Generally, an issue you didn’t know existed meets an answer you didn’t got down to construct. That’s exactly what occurred when a routine dialog about our developer docs became an AI-powered software for our content material workforce.
Right here’s the way it all got here collectively:
How It Began
Initially of 2025, my workforce at Automattic (WordPress.com’s dad or mum firm) had taken over accountability for updates and upkeep of the WordPress.com Developer Docs.
As a part of this course of, we met with Alexa, a content material advertising supervisor for WordPress.com, to talk about processes and pointers round this handover.
Towards the tip of the assembly, Justin Tadlock, my workforce lead, requested a easy query, “Is there the rest we might help you with?” Alexa replied, “I’ve a query, however it’s unrelated to the docs. Do any of you already know if a calendar view of GitHub Points is feasible?”
The Drawback
Alexa defined that her workforce makes use of GitHub Points and Initiatives to handle content material publishing, a typical apply in software program firms.
The method usually labored nicely however was lacking one essential piece for visibility into scheduled content material and content material gaps: a calendar view. Their workforce couldn’t view these content material points in a month-to-month calendar to get a fast snapshot of what was deliberate for the month forward, because the GitHub Initiatives view solely gives a roadmap-style view—useful for creating software program however not splendid for content material planning. Nick Diego was in that assembly. You most likely know Nick from his Revolutionize Your WordPress Growth with Cursor AI video.
Since management at Automattic inspired us to discover AI instruments like Cursor, Nick and I had spent the previous few months experimenting and had some implausible successes, constructing every little thing from WordPress plugins to a desktop app for viewing a neighborhood WordPress debug log.
If you happen to’ve by no means heard of Cursor, it’s an AI-powered code editor. Forked from the favored Visible Studio Code editor, Cursor takes the thought of AI-assisted coding to the following stage by including a ChatGPT-like agent to the combination. This AI-powered chatbot interprets your written prompts into working code. This idea has grow to be so fashionable it even has its time period—vibe coding, or utilizing a big language mannequin for coding.
We had been each impressed by an Automattic City Corridor in December, the place certainly one of our engineering leads used Cursor to routinely generate a pull request for a undertaking he was engaged on.
Now, I can’t say this for positive as a result of it was throughout the digital divide of Google Meet, however I’m virtually positive Nick and I appeared instantly at one another and had the identical thought on the identical second.
“This isn’t that tough to unravel with AI.” Nick was the primary to specific this concept, so we spent the following jiffy chatting about it. We each agreed that the one difficult half could be determining the correct method to question the information we would wish to fetch the related GitHub Points from the undertaking view; constructing the precise calendar app could be fairly straightforward.
Fortuitously for us, our colleague Birgit Pauli-Haack was additionally on that decision. She talked about that, sure, GitHub Initiatives have a publicly queryable API. Nevertheless, it was not a REST API however a GraphQL API. She had explored querying GitHub Challenge knowledge with some success, however she hadn’t dived into one thing like this earlier than.
Sitting on that decision and listening to all this data, I felt like producing a calendar view for these GitHub Points could be doable and comparatively simple. I simply needed to:
- Work out the GraphQL question.
- Use Cursor to construct an internet app round it.
How onerous might or not it’s?
Perplexity to the Rescue
The very first thing I needed to do was determine the GraphQL question, and I’ll be the primary to confess that my information of GraphQL is proscribed.
I’ve written tutorials on utilizing GraphQL to construct static frontends for WordPress web sites, often known as “Headless” or “Decoupled” WordPress websites. In that course of, I used the WP GraphQL plugin that our colleague Jason Bahl developed; nonetheless, I’ve by no means used it in a manufacturing atmosphere or used it to construct an internet app.
So I puzzled if I might take a shortcut on this studying curve with AI and craft the GraphQL question with out understanding it deeply and utterly.
Then I remembered that WordPress.com had partnered with Perplexity. Perplexity is comparable in some ways to ChatGPT, however it describes itself as an AI-powered reply engine that permits you to analysis the online.
What I like about Perplexity is that when it outputs the generated AI response, it additionally contains all of the sources of knowledge it has used. If the reply you get from Perplexity isn’t fairly what you had been searching for, you possibly can open the supply hyperlinks to do additional guide analysis.I additionally wanted a method to take a look at and validate the GraphQL question that Perplexity would offer. Fortuitously, GitHub has a GraphQL API Explorer, which lets you authenticate together with your GitHub account and run legitimate GraphQL queries on any non-public or public GitHub repositories you possibly can entry.
It took me a complete of two Perplexity Threads (back-and-forth conversations with Perplexity) to determine the GraphQL question. The primary Thread was me asking it to construct primary variations of the question, working them within the explorer, after which I fine-tuned the question based mostly on the information from the explorer. After three rounds of fine-tuning, I began a brand new Thread and created a extra particular immediate, utilizing all the teachings I’d realized from the earlier Thread. Testing that question within the Explorer returned the precise knowledge set I’d want for the online app.
The Cursor Construct
The subsequent step was to start app improvement with Cursor. One of many issues that I found early on (each via Nick’s video and this Twitter thread) was the thought of utilizing Cursor to generate an directions file. The thought is to have a “dialog” with the Cursor agent and share the fundamental necessities of the appliance.
You then instruct the agent to write down out these necessities to a necessities.md markdown file. Subsequent, you ask the agent to undergo the necessities file itself and ask you any clarifying questions in regards to the necessities. As you reply the clarifying questions, you instruct the agent to replace the necessities file.
After getting the necessities file, you instruct the agent to create the directions.md file based mostly on the necessities file, which outlines the steps wanted to construct the appliance.
As a result of I already had the GraphQL question, I additionally added it to a file within the undertaking listing. Armed with my necessities.md file, directions.md file, and the file containing the question, I instructed the Cursor agent to start out constructing the app step-by-step, following the directions.
In about two hours, I had a working prototype. I returned to Perplexity a few times to assist me remedy one thing that Cursor couldn’t do by itself. I wrote no different code past working one or two terminal instructions to arrange the undertaking listing and its dependencies or working the event server to serve the app for testing.
It took me possibly one other hour to shine the app the next Monday morning (largely including colours and making GitHub Points clickable). By noon, I felt I had one thing very near what Alexa’s workforce wanted.
All that remained was to seek out an inner location to host it, which paradoxically took longer than the app construct as a result of this was so new.
By Wednesday, I shared the hyperlink with Alexa and my workforce in an inner put up.
Everybody was actually proud of the result. Alexa and some of her teammates examined it out and picked up one or two small points.
A kind of bugs was attention-grabbing; evidently Alexa noticed all of the objects within the calendar at some point earlier than they had been on account of be printed. I fed this data again to the Cursor agent, and it confirmed this was on account of some timezone-related bug, which it promptly fastened.
One week later, we had a working GitHub Initiatives Content material Calendar.
Closing Ideas
With all the AI instruments accessible at the moment, software program improvement is at an inflection level.
I felt the identical manner when Google launched its search engine and Gmail. That is most likely how individuals felt in 1886 when the first sensible trendy car was invented.
LLMs and AI brokers will certainly change how we develop software program, from dashing up how we study new issues to helping us in our precise improvement.
I don’t consider they’ll change the necessity for human builders as a result of if we don’t use them responsibly, we’ll find yourself with mountains of inefficient code. However it’s definitely doable at the moment to construct issues with a better success fee than it was. Based mostly on my expertise creating the content material calendar app, I made one other GitHub-related internet app that permits you to view any GitHub points assigned to you throughout all of your GitHub repositories.
As somebody who works throughout a number of repositories, I’ve wished a dashboard like this for years—and I used to be in a position to construct it myself!
Getting used to how these instruments work additionally has different advantages. For instance, you will have observed that this weblog put up contains hyperlinks to my colleagues’ on-line accounts or merchandise’ web sites. I added all these hyperlinks just by exporting the content material to a markdown file and asking Cursor to “search for all correct nouns within the textual content, discover the net URL for each, and add the hyperlink to the right noun.”
I wanted to spot-check the hyperlinks, however discovering and linking them myself would have taken me half an hour. I did it in minutes with Cursor.
If you happen to’re a developer, I encourage you to experiment with these instruments. Your subsequent nice product concept could also be simpler to construct than you suppose!
Subscribe to MarketingSolution.
Receive web development discounts & web design tutorials.
Now! Lets GROW Together!