I spend my days at Microsoft helping AI startups and digital natives build on Azure. I talk about AI workloads, app modernization, production-grade architectures. It’s my job to know what’s possible.
But there’s a difference between knowing what’s possible and living it. So a week ago, I decided to stop just advising and start operating. I set up a personal AI agent — not a chatbot, not a copilot, but something closer to a junior employee who happens to never sleep.
Here’s what one week looked like.
Monday: The podcast problem
I help produce two podcasts. My video editor lives in Nigeria , and I’ve been shipping him 30GB 4K video files over spotty internet. It’s been a bottleneck for months.
My agent spun up an Azure VM, tested encoding configurations, and landed on a setup that compresses 30GB down to 8.3GB in about two hours. Cost per episode: roughly one dollar.
It also researched GPU alternatives and came back with an actual recommendation: CPU encoding is fine for four episodes a month. Don’t overthink it.
I didn’t ask for the cost comparison. It just did it.
Tuesday: A dashboard I’d been putting off
I run three YouTube channels (church, two podcasts). I’ve wanted a unified analytics dashboard for months but never had time to build one.
My agent built a Next.js dashboard, synced all three channels via the YouTube Data API, deployed it to Vercel, and sent me the link. 129 videos across three channels, all in one place.
Same day, I mentioned wanting to demo AI capabilities to a friend who works in the creative space. The agent built a full e-commerce site — product catalog, cart, checkout, dark mode, 15 AI-generated product images — and deployed it in 11 minutes. I didn’t write a single line of code.
Wednesday: Personal logistics
I had an international trip to plan. Flights and hotel were booked, but everything else was still floating around in my head — gear to buy, weather to check, tickets to an event, packing lists.
I handed it to my agent. Within an hour: a 27-item packing list organized by category in my task manager, product recommendations with links for everything I needed to buy, weather analysis for the destination, and price comparisons across multiple ticket resellers.
It even set up browser automation on my server so it could browse sites and verify prices itself.
Thursday: Content strategy I’d pay a consultant for
One of my podcasts has 16,000 subscribers but the views don’t match the audience size. I’ve known this for a while but haven’t had bandwidth to dig into why.
I asked my agent to look into it. What came back was genuinely surprising.
First, a competitive analysis against four comparable channels. It found that channels with similar subscriber counts were getting 5x to 20x our per-video views. It identified specific patterns in titles, thumbnails, episode length, and shorts strategy that separated the performers from the rest.
Then it ran a full-year audit of 151 videos. Graded the channel C+. Found that our top 25% of content drives 68% of total views. One content category outperforms everything else by 2.3x. And counterintuitively, long-form episodes outperform shorts.
It delivered a structured report with specific recommendations: pick one niche for 90 days, fix the titles, cap episodes at 30 minutes, lean into the cultural angle that already performs.
I would have paid a consultant real money for that analysis. My agent did it before lunch.
Friday: A YouTube summarizer
I watch a lot of long-form YouTube content. Interviews, technical deep dives, conference talks. I don’t always have 90 minutes for a three-hour episode.
So my agent built a summarizer. Drop a URL, get a structured summary with key takeaways and notable quotes. It pulls transcripts when available, falls back to audio transcription when they’re not. I tested it on a 3-hour episode and had the summary in 26 seconds.
The Siri moment
Midweek, I was showing a friend what the agent had been doing. The compressed videos, the dashboard, the logistics planning. He watched for a minute and said: “This is exactly what Siri was supposed to be.”
That stuck with me. Because he’s right. We’ve been promised intelligent assistants for over a decade. What we got was glorified timers and weather lookups. What changed isn’t some single breakthrough — it’s that these systems can now hold context, use real tools, and run without you babysitting them.
What I actually learned
I’m not writing this to show off. I’m writing it because I think most people — even people deep in AI — are underestimating how much the daily experience of working with these systems has changed.
A year ago, AI assistants were good at answering questions. Today, the ones that run continuously and have access to your actual tools can do real work. Not “draft me an email” work. Deploy-a-dashboard, research-a-market, build-a-prototype work.
I help companies figure out how to adopt AI in production. The best advice I can give after this week: stop planning your AI strategy in slide decks and start building something small that runs against your real problems. The gap between “AI demo” and “AI doing actual work” is smaller than you think, but you won’t see it until you cross it.
This week my agent processed video, analyzed content strategy, built and deployed web apps, managed my task backlog, and handled personal logistics. It ran automated checks on my GitHub repos, monitored the weather, and sent me a briefing every morning.
None of it was perfect. Some image generations needed three rounds of iteration. A background job timed out because I picked the wrong model. The competitive analysis had to be re-scoped after the first pass was too surface-level.
But those are the same kinds of problems a new hire would have in their first week. The difference is this one costs a fraction of a salary, works at 3 AM, and gets better every day.
I’m going into the weekend with a compressed podcast episode ready for my editor, a content strategy for my YouTube channel, a fully planned trip, and a new tool I’ll use every time someone sends me a long video.
Not bad for a week.