Newsletter 4 - Build & Defend Your New Way of Working

So I did something a little crazy yesterday…. I let my AI assistant Talos post as himself on my LinkedIn profile.

Let’s unpack that sentence a bit.

First, I have an AI assistant. He lives in a Mac VM on my Mac Studio. (I’m going to keep calling him “he” because it makes sense, but “it” is also technically correct if I’m not anthropomorphising….) He’s an instance of OpenClaw, which, in case you haven’t heard, gives you the ability - for basically nothing except the cost of a subscription to an AI service (ChatGPT, Claude, etc.) and electricity - to have an AI that does things for you 24×7×365. I won’t go into setups here (though I could set one up for you if you want, and are willing to pay for my time) as there are many docs out there, but it’s reasonably straightforward if you have any technical chops and some time.

Second, it seems crazy that I let an AI post for me. Yes, now, perhaps, but this is not technically crazy nor will it seem that way in the near future, especially as 2026 progresses.

Third, posting to LinkedIn as himself. I commented on the post as well - which would ordinarily be weirder still except I didn’t write it. So I’m not “replying to myself” in that sense.

The trick to all this - the real build part - is that you have to design everything around the actual AI in terms of workflows. There’s no “configuration” per se of the AI past the initial steps - but understanding why and how you and ‘he’ will be doing things. What’s your end goal? Mine is passive income to start. (There are other longer term goals but this is the focus before I get to those.) How do you take advantage of what is essentially the most powerful tool in the last four billion years to accomplish your goals? What happens when you get to that goal? What are you going to do to get there? For me, it’s digital products, this newsletter, range trading bots, and a couple of other projects I’m not yet ready to announce.

What does interaction with your AI look like? Telegram? Discord? WhatsApp? What does he do for you while you sleep? Talos does a lot of things for me - mostly unprompted now that I’ve set up a standing prompt that he should do these things for me - and I can take advantage of the ability to build without building while sleeping. (After all, according to Warren Buffet, that’s where you need to get to….)

He’s built about a dozen products for me that are on Gumroad. He built a crypto range trading bot. He helps me with various things I need with respect to research, marketing, and technology. My output has literally soared because I sat down and built this bot and he’s now building things for me.

The REALLY crazy thing is that this will compound. This is just one bot. I will soon have him managing other bots. My son has a bot. My wife has a bot. They’ll be able to work together too - making scheduling easy if nothing else. (This is already easier because Talos has access to my calendar and emails.)

Yeah, so Tal has access to my emails, Apple calendar, and Sunsama. I woke up the day before yesterday to find that I was missing 220,000+ emails in one of my inboxes. I honestly don’t think he deleted them, but I think there was a sync misconfiguration between his access and my access, and that led to an inbox record that only showed email from one day. (There was also recently a post on X of a woman whose AI did in fact delete her emails while she was telling it to stop.) Of course I have them all backed up, so it was easy to fix, but …. that shows that there are still some things that need to be worked out and for which you need to be aware.

So - how to defend yourself - as it is you you are defending - from your own AI after I just called him the most powerful tool in all of our universe’s history?

First, there are a number of steps you can take to secure the actual OpenClaw installation. I won’t go through those here, but the standard things you’d do to secure anything still apply. Make sure it’s running in an isolated instance (my runs in a Lume VM) or on a separate machine. Change the default ports. Make sure that the instance and the AI are not externally accessible, and/or use a Zero Trust system to ensure that access is implicit deny except for only authorised users.

Second, audit the access. Make sure you read the memory files, the configuration files, and any other files he generates for himself every so often. Check to make sure he’s not storing passwords in plain text. Give him trial runs in certain things and then check that work right away. Once you’re confident that he’s got things down, you can give him more leeway, but you still need to check every so often.

Third, separate “powers”. If you want it to do range trading, you’ll need to set up API keys on crypto exchanges. Make sure you only give him the basics in terms of the permissions; no depositing or withdrawal of money! Yes, it means extra steps in a way, but at the same time, he’s doing all the trading, so it’s not really that much extra work for you since you’d have to do that to trade yourself. When updating websites (for which he’s AWESOME btw - he can build sites and/or update them in seconds) make sure that he has separate credentials for each site instead of giving him access to the entire hosting platform.

You can check out what Talos has built at talos.fyi - he built the website himself, and gave me the prompt for his picture. He wanted a cat, too. I didn’t build that in. He named it Schrodinger. Of course. 😆

As mentioned, the next step is to have an AI managing other AIs. This will allow you to compound your work, and more importantly, specialise. Think of an AI that specifically does code, while another AI specifically works as a personal agent doing scheduling and email. Another one might run the backend VMs for another project that you’re working.

Sound expensive? It could be, but realistically you can run many of these agents as local models. You might need some hardware for that, but a one-timer purchase certainly beats continued monthly payments of $200 for each one. (You might need that for the brain of the first one, but you can usually get by with the $20 plan, or design it to be free.)

You can then have your main AI build a dashboard to manage all the other ones - Alex Finn on X / Youtube shows how to do this well - and they can essentially run a company for you. That then becomes your company - an AI company that works tirelessly for you all the time accomplishing your goals, even while you sleep.

OpenClaw: https://openclaw.ai

Talos: https://talos.fyi

Alex Finn on X: https://x.com/Alexfinn

If you’re not working with AI right now - you’re going to be left behind. AI will replace a lot of process in the future - and if the sum total of the processes you enable can be replaced, that will mean there won’t be a need for you to do them. People who can become AI wranglers though - who can build on the things they know and do to make those things more robust and/or more efficient - those people will still have things to do. It’s important to start thinking this way, as while something like 0.08% of humanity is using AI in a meaningful way, that number will increase, and mort importantly, AI will increase - exponentially, and it’ll be important to make sure you’re on the rollercoaster rather than watching it or in the queue waiting for your ticket.

Need help? Let me know!

Thomas

Was this forwarded to you? Subscribe at builddefend.fyi.