In this post, I share two ways AI has saved me time and unlocked new capabilities since January.
Both follow a pattern worth knowing: any task that is Repetitive, Data-heavy, and has Low variability is a prime candidate for AI automation. Annie Tsai calls this the RDL framework (skip to 53s), and once you’ve seen it, you’ll start spotting automation opportunities everywhere.
1. Automated a manual copy/paste chore for publishing blogposts each week
I volunteer at a non-profit, and as part of my duties, I create three blogposts on the organization’s WordPress site every Friday. The contents of these blogposts are pre-written by the non-profit, and they’re all stored in a templated Google spreadsheet.
For years, I dutifully copy/pasted bits of data from the spreadsheet into the WordPress Admin pages of the non-profit’s website. This task took roughly half an hour each week, due in part to finicky formatting requirements. It was repetitive (same steps every Friday), data-heavy (pulling from structured spreadsheet fields), and low-variability (the format never changed). Classic RDL.
So why did I keep doing it manually for so long?
I took on that responsibility back in 2007, when the org was new to the web, WordPress itself was far less mature, and just getting to publish posts regularly while keeping the website up was considered a win. Blogpost content was sent to me via email, and it wouldn’t be until over a decade later that we added some semblance of rigor to the workflow through Google spreadsheets. By then, I’d fallen into the mindless ritual of opening my data source, copying each field, and formatting the post week after week. And it felt like real work: it occupied time and produced a visible result.
But one Friday last month, after realizing that I’d mis-scheduled a post for the wrong day again, I stopped and asked myself: What if I just described this entire workflow to Claude Code and asked for help?
Less than four hours later, I had a working publication pipeline built from three scripts:
- Blogpost creation: the first script reads the Google Spreadsheet and assembles a Markdown file containing the full blog post, pulling data from multiple cells across several tabs into a cohesive whole. I review and finalize it on my local drive. I use a dedicated Google account with limited access to give the script access to the spreadsheet, and the script uses OAuth through that account. Claude walked me through the setup.
- Blogpost publication – the second script takes the Markdown file and creates a new post on WordPress via the API, with all required formatting applied automatically. Post metadata (publication date, excerpt, featured image, categories, and tags) are populated by the script. The script uses a “Staff” account on the WordPress site, with an Application Password that can create posts but can’t edit user profiles. Claude guided me through that setup too.
- Google Spreadsheet update – once the WordPress post is created and scheduled, a third script updates status fields in the Google Spreadsheet and saves the published post URLs for easy cross-reference.
A task that used to take half an hour each week now takes only a few minutes. More importantly, the mindless and error-prone copy/paste work is gone. I now get to focus my time on the true human work: copyediting and finalizing the content.
2. Created a sustainable way to capture and save useful ideas from YouTube videos as Snippets
If you watch a lot of YouTube videos but struggle to remember exactly where you heard a particularly useful nugget of wisdom, then you understand the itch I wanted to scratch.
This problem also fits the RDL pattern: processing video transcripts is repetitive, the source material is data-heavy, and the shownotes format I use is consistent. Perfect for automation.
With Claude Code’s help, I built a semi-automated process that:
- downloads transcripts of YouTube videos;
- generates shownotes with promising takeaways from each video transcript;
- creates a “Snippet” for any takeaway I deem worth saving;
- suggests categories and tags for each Snippet based on its contents, and
- publishes the Snippet (after I’ve reviewed, edited, and finalized it) to a dedicated WordPress site.
What’s a Snippet? Each one is a short, standalone article with a concise takeaway that anyone can read in a few minutes. Here’s what’s included in each:
- a title that expresses the key idea fully but succinctly
- a direct quote from the original video that captures the spirit of the Snippet
- a timestamped link that, in the case of YouTube videos, takes the reader to the exact moment in the video where the quote is from
- a short explainer that adds nuance, depth, or examples that illustrate the key idea
For example, today’s Snippet is entitled, “Use AI to automate away the annoying things first.“ It opens with a quote from Annie Tsai explaining the RDL framework, links to the video where she says it, then unpacks why starting with annoying tasks (rather than ambitious ones) builds momentum and trust in AI tools.
Most takeaways from shownotes are discarded. The takeaways that become Snippets are handpicked, reviewed, and edited by me, with a focus on clarity, accuracy, and whether the idea is actually worth saving.
The first Snippet, AI enables “compounding learning” through iterative feedback, was published on January 4. You’ll find a list of all January Snippets on a monthly recap page. For the curious, the technical details of the Snippet-creation workflow are described on the site’s About page.
Without automation support, I wouldn’t have even tried this kind of curation; dealing with video transcripts manually entails too much work! Claude Code automated the tedious assembly process, so now I can focus my time and attention on the actual curation task.
In Closing
Both examples started the same way: I realized the task was repetitive, data-heavy, and had low-variability. Then I asked Claude Code if it could help. The answer, both times, was yes.
A small note on what it means to be data-heavy: In both cases, AI could automate the workflows because the data was readily accessible and was stored in a sufficiently structured way. This structure allowed Claude and the scripts it wrote to retrieve and understand the data well enough to do the work. In the non-profit use case, we had a templated Google Spreadsheet. In the Snippets use case, the transcript downloader preserved the timestamps and stored useful video metadata.
If you’re wondering where to start with AI automation, Annie Tsai’s RDL framework is a reliable filter. Find the tasks that fit those three criteria, and you’ll find your best candidates. The annoying stuff gets fixed first, and the wins build from there.
