes. I did not need to ask my EA to retype it, and I avoided using a PDF app (which usually requires a subscription) or an online service (where proprietary information could be at risk). Time and money were saved, and my EA can focus on more important projects.
Extracting recommendations from an 80-page report. Late last week, I received an accreditation report with many chapters. Each chapter contained a summary of the review team’s findings, commendations, and recommendations. I needed all recommendations in a separate document so they could be reviewed, accepted or rejected, and rolled into the three-year strategic plan I am developing. I uploaded the PDF to AI and instructed it to extract the recommendations and create a summary document formatted for Google Docs. The document was to include headings for each chapter with corresponding recommendations as subheadings and bullet points. AI completed this in about six minutes with perfect formatting. This would have been an intensely manual process involving repeated copying and pasting.
Document conversion, especially the gnarly edge cases. As I’ve mentioned previously, I’ve used both Claude and Gemini to convert and/or clean up all kinds of unruly documents while I attend to other things. Example: I purchased an ebook that was formatted to mimic the printed book’s pages exactly. (I have no idea why; it’s not the kind of text where looking just like the hardback’s printed pages matters in any way at all.) Converting that kind of document into a re-sizable, re-flowable ePub is maddening: it’s all in-line html designed to place each letter in precisely one spot and one spot only on the screen. But I need to be able to read this book on my iPad mini or my phone! What to do? I threw the task at Claude, and it gave me both a markdown file and a properly structured ePub document with 156 images embedded in the proper places in the text.
Getting detailed instructions on how to do very specific things or getting a better understanding about how something works so I can use it more effectively. It’s almost always more straightforward than parsing manuals (if there even is one, which is rarer and rarer these days), help pages, forums, or YouTube videos. Lately, Gemini has been warning me about what NOT to do as well. Example: I asked to recommend some apps to quickly and easily split videos into AI digestible chunks so I could upload them without exceeding the maximum file size, and, at the end of the list of recommended apps and how to use them, it added this, unprompted, “Please do not use a file splitter that cuts strictly by file size (like ‘Split into 2GB chunks exactly’). This often cuts the file in the middle of a data frame, corrupting the video file header. Always split by time/duration (e.g.,‘10 minutes’) to ensure every file is a valid, playable video.”
Parsing web pages into tables or lists. I find this especially useful when someone embeds a longish list of books to read or exhibits to see or whatever into several paragraphs of appreciative or explanatory prose. It’s nice to have the analysis, but sometimes you just need a checklist, and an LLM will cheerfully churn one out.
Converting text to speech so I can listen to something I need to read while I’m doing chores. A lot of apps now have built-in TTS capability, and some have even paid up for the highest quality voices.
Converting speech to text so I can take proper notes without having to re-wind several times to make sure I’ve really captured something accurately.
Here’s the one I couldn’t live without: the AI masking tools in Lightroom and Photoshop. I can spend less time painstakingly painting complicated masks on complicated objects, and more time doing creative things with my images, not to mention more time actually taking photos.
One of my frequent uses is pasting a large batch of transactions (from a pdf usually) and having AI organize into table format or csv file I can then manipulate or import into accounting software. Saves A LOT of data entry.
Creating masks in photo/graphics programs, e.g. to emphasise the subject - it’s gone from a multi-hour to couple of minutes job.
Like most things tech, though, it’s more about how you use the time than giving you back loads of time. Because it’s now much easier, you do it more often and hold yourself to a higher standard.
But! It was surprisingly helpful planning a vacation using the prompt “provide an x day itinerary for a photography focused trip in y location” often lead to a schedule that was physically impossible to achieve, but definitely cut down the amount of research into the top touristy things to do along the way.
Uploaded my sermon outlines on 2 Corinthians, totaling 180 pages. Had Claude generate sample quizzes, syllabus, and summary for a grad class I am teaching. Saved me a lot of time. Did not use anything completely but was enough of a head start. I think Claude worked on it for a total of 45 minutes. Saved me hours.
By coming up with advanced code snippets that I can use to further test and expand upon and write tutorials on for WordPress community.
By giving me context-aware suggestions of what I am going to type next. All I do is press Tab to accept the suggestions or just keep typing. https://cotypist.app/
By giving summaries of what the current webpage is about via Comet browser.
By generating an excel sheet with detailed daily plan to complete a few Chessable courses that I bought.
I recently moved my own e-mail and web domain from one provider to another and from one country to another (Austria to Germany) I had no idea where to start and how to minimize the downtime for my e-mail domain.
The AI (Copilot) came up with a step by step plan that was easy to follow and it actually worked.
I moved everything and was up and running in less than 20 minutes without loosing anything.
My three biggest use cases, one of which saves me a couple hours each week:
Meeting and phone call summaries - after a call or a meeting, I hit a keystroke, and instead of spending 15 minutes typing up my notes, I spent two minutes cleaning up and augmenting the AI summary. This might save me a couple hours a week. I use Alter for this. Two caveats: I lose a few seconds of that gained time at the start of the call/meeting, explaining what I am doing and getting consent from the other person (I have never had anyone say no so far). Also, depending on your situation, you may want to make sure you’re using a local or private model.
Dictation - I bounce between Superwhisper and Spokenly. When I started, I went crazy with different modes for different use cases, such as one dedicated to cleaning/drafting e-mail, but mostly now use three:
one with no AI post-processing. This is nearly instant;
one that does minor cleanup
Note summaries - another huge time saver. On those rare occasions where I haven’t made a meeting summary (above), I can dictate top to bottom from my notes, without any worry about organizing what I’m saying, and get an organized summary back.
Research - I do quite a bit of research, either in one of the main providers’ apps, or using Alter. I might even do the same research across multiple providers. For example, I recently researched new skis. I told it about my ability level, how often I ski, what skis I already have, where I ski, etc., and asked for recommendations to fill out my lineup. It was interesting to see there was significant overlap in the recommendations among providers, and it matched my own research.
(I’m also using Cotypist, which @sridhar mentioned above. It’s pretty impressive)
I feel this. I use AI a lot and mostly to scaffold psychological readiness for a task, which is helpful but not time saving. But, there are two uses that come to mind that actually save time:
Expense Reporting
I have/hate to do expense reporting. The software we use (Concur) is horrible. I’ve crafted a prompt in ChatGPT that allows me to (using WisprFlow) describe the expense and it will give me an accurate category, business purpose, and justification comment. This saves a couple hours a month.
Script Writing
We have an online professional development course that is very scripted and professionally produced. This is not a skill I have nor want. My colleague and I had to recently write several scripts so we recorded ourselves having informal conversations about the topic, then using prior scripts in the course crafted by others, we asked Claude to write scripts for us based on those as a model. Then we edited the scripts to sound more like us. This task is rare, but it likely saved us 10-20 hours of work that required building a skill neither of us wanted to build.
I think it’s interesting that these two timesaving uses require sophisticated prompts, which is something I rarely do (I’m more chatty in my use than prompting).
Using Claude’s Chrome Browser plugin, I had it create a JSON payload of all the books I’ve purchased this year. For each book, it collected a fair amount of metadata. This simple task saved me a good few hours. And it’s in support of a larger project involving Tinderbox.
DxO PhotoLab is the best for denoising. Some say Topaz can match it, but in my experience, DxO is far more consistent. Not to mention it’s part of a full editing workflow and not a separate tool.
However, for the occasional upscale (usually less than 2x) and fixing just-missed focus, or slight motion blur, Topaz Photo does a decent job.
It sounds like you have this figured out from watching the demo, but just in case…
As far as bouncing between Superwhisper and Spokenly,I have a Superwhisper license. Spokenly sells licenses, too, but can also supports a “bring your own API key” option as part of its free option. Alter, mentioned in my original post, gives you an OpenAI-compatible API key to use in other apps, which gives access to many, many models. So, I can get Spokenly for free, and wanted to see how well it works, and if next year it could replace Superwhisper (I do wish the developer had a lower tier for use cases like mine, so I could support it). I’ve found Superwhisper to be a little faster, but that might be because of me using my own keys with Spokenly? Not sure. So I will probably stick with Superwhisper.
My “no processing” mode uses one of the AI transcription models that just transcribes, and doesn’t alter the text beyond what it automatically decides (deciding on its own where punctuation goes). My other one leaves me in control of the punctuation, as it just replaces spoken punctuation (comma, period, new paragraph) with the actual punctuation.
I’ve been using Claude Opus 4.5 via Poe for revisions to my Obsidian workflows. It is a mix of chatting to get feedback and suggestions as well as generating lists based on the results of the chat.
Claude Sonnet 4.5 has been great for code generation for SAS, Obsidian plugins, and Python. Still needs work and sometimes iterative feedback to get it to refine or correct code, but way faster to get an initial version of the code than coding it myself.
Expense Reporting
Script Writing