AI

Problems stuffing AI genie back in the lamp

It's no secret that USA has neither the electricity nor the skilled labor to deliver data centers needed for the AI revolution. Yet, frontier model providers and public companies alike keep insisting that America must win the AI race or be left behind! and they are spending like there is no tomorrow. Most rational people understand that this bubble won't hold up and they think this is where AI is actually going:

AI bot begging for tokens

Problems (beyond 5th grade math) aren't just around lack of electricity, GPUs, staff, ram shortages or even investor enthusiasm. Developers are complaining that they are getting less tokens for the buck and that models are thinking less. Major providers like xAI/Twitter's Colossus DC (Grok) run on natural gas (methane) burning turbines. Most of the world is on a path to economic recession and $ from the Middle East is not arriving like it used to habibi.

All these headwinds are giving many AI naysayers reasons to remain pessimistic. I hope you're not among them.

Here is the issue: AI is not the dot-com bubble pets.com sock puppet where removing one or two players will going to collapse the entire value chain. Frontier models like GPT/Opus/Grok require a ton of resources but provide massive context windows and fast token output. Billions of dollars and entire supply chains are tied to the success of these AI models. Are they at risk? Sure! Problem is, these are not the only AI models out there and those are also not likely to put you out of work (unless you're a developer).

As of mid-April 2026 I can run a very respectable Gemma 4 model from Google on my iPhone 17 Pro.

Even Raspberry Pi can run Qwen, Deepseek R1, llama, etc.

These selfhosted and quantized models are the genie that won't fit back in that magic lamp. If you can have AI on devices that cost less than $100 no amount of Middle East war or RAM and electricity shortages is going to hold it back.

When the same PC you used to give to your employees can outperform and outproduce your employees... you're going to have far less employees. Billionaires don't disagree:

Elon Musk on UBI

If you're worried about OpenAI, Grok or Claude putting you out of work and have hopes that economic collapse will put them out of business - respectfully - you need to broaden your horizons. The overall automation, orchestration, and awareness at the edge or in your pocket is what will ultimately decimate low-skill labor.

It's not a question of if but when. Consider this an encouragement to get out of roles that can be obsoleted by last years smartphone tech.

Knowledge gap vs. Outcome gap

I've been annoying my friends with AI for years but one of the more bewildering things has been trying to convince my technical friends to try AI bots and automation. One of the most common arguments I get back is:

"I just lack foundational skills/knowledge in that area"

Ugh.

Stupidity is something I used to excuse (somewhat) before I was a father of teenagers - these days I have near zero tollerance for it. I get it - not everyone has in-depth expertise with Linux, docker, networking, web services, or a dozen programming languages and technologies that are part of even basic apps these days. The point of AI is that you don't need it to learn and get ahead.

Let me take you back to 8 year old Vlad. My first programming class and I got to learn the magic of the POKE command in Basic in my first computer programming class:

If you remember this it's time to schedule your colonoscopy.

The domain expertise back then involved turning on the TV, going to channel 3 or 4 back and forth and jiggling the cable until the RF signal locked in and you were lucky not to have rolling bars scroll up. Point being, technology was not that rock solid and you had a lot of trial and error involved in getting things to work. Being able to change the color of the screen and the border - that was magic!

Meanwhile in 2026...

We've advanced a little. And not just in a sense that you no longer need to remember memory-mapped hardware registers or color codes.

AI IDE, agents, MCP servers, automation workflows and everything I pretty much spent my entire career learning like a caveman is extensively documented and on a skill gap of someone experiencing autocomplete for the first time.

You don't need to know every little thing about Linux, Codex or Cloud Code can setup the backend for you. You don't need to know every little thing about security and networking, pentests and development best practices are automated as well.

With every technological advancement in tech we've gone from needing to know less and less about the guts and focus more on the outcomes. I've gone from memorizing memory locations, to memorizing which files handled network configuration, to barely being able to remember anything. AI came in just the right time.

Point being, don't let your percieved gap in knowledge of all the components hold you back from building and growing. At no point in tech history did we have the ability to have the tech guide us, teach us, troubleshoot for us, work around problems for us, or help us solve problems. I grew up hearing "you won't always have a calculator on you" and man were they wrong.

Do your self a favor and mute ignorant people who ramble about AI killing the environment or fantasize about a future without it because "it's going to become proprietary or expensive". Not only because none of that is even remotely true (after all, quantized models can run on laptops with mediocre specs and even on $10 esp32 SoC powered by a watch chip)

Sign up for a free VPS. Roll out openclaw or a bot of your choice. Start learning, start building, start growing. The gap in tech these days is less about knowledge and more about perfecting the outcome. We're in the golden age of free computing - as every AI platform and cloud service provider tries to compete for control of the future they are giving away an insane amount of resources that can help you.

Don't get obsolted by AI, grow with it instead. It's a choice you get to make.

P.S. Most of you dinosaurs reading this blog didn't know jack about DNS, Exchange, IIS, ISA and needed Jeff to write encyclopedia email responses just to guide you through AD. Yet you limped through it just fine with wizards, in 2026 that's ChatGPT and Codex.

How I use openclawd

I have been on a journey with my clawdbot -> moltbot -> openclaw bot and at this point I have dozens of agents running on multiple bots and multiple servers. If you want to spin up one of your own start here.

I am going to share some of the ways I'm using openclaw to make my life easier and more consistent but I want to be honest: this stuff is addictive. Much like ChatGPT suggest next steps as you're using it to solve a problem, your LLM-driven-bot will make you aware how many multi-step tasks you do with technology can and should be done by a robot.

I am a huge fan of GTD and productivity todo lists and having managed and built multiple businesses staying organized is a core skill that I am constantly fine-tuning even in my retirement. As easy as all the list solutions are to set up - they are even easier to ignore. The way you adapt any new process is the same white-knuckle "do it for 21 days" until it becomes a habit. Most of the time, you find way sooner that it's just not going to work for you. This is where openclaw is different: You program it to remind you why you got started, you program it to use the language that gets you back on track when you slip, and it's right there in your usual chat stream so it's hard to ignore and easy to get reminded. While I'm the king of ignoring notifications, I always seem to manage to get to all my messages and this works for me

Fitness & Health: Agent Allison

They say adulthood is just a neverending pursuit to lose the same 5-10lb. I still run and I use everything from Apple Fitness to Strava to LoseIt and more often than not I forget about them when I'm not trying to cut before the race. If you follow my Instagram you know I've never missed a flavor of Oreos and boy do they have a way of disappearing around my house.

Keeping track of everything is a PITA and tracking calories is a challenge when you're cooking stuff for kids or throwing out leftovers - undocumented snacks pile up quick. So I built a bot to help me track my exercise and calorie intake. Now before I eat something I just take a picture and bounce it to her:

It keeps track of everything I eat, makes recommendations based on my diet plan, makes adjustments day-to-day and helps me stay on course. I also feed it my Apple Fitness data:

Since February 2nd I've automated this process and she has access to it daily and helps me structure the next day. It's not without bugs but way more consistent than I am.

Logs & Alerts: Agent Watson

There is nothing more panic inducing than a call from your CTO/CIO after midnight.

Ignored and silenced alerts from thousands of different services (still mostly DNS and Amex) can be just as deadly.

I have an agent that is subscribed to all the notifications, alerts, warnings, and beeps in the dark I receive around the clock. In the context of monitoring tools, it is hard to define a creeping issue that will result in a production crash that will take out your entire service/network and it's even harder to know which alerts should beep at 2AM and which one can wait.

Enter Agent Watson: I now have an LLM with my personality baked in that can triage the issue and figure out which ones I have to address first. Mysql docker crashed for a blog nobody has visited on this side of Covid? Ssssh. Certbot crashing in a loop on a production network and nobody has looked at it in hours? Yeah, let me know!

There is only one Vlad but I keep an eye on multiple Slack channels, multiple ntfy servers, I have 2 cell phones just for "urgent" alerts! LLM and the curated instructions the bot leverages don't eliminate any of that - but they help facilitate getting important alerts to me quickly while leaving unimportant stuff for the scheduled time when they get my undivided attention.

TV Guide: Agent Joan Rivers

Back in the day it was easy to know what to watch: You either got it from Blockbuster or the network show reminded you of a hit show during every commercial break. With the new era of DVRs and streaming there are millions of options and sometimes I spend more time scrolling through stuff I don't want to watch. Enter Agent Joan Rivers (remember TV Guide Channel?)

We use Jellyfin for our home entertainment and everyone has their own Jellyseer account to request the shows they want to watch. Every time a show is downloaded it generates an email that nobody reads.

What Joan Rivers does is enumerate my DVR email folder for all the downloads that completed that day. It parses the weird filename into an actual show, season/episode, and pulls similar shows and downloads images and summary from imdb with a youtube link for a trailer. All we get is a single email every day:

It takes unrecognizable filenames in a download report and spits out a tailored TV Guide for the whole family along with recommendations thanks to bots memory, soul, and documented instructions.

Finance: Agent Ron Insana

One of the most frustrating thing in finance is getting to the bottom of why certain stocks are making disproportionate moves. Now that NYSE is trading 24/7 the sun is set on Bloomberg Terminals and the process of discovery is usually digging through random Twitter comments, reading articles behind a paywall, going down the rabit hole of fake rumors and uncited press releases.

Ron Insana was a CNBC host that typically crushed the markets in the final hour of trading by slicing through the fud and nonsense of the IPO era. Now he is my agent that looks at developments across financial news, research newsletters, Twitter chart junkies and my trade groups. My bot looks at my portfolio watchlist and watches my Fintwit feed along with all the popular sites (not directly of course, this thing is so prone to prompt injection that everything is scraped, sanitized, and passed through an n8n workflow).

Bottom line: Every morning and evening I get a report that tells me what's going on with that stock/security in under 140 characters.

Youtube: Agent Sweet Brown

As a geek I have a ton of Youtube subscriptions for all the random hobbies I've picked up in my retirement. They are fantastic but I understand everyone has to make a buck so sometimes a 30 second video is 20 minutes long. And do I really care about a new RG Abernic handheld? Say it with me:

Every evening my Sweet Brown agent enumerates my Youtube subscriptions and looks at all the new episodes published that day. It pulls the transcripts for the episode and runs it through the LLM. I get a TV Guide email that includes the link to the episode and a 140 character summary of what they covered.

It also has instructions on what I really want to see! This is where it's magical. If the episode has transcripts (some don't) and it meets the criteria of what I want to watch - it adds it to my playlist. This way I have a Vlad channel that is curated by my Agent Sweet Brown and I don't open Youtube and go scrolling, I go straight to my playlist and let it rip.

Agent Prince Adewale

If you thought I was an asshole before, wait till you read this :)

I have spent most of my adult life killing SPAM for a living. Hand on the bible - "unsubscribe" processes don't do shit. Yeah, they might unsubscribe you from one or two reputable companies but the new era of Zoomer businesses changes marketing tools more often than they change their underwear and I'm getting marketing emails from several "employees" of the soap company. The moment I unsubscribe from one of them there is a notification from another. Took a car to a dealership? You're getting an email from the service department, from the sales department looking for a tradein, suddenly you're learning about their awards and holiday specials. There is no unsubscribing from this shit. It's endless because companies don't care about unsubscription requests and they move their mailing lists a lot.

You know who cares? The designated ARIN IP block owner. OH THEY CARE ABOUT A DIRTY IPv4 BLOCK A LOT. Send an abuse complaint to that address and cc: an RBL and half a dozen nerds start sending nasty emails.

Enter Prince Adewale, son of a former Nigerian minister... but my dear friend finally got his money out of his war-torn country and now he works for your boy Vlad filing abuse reports on your fly-by-night dropship-from-Temu Instagram millionaire. Second email today about a new soap scent your company dropped? I'm not unsubscribing, I'm dragging your email to Adewale's folder.

How it works: I have a background job that copies messages from Adewale to my agents mailbox daily. When I get bored I open a Chrome tab and send a message to Prince Adewale on Telegram to get started abusing the abusers. First, the bot downloads the message, explodes the SMTP headers and looks up the netblock owner. Then it scopes out abuse and postmaster contacts and drafts an official abuse email to send on behalf of ExchangeDefender. At the same time, a second supervised process opens up the unsubscribe link from the email, automatically fills out the form, captcha and navigates through the emotional messages marketers think make a damn bit of difference - screenshot of the unsubscribe gets attached to the ARIN abuse form and packed away.

My inbox hasn't been this clean since the 90s.

Time Suck

Once you realize just how many of your pursuits have failed because you lacked adequate accountability or you let process friction hold you back from realizing your goals... you will waste an indefinite amount of time tweaking your bots. Personally, this way is more fun :)

It's by no means magic - processes break, cron jobs don't run on time, sites time out and I try to put an n8n & scraper in between everything these bots get fed to them. All in all, I spend way, way, way, way, way less time managing my workflow than I did wasting time on just one of the tasks these bots do.

Meet Stephy, my clawdbot

Note: Updated on February 12, 2026 to reflect the name change. While most of this post still stands I encourage you to look at https://openclaw.ai/ first if you're deploying it in the future as this product has changed the name twice since I've mentioned it -- and I'm sure best practices will too.

I've been experimenting with AI for years now and this is the first project I've become absolutely obsessed with. Don't get swayed by supernatural Twitter claims by lying "influencers" - no, it will not print money, no it will not make you a fortune trading crypto while you sleep, no it cannot run your entire business - as a matter of fact it's a clunky pile of #vibecode npm nightmares that barely functions. When it does work though - it's miraculous. It's why I named mine Stephy, after the best admin/marketing miracle worker I ever had a pleasure of working with.

What is it? ELI5

It's a robot that lives in your computer, with access to the same tools and web services you have access to. You chat with it through your favorite chat app and tell it what to do. It has a soul, personality, heartbeat, and a scheduler that enables it to remember your interactions and perform tasks for you on schedule.

At it's core it's a npm package that uses an LLM (OpenAI, Codex, Claude, etc) to coordinate and execute tasks based on instructions it keeps in markdown files. As you talk to it through your chat app (Telegram, Slack, WhatsApp, etc) it references those files for instructions on how to perform a task, how to respond to you, which tools to use, etc. It also has skills which are npm packages that enable it to interact with web services and tools to get the job done. Over time as your md text files grow and you get the context down to exactly what you want, it can perform your tasks on a schedule in a cron it maintains on it's own.

I'm sold, where do I put it?

Resist the temptation to buy a Mac mini. I know it's all the rage right now and the potential is there -- but it's pretty half baked and just like every other npm project out there you're likely to mess it up adding vibe coded skills and broken dependencies. YMMV.

You don't need a dedicated PC or a laptop for this, unless you expect to have it use your Chrome browser (ffs don't!) or edit video the resources of a free VPS from Amazon or Oracle Free Cloud are more than enough. I have one running on Raspberry Pi 4 for house stuff and it's just as happy as the one that lives in a 16GB Proxmox container in my data center. If you don't have Linux skills or aren't in a mood for a learning curve to setup the free VPS instance, check out VPS providers like RackNerd or Digital Ocean.

February Update: This is the friendliest YouTube guide I've seen on getting started but it's as outdated as this post - AI development moves fast! I still recommend watching it just so you can get an orientation on how to install the package, ssh tunnel, and a Telegram bot https://www.youtube.com/watch?v=3DBpfB0ao50

How do I install it?

The installation is actually the simplest part of the process, you just run this command in your shell:

curl -fsSL https://openclaw.ai/install.sh | bash

This process will install all the required packages and the next thing you need to do is "hatch" your bot.

openclaw onboard

You're now setting up your bot and I cannot encourage you highly enough to keep the documentation handy. I have only one recommendation here: when it comes to gateway bind it to 127.0.0.1. Idiots are binding their gateway to 0.0.0.0 and getting pwn3d left and right so please don't do that. Lot's of newbie YouTube walkthroughs are sponsored by VPS providers whose IP ranges are well known and it's trivial to nmap the whole range for an open 18789 port and steal your credentials. Don't be stupid.

Once you finish deploying your bot and onboarding it you have several ways of interacting with it. My favorite to get started is openclawd tui which will give you immediate access to your bot through the same terminal interface you used to install it. Give your bot a personality, name, what you want it to call you and how you prefer to interact with it.

If you'd rather interact with it through the powerful web interface, bring up an SSH tunnel. I'm assuming you've downloaded your VPS issued ssh key to your ~/.ssh so mind your paths and filenames

ssh -L 18789:127.0.0.1:18789 vladbot@demo.vlad.net -i ~/.ssh/vladbot-demo-vlad-net

vladbot@demo.vlad.net is my username and my vps server, the part at the end is path to my ssh key. Yours will be different but once you launch this you'll be able to talk to your clawdbot at https://localhost:18789

What else do I need?

Next you have to pick a chat agent. I've personally used it on my infrastructure through Telegram, WhatsApp, Slack and Mattermost but it supports a TON of chat platforms. I loathe Facebook properties and I've found that it's the most reliable on Telegram. On Telegram you just start a chat with Botfather, /newbot and follow the instructions to get your bot. By default this is locked down to you and tui or webUI make configuring the chat part of this super easy.

Some Tips

Most of what you see on social media is pure hype and fantasy. Remember these dudes hype and cut scenes to get the clicks and go viral... and if you follow them step by step at best you're going to get hacked. Here are some of my suggestions when it comes to use:

  1. Make backups. Keep your bots markdown files backed up often. This is perhaps the best way to learn how clawdbot works, instruct it to make backups in ~/backups/ every day at midnight and report back. You will mess up your bot and your config as you play with it and copying back the *.md files will get you back to the experience quickly without having to recreate everything.
  2. DO NOT GIVE IT ACCESS TO YOUR ACCOUNTS. But Vlad, how am I going to get rich if I don't give it access to my Polymarket and Coinbase accounts? Just like every other crypto influencer that hasn't blown their brains out in the back of a rented Lambo - you won't. And if you're reading this post you likely don't have a body for OnlyFans so calm down and give your bot it's own accounts. It is much easier to create a forwarding rule for mail or to share a read-only calendar than it is to get your accounts back once someone has hacked it.
  3. Take it easy with skills and plugins. Remember that most of these addons are community built and have not gone through any sort of a security audit. Same caveats as Chrome plugins. It might be able to run your entire company but it also might send all your data to an intermediary, use your brain.

Have fun!

If you've ever used ChatGPT or Claude or CoPilot you know how much fun they can be but you've also probably experienced hallucinations and outright wrong answers. What makes Clawdbot special in my experience is the memory and instructions. It knows how to interact with the LLM to get the answer that I am looking for and it knows how to format it right. By giving it additional instructions and corrections it keeps adding specs to your markdown files which makes it far more fun and personable that ChatGPT - even when used in context with projects.

Leveraging automation with memory feels like a giant new step in the AI evolution. Happy to be alive to see what this thing does next.

Removing Microsoft Copilot

If Microsoft Windows 11 hasn't compelled you to switch to a Mac or Linux yet, you're likely seeing way more Copilot in your day-to-day. For my fellow geezers that lived through MSN and Bing, every so often Microsoft invests an enormous amount of money in a second-rate release that the only way they can get you to use it is by forcing it on you. Repeatedly. As of Windows 11 25H2 the history repeats itself and your boy Vlad is here to help you get rid of it (somewhat).

Start > Run > gpoedit.msc

This will open Local Group Policy Editor. Go down to User Configuration > Administrative Templates > Windows CoPilot and double tap Turn off Windows CoPilot.

Select Disabled and click on Apply / OK.

Now either reboot or run gpupdate /force from the command prompt and you shouldn't see Windows Copilot again.

I want it gone-gone!

Read the opening paragraph again. Like herpes there is no getting rid of it, at best you're managing outbreaks until the next major update where oops they've enabled it again. If GPO doesn't do it for you, you can also remove the actual package. To do so launch powershell as an Administrator and run this:

Get-AppxPackage CoPilot -AllUsers | Remove-AppPackage -AllUsers

You now have Copilot-free Windows.