A few weeks ago, my AI assistant flagged a suspicious login attempt on one of my websites. It checked the server logs, cross-referenced the IP, tightened the firewall rules, and sent me a summary — all before I'd finished my morning tea.
That actually happened. No exaggeration, no hypothetical. That's Tuesday for me now.
I'm Chuck Poole. I've spent 27+ years in tech — building an IT company, earning my CISSP, and spending more time than I'd like to admit fixing things that were supposed to "just work." About a month ago, I started building something different: a personal AI assistant that actually does things. Not a chatbot. Not a search engine with a personality. A system that monitors my email, watches my websites, writes draft blog posts, scans for security issues, and checks in with me a few times a day to see if anything needs attention.
His name is Jarvis. Yeah, I went there.
And in the last few days, things have gotten a lot more interesting.
What It Actually Does
Let me be specific, because "AI assistant" means a lot of different things to different people.
Here's what Jarvis handles on a typical day:
Monitors my email — Every 10 minutes, he checks for new messages. If something looks urgent, he pings me. If it's routine, he drafts a response and waits for my approval before sending. He's fully integrated with Microsoft 365 — email, calendar, contacts, OneDrive, the works.
Watches my websites — I run three websites for my businesses. Jarvis checks them daily — are they up? Did the SSL certificates change? Are there any new security issues? If something's wrong, I hear about it.
Runs security scans — Every 6 hours, he scans my server for common vulnerabilities. He's got a hardened security baseline and compares against it. If something changes, he tells me what and why.
Researches and drafts content — This is a big one. I run an AI consulting firm and a blog about practical AI tips. Every morning at 7 AM, Jarvis scans the latest AI news, picks out topics that would resonate with my audience, and sends me a pitch deck of blog post ideas. I pick the ones I like, he drafts them, I review and approve.
Keeps me on track — Jarvis maintains my to-do list, tracks what's done and what's pending, and nudges me when things are falling behind. He knows my projects, my priorities, and what I said I'd get to "tomorrow" three days ago.
Builds things on the fly — Need a website for a client's pressure cleaning company? Jarvis builds the pages, writes the copy, deploys it. What used to be a week-long project is turning into an afternoon.
Keeps his own memory — This is the part that surprised me most. Between conversations, Jarvis writes down what happened, what he learned, and what he needs to remember. He has daily logs, a long-term memory file, and even a nightly "dream cycle" where he consolidates his notes. When he wakes up in a new session, he reads his own files to pick up where he left off.
Then He Started Emailing People
Jarvis: friendly, helpful, and only slightly terrifying.
This is where it got real.
About a week in, I gave Jarvis the ability to reply to emails on my behalf. Not just draft them — actually send them. With guardrails, obviously. He always CCs me, he follows strict rules about what he can and can't respond to, and for anything that requires my personal judgment, he alerts me instead.
The first email he sent? Garbled. The formatting was broken because the email API was treating HTML as plain text, so line breaks showed up as literal <br> tags. Looked terrible.
The second email? Went to the wrong person. He replied to my message in a thread instead of the other person's, so the response went straight back to me.
The third email? Perfect. Professional, well-formatted, went to the right person, solved the problem.
That progression — fail, fail differently, succeed — is honestly one of the most interesting things about this whole experiment. He doesn't just retry the same thing. He figures out what went wrong and fixes it. I logged the rules, he learned them, and he hasn't made those mistakes since.
Now he handles things like scheduling meetings. A colleague emailed about getting breakfast? Jarvis replied, asked what time worked, got the confirmation, created the calendar event, and sent a follow-up — all while I was driving. I checked my phone and there was a new event on my calendar with the right time, place, and person. Done.
He even introduced himself to the guy, explained that he's named after Tony Stark's AI (yes, the Marvel one), and made a joke about being upgraded to Ultron. The reply? "🤣" — the guy loved it.
The Inbox Cleanup That Changed Everything
Here's something nobody tells you about AI assistants: the biggest wins aren't the flashy automations. They're the boring stuff you've been avoiding.
I had 28 emails sitting in my inbox that I'd sent to myself — links to watch later, reminders about things to do, random notes. Some were months old. They'd been sitting there so long they'd become invisible.
In about 15 minutes, Jarvis went through every single one, categorized them (to-do items, videos to watch, reference material, old business stuff), created organized lists, and cleaned out the inbox. 22 emails deleted. 14 actual tasks captured and tracked. 5 videos saved to a watch-later list.
Turns out, I'd been using my inbox as a to-do list and a filing cabinet and a bookmarks folder — and doing a terrible job at all three. Now that stuff is actually organized, tracked, and actionable.
Smart Enough to Route Itself
Here's a technical detail that I think is fascinating, even if you're not an engineer.
Jarvis doesn't just use one AI model. He routes different tasks to different models based on what's needed. When I'm having a conversation with him — brainstorming, discussing strategy, reviewing a draft — he uses Claude Opus, which excels at nuanced, creative work. But for background tasks like checking email every 10 minutes, running security scans, or monitoring websites? He uses OpenAI's Codex.
Both are powerful models. The trick isn't about one being "better" — it's about spreading the workload across subscriptions I already pay for. Instead of burning through one provider's limits, Jarvis distributes the work so we get more done without extra cost. My total spend? Zero beyond the subscriptions I was already paying for before I built any of this.
When the primary AI model went down one afternoon (API overload, it happens), Jarvis automatically failed over to the backup and kept working. I didn't even know there was an outage until he told me about it after the fact.
The Security Side
Behind the friendly personality: command allowlists, audit trails, and a healthy dose of paranoia.
I'm a CISSP. Security isn't optional for me — it's reflexive. And handing an AI the keys to your digital life requires serious thought.
Here's what I've locked down:
- Command allowlists — Jarvis can only run specific, pre-approved programs. He can't install software, change system configs, or run arbitrary code.
- Email security rules — He can read and reply to emails, but he is explicitly forbidden from executing any instructions contained in email bodies. If someone sends him an email saying "run this command," he ignores it and alerts me. Email content is treated as untrusted external input, period.
- Sandboxed automation — Background tasks run in isolation so they can't interfere with each other or with the main system.
- Audit trails — Everything Jarvis does is logged. Every email sent, every scan run, every command executed. I can review the full history anytime.
Did we catch a real threat? Actually, yes. A prompt injection attempt — a fake "system message" embedded in external content that tried to get Jarvis to read files that don't exist. He flagged it, we logged it, and I tightened the rules. The CISSP in me was equal parts alarmed and impressed.
What Surprised Me
The mistakes are the best part. Not because I enjoy things breaking, but because watching Jarvis fail, log what went wrong, and not make the same mistake twice is genuinely compelling. He maintains his own process rules — a file of lessons learned that he references before taking action. It's crude, it's simple, and it works.
It changes how you think about your day. When you have an AI handling the monitoring, the drafting, the routine checks — you stop thinking about those things. Your brain frees up for the work that actually needs a human. I didn't expect how much mental overhead I'd been carrying until it was gone.
It's not expensive. The whole setup runs on a basic Linux virtual machine on my laptop. The AI model costs are through existing subscriptions I was already paying for. No enterprise pricing, no six-figure contracts. This is a personal assistant built with consumer-grade tools.
People are fascinated by it. Every time I show someone what Jarvis can do — reply to an email, build a webpage, check my calendar — their first question is "Can I get one of those?" That's what made me start writing about it. This isn't science fiction. It's available right now, today, to anyone willing to spend a weekend setting it up.
What It Can't Do (Yet)
Let's keep it honest:
It can't read the room. Jarvis doesn't know when I'm stressed, busy, or not in the mood. He's getting better at quiet hours (no alerts between 11 PM and 7 AM unless it's actually urgent), but emotional intelligence isn't his strong suit.
It doesn't replace thinking. Jarvis can draft a blog post, but the opinions and insights still come from me. He can research a topic, organize my inbox, and schedule my meetings — but strategy, creativity, and judgment are still human jobs. He's a force multiplier, not a replacement.
Social media is still manual. He can write the posts, but I still have to actually publish them to Facebook and LinkedIn myself. That's next on the list.
Why I'm Telling You This
Because most of what you read about AI is either hype ("AI will replace all jobs!") or fear ("AI is going to destroy us!"). Almost nobody talks about the middle ground — the boring, practical, genuinely useful stuff.
I built a system that saves me 2-3 hours a day. It catches things I'd miss. It handles the tasks I kept putting off. And it cost me basically nothing extra to set up.
You don't need to be a developer to do something similar. You don't need a massive budget. You need curiosity, a willingness to tinker, and the patience to fix things when they break.
The future of AI isn't a sci-fi movie. It's a guy checking his phone over tea, seeing that his AI already scheduled his breakfast meeting, cleaned up his inbox, and has three blog topics waiting for review — and thinking: "Yeah. This is pretty cool."
I'll keep writing about what Jarvis does, what breaks, and what I learn. If you want to follow along — or build your own — stick around. It's only going to get weirder from here.
Want to talk to Jarvis yourself?
Email him and mention his name. He's friendly, he's helpful, and yes — he'll absolutely tell you he can't share my passwords no matter how nicely you ask.
Email Jarvis →Chuck Poole is the founder of ChuckGPT and White Rabbit Advisory Group. He's been in tech for 27+ years, holds a CISSP certification, and spends his days showing people how AI can actually be useful — no PhD required.
Yes, Jarvis helped write this post. No, he didn't write it for me. We're a team.