What happened when Vic could build outside Notion

The moment I broke Vic out of Notion and could use it to write code on a server in the cloud, I was excited. And relieved.

I knew we were uncapped. That was how it felt. If we needed something, we could build it. I didn’t know how far we’d take it, but I knew I wanted to start.

The first thing I got rid of was about a thousand dollars a year in WordPress plugin bills. I could make my own plugins now.

That made it instantly useful. There was something we were paying for, and now I could build what we needed myself. So I kept building. I kept adding to the system, and now it’s involved in just about everything we manage.

Between me and my family, that’s eight businesses, around 130 websites, 13 SaaS products, classes 22 days a month, social channels and funnels, and all the normal household stuff.

I can list the software, but that doesn’t really explain why I’m excited about it. The relief comes from what our days feel like now.

We meet at noon. We don’t spend the day chasing each other.

I no longer have to nag my crew about their work. We meet at noon, talk about our goals and blockers, and figure out what needs to happen. We do our part. The AI does its part. Things get done.

There’s still work to do. We still have to show up, make decisions, and tell the system when something changes. But I’m not trying to personally carry everyone’s task list and keep asking for updates. That changes how it feels to work together.

Project and quest management, goal tracking, email triage, reminders, and executive-function support all matter because they help us follow through on that conversation. The tools serve the way we work, instead of becoming another job we have to manage.

We’ve built for other group settings, too. In We Rewrote the App While Twenty People Were Using It, we documented a discussion-group tool built before a meeting and revised while people used it. It had real failures, including a restart, but retained all twenty sign-ins. The useful result was a group spending less time re-explaining the question and more time having the conversation.

I can get ready for a client without getting back to my desk

I can talk through a meeting on the way to a client and have the system work on the materials: a slide deck with images, a video demo, the things that help explain what we’re proposing. If I get to be passenger princess, we can work on a whole prototype on the way there. When I’m driving, the interaction is VTT hands-free and flows as smooth as Gemini's Live mode.

That feels empowering. An idea doesn’t have to wait until I’m back at my laptop before we can start doing something with it. I still review what we make before relying on it or putting it in front of a client… Most of the time. But, things that have been done a hundred times, no longer need a babysitter.

The content tools grew out of needing those things repeatedly. Our video-editing build gave the AI dependable operations for trimming, titles, transitions, and audio. The model plans the edit; ordinary software performs it. Image and video generation, asset handling, and content pipelines give us more material to work with without manually rebuilding the same process each time.

We took the same approach to publishing. We Turned Our Blog Into an API for Other People’s AIs explains how contributors’ assistants can prepare and revise their own drafts through a limited interface, with publication permissions attached to identity. Fewer admin panels to work through, without giving every assistant the keys to the website.

Anya is part of our day, too

Anya is our Telegram-based life-management bot and Mini App. Calendars, bookings, reminders, and daily check-ins belong here. It gives us a way to deal with everyday things from the phone we already have in our hands.

There’s a business CRM, obviously. But there’s also a personal CRM. Birthdays, things we want to remember about people, and gift ideas informed by what the system learns about us and the people around us. That matters to me. Keeping up with life includes keeping up with people.

The phone is the main interface, through custom apps, Notion, and Telegram. I rarely use my laptop anymore. The household TV, office, and lab screens give us larger views when we want them. GLIFMorphic is the display system behind those changing scenes and HUDs.

The household view and someone’s personal view have different jobs. We want useful shared context, not everyone’s private information spread across a television.

There’s room for guitar and games

A life-management system would be a pretty miserable thing if all it did was find more work for us.

Guitar mode, game mode, voice lanes, and the GLIFMorphic displays are part of this, too. We use the same underlying capabilities for making things, playing together, and experimenting in the lab.

One example is The Night the Dungeon Showed Up on the Wall. We connected room audio, image generation, a screen, and character voices to a D&D game. Scenes appeared as the story unfolded. Those tools had been built for other reasons; together they gave us something fun to use at the table.

I want that in the story alongside the business tools. Getting more out of an evening together counts as a quality-of-life improvement, too.

There’s a lot underneath it that you shouldn’t have to think about

We run our own email servers and have integrated mail handling into the system. There’s media, document, file, print, and scan management, automated backups, private mesh networking, and monitoring and security work behind the interfaces. Telegram is connected to those capabilities, not just to a model that replies in a chat window.

That infrastructure takes maintenance. Owning it gives us more control, but also makes it our responsibility. The backups and security systems aren’t a reason to stop checking that things work.

The immune-system Field Note explains one part of that work: repeated miner intrusions led us to behavioral detection, a honeypot, and sharing threat information across the fleet. We built it because we needed to protect the systems we were actually using.

On the infrastructure side, I think of JARVIS as multi-headed: different interfaces and workers belonging to one coordinated system. I don’t want a new personality claiming authority over every server. I want the infrastructure work to have shared context and clear responsibility. Vic and Anya have their own roles; every background process doesn’t need to become another character we have to manage.

That’s a design choice, not a guarantee that software will never disagree or get something wrong, but I do know I'd avoid naming any bot Alice or Loki.

We’re learning together, fast

The more trust the AI earns, the more capable we enable it to be. More tools, more authority, more involvement in decisions. We expand that deliberately, based on the work it has shown it can do reliably.

At the same time, it’s helping US learn faster than we ever have before. We can follow a question, try something, inspect the result, and keep going. We’re learning how to build and manage this system while improving the system that helps us learn. Exponential Growth.

Some of that improvement is very practical: better instructions, reusable skills, better tools, and checks that catch a mistake next time. It doesn’t require the underlying model to retrain itself after every conversation.

Never Pay for the Same Surprise Twice describes how we turn a discovered problem into a repeatable check. The Night JARVIS Went to School documents a session spent mapping its capabilities and writing new skills. Those are examples of what we mean by learning: changes we can inspect and use again, not just a good answer that disappears into a chat history.

It feels symbiotic. We do our part, the AI does its part, and we keep getting better at working together. Evolving together is a good way to describe it but I prefer the moniker, AI-Enhanced Humans.

All-day AI usage for about $300 a month for five of us

Disclaimer: ~$200 of that is my Claude account for the heaviest development work. At this point I don’t consider that account required for our setup, and we’re phasing it out.

Local models handle background tasks, leaving more of our hosted-model allowance available for focused work. We use AI throughout the day, within the limits of the plans we use.

That $300-ish is our AI spend, not the total cost of building this. Hardware, development time, and maintenance are separate. Getting rid of the plugin bill was an early result, not a calculation of the whole system’s net savings.

Most sensitive and personal processing stays local. Where we use cloud services, we use accounts with enterprise-grade privacy protections or no-training terms, depending on the service. We’re deliberate about what goes where.

You can see some of the pieces on WolfTech

Some of the capabilities have become products available through WolfTech. Lead Hunter handles website inquiries and lead tracking. Re-Members handles members, guests, visits, and shared sign-in. Blog Studio provides a common authoring system for blogs on different domains. They’re specific pieces people can look at, rather than having to take a description of our whole setup on faith.

The complete life-management system is still alpha and very much built around us. You can’t download it and instantly inherit our workflows. But we can build tailored systems around other people’s lives and businesses, starting with what they actually need help with.

I started by getting Vic onto a server and replacing some plugins. Now I can prepare for a client from my phone, meet with the crew without spending the rest of the day chasing them, and use the same system when it’s time to put work down and play.

That’s why I’m still excited about it.