Learning by Prompting AI for Answers (June 2023) to Prompting my Agent to Write & Do Tasks for Me (2026)

Since June 2023, right when ChatGPT first dropped, I’ve been prompting AI non-stop to learn anything and everything I can about the tech world. Unfortunately, I grew up in a house where computer were seen as adult toys and email address, social media accounts, and live chat were seen as unnecessary. So when I started prompting AI, I started with basically zero background other than AI output needs to be verified.

Back then, AI was pretty much a supercharged Google search engine. It could summarize huge piles of information and explain things at whatever level I asked for, in multiple tones. I used it to get my head around search engines, HTTP, HTTPS, middleware, controllers, basic server stuff, email certificates, and how API calls actually work. I never took the first answer and ran with it. Because I was starting from scratch, I kept things small and grew them gradually. I’d make it clarify terms, define everything, give me examples, show contrasts, and expand until the explanation felt solid. Then I’d go check trusted sources to make sure I was actually learning and not just absorbing AI hallucinations.

I took all that information and turned it into Canva presentations, each with its own unique styling so they felt a bit more personal. I tried to structure them as professionally as I could and stayed organized by constantly picturing how someone else would read what I made. That little mental check kept me from dumping raw notes and calling it done.

Then early 2026 happened and agents changed the game. Once I saw what they could do, I wanted a real reason to connect them to platforms so I could start creating agent guides, reports, analytics, and articles based on what the agent gathered and consumed. So I made up DuckBuddy—a concept company for social networking aimed at remote workers. Something different from Discord, Reddit, Slack, or LunchClub. Just a concrete example I could actually work with.

My very first prompt to my personal agent was to connect to my personal Jira account and create all the epics, stories, and issues needed to build DuckBuddy. I told it JavaScript React for the front end, Python for the service layer, and Postgres SQL for the database. (The agent later corrected me and explained that I’d also need like four non-SQL database types if I wanted the platform to run efficiently.) Watching it work while explaining what it was doing in real time was crazy. Shocking. Mind-blowing. Hard to even describe. Technology has come so far that it just… did it. Within minutes.

In a short time I had the whole plan sitting in my Jira account on a Kanban board. Then I asked it to do the same thing in a SCRUM board with sprints, and it knocked that out super quick too!

Once DuckBuddy’s programming architecture was laid out, that was just the starting point (which was exactly my goal). Having that foundation finally let me ask the agent to create reports explaining how long different software bugs would take my development team to fix. I had it write a QA test guide that walks an agent through reading the reported bug in Jira, analyzing the codebase to find where the bug lives, suggesting code changes for the team to review, estimating the time for the fix and the impact on the rest of the code base, plus the QA tests that need to happen after the fix hits the QA environment. And a bunch more documents, guides, and reports after that.

I’m still asking AI to teach me, but now I’m also having it write the articles and documents while it does so much more. I’ve seen the kinds of things people at real companies are using it for—writing knowledge base articles for customers and internal use (when it has access to the company's source code and software account), creating new feature release documentation and proposal standards, building new feature evaluation guides, writing swim-lane guidelines that explain when and why each department needs to be involved in big feature releases, and tons more. I’ve had my agent do all of those for DuckBuddy just so I could see what the documents actually look like and have the experience of prompting and guiding an agent to do so.

I even had it do market research and create a document that lays out the benefits, the market gap, and the potential competition risks for DuckBuddy (which are some of the realistic things that companies have to consider). With everything it produces, I get to learn more—what it included, what it left out, how it explains things, what it suggests. Having it actually do the tasks and then looking at the output gives me way more to learn from than just chatting.

That’s the shift. I went from carefully questioning a smart search tool to collaborating with systems that can plan, execute, document, and iterate at a speed that still feels a little unreal. DuckBuddy is still just a concept, but building it this way has already taught me a ton about modern software development, product work, and what agents can actually handle inside real teams. The tools keep getting better. My habit of making them clarify, verify, and organize the results stays the same. Honestly, that combination is still the best way I’ve found to keep learning. And I definitely still use Canva, Cavan pro now so I can do even more there too!