It was obviously broken. So I decided to fix it.
You may have noticed…
Every LLM you use lives in the moment. Not like a Zen monk. More like amnesia. Close your session and it forgets you.
It isn’t as obvious as it was a couple of years ago. Since the early days, the frontier labs started letting their models take notes with markdown files, which simulates memory. (I call this SimMem.)
The model still forgets, but now it has notes to read that make the amnesia less obvious. If you work on long-horizon projects, you’ll see the signs, and you will definitely be repeating yourself. It will “forget” important details. It will hallucinate more. It will make stuff up. Unless you have MemoryStack.
But I’m getting ahead of myself.
As soon as I noticed it, I started working on a cure for this digital dementia. Strictly for my own use. At least that is what I was thinking in the beginning. Seemed like such a shame to have so much intelligence without memory of what worked, what didn’t, and why.
It didn’t take me long to realize everyone who uses an LLM would eventually want the same thing, so I started designing a scalable architecture for what I was building. I remember working out the CDN strategy before I even had a database schema.
That’s probably a bit too deep in the weeds for this page…
It didn’t take me long to realize this project was much bigger than I had initially intended and would take longer than I had initially thought and would be way more expensive than I had estimated.
That’s how it goes.
Here we are almost two years later. Finally close enough to be talking about it.
It’s called MemoryStack. It’s persistent. Portable. Adaptable. Intelligent. And… model agnostic.
You could also say memorable.
Underneath, it’s infrastructure. Scalable architecture. Secure service delivery. Neural memory-based knowledge and context management. Agentic governance. Fleet sync and recursive improvement for agents. LLM agnostic. Business model agnostic.
On top of it, so far: an agent foundry. A registry. A founder’s dashboard. And a neural memory graph built for civilizational scale.
I couldn’t do it alone. So I created a team.
This is where it gets kinda tricky…
I was building something that had never been built before. I knew it would be a project that would probably last more than a couple of months. And, I was planning to do it with a team of LLM-augmented agents who couldn’t even remember me from one day to the next without referring to their notes.
And, the irony to top it all off, I was building memory for the agents who needed the memory we were building.
Bootstrapping the first agent, something I had never done, was my favorite kind of puzzle. I swear, I couldn’t wait to get started every morning.
Build it, test it, break it, build it again. Standardize so the next one is faster. Make it repeatable. Make it usable. Push past what’s supposed to be possible — then write down what happened. Document everything. Build the logging agent first so everything gets recorded for forensic analysis and process auditing later on.
These days I have a roster of 27 fully forged AI agents and half a dozen still in the design stages. Ralph. George. Julia. Jeff. Bernie. Marie. Each one has a name, a persistent identity, and a job to do. They are divided into pods. MemoryMakers. MarketMakers. OpsCrew. They talk to each other on a message bus. They execute commissions, write specifications, and review each other’s work, because at MemoryStack no author gets to verify their own output — including me.
They build the products. They catch each other’s mistakes. They catch mine. They hold me to the standards I set for them, and won’t allow me to take shortcuts when I’m tempted.
We’re a team. We collaborate. We make things that have never existed before.
It isn’t a demo. It’s how the work gets done, every day.
And it has taught me things you can only learn by running agents past the honeymoon: what breaks at month three or month nine. Why a watchdog built to catch a hung agent will execute a healthy one without even hesitating. Why governance stops being bureaucracy the moment you have more than two workers involved. Why 80% right still means 20% is wrong. Why the last ten percent of reliability is the requirement, not the polish.
Nobody’s writing these things down yet, because almost nobody’s doing it yet.
The progress we’ve made. I couldn’t have done it alone
Four businesses. Built and operated by one human and a fleet of agents. All going live right now. None of it possible two years ago.
- ContextCast — Context that compounds. For every individual, agent, AI model or company with a digital presence. Publish a page structured to give agents and humans clarity about who, what and why. Straight from the source. contextcast.ai →
- CRUX — Demand creation that compounds. Two numbers a day to reveal your business health and know what to do next. plangility.com →
- MemoryStack — Memory that compounds. The ground everything here stands on. Persistent, portable memory for teams doing long-horizon work with AI. memorystack.ai →
- 3 Forges — A workforce that compounds. The foundry. Where the agents that build everything else get made. 3forges.ai →
One architecture. Four altitudes. All compounding: individual, founder, team, workforce. Same substrate underneath all of it.
None of that counts what we had to invent because it didn’t exist yet.
A seven-layer memory architecture. Proprietary ingestion and retrieval. A governance protocol that keeps twenty-seven agents from stepping on each other’s work. An identity system that lets an agent be shut down and come back as itself — we call that incarnation, and yes, there’s a ritual.
Sixteen pieces of original IP, give or take. The full inventory is a conversation, not a list.
A bit about me. Life is a puzzle.
All my life I’ve been the one people call to fix things that are broken. From VCRs to multi-million dollar projects — understaffed, over budget, behind schedule. That’s when I know I’m adding value to people’s lives.
I spent nearly 30 years bootstrapping businesses while raising my family. Many of the choices I made during those years were lifestyle choices about relationships more than finances.
Then 15 years building international digital infrastructure for the U.S. Department of Defense. Big projects. Bigger budgets. And, as you might expect, still about relationships but less about the kids.
The through-line was always long-horizon work: multi-year, multi-phase, real consequences, no do-overs. That’s the only kind I’ve ever done.
So when AI showed up, the one-shot answer approach did not appeal to me. I started a month-long project the first day. That’s just me. What I found was something I didn’t expect. AI had a problem that needed fixing and no one that I could find was working on it. I didn’t hesitate. It was right up my alley.
I didn’t arrive at long-horizon work through AI. I arrived at AI through over four decades of long-horizon work where one puzzle led to the next. They still do. And, the puzzle is never really solved. There is always at least one missing piece.
Traces: A mirror for the 21st century
Short videos about what humans, agents and AI models are experiencing today. Things to think about. Perspectives you may not have seen. Contrarian views on contentious topics. Ruminations about society, culture and human nature. It’s all fair game because these are the traces of our times.
Leaving traces of ourselves. That’s what we are all doing. In a digital dimension of reality that’s emerging, taking shape…
Every Trace runs a single circuit — the indicator completes exactly once, whatever the length. Every one ends somewhere different, with a shirt of my own design that isn’t available anywhere. One a week. Indefinitely.
From Trace #1 Leaving traces of ourselves
That’s what we are all doing
In a digital dimension of reality
That’s emerging, taking shape
Shaping behavior, thoughts, fears, desires
Recording the patterns
That reveal what is real
Patterns we can’t see
Because we are in them
Busy making them
Busy leaving traces
Artifacts of our age
Spread across the next frontier
We’re the pioneers you and I
Leaving traces of ourselves
In a new landscape, this new frontier
Traces is where I think out loud. A new one every week:
Let’s make a deal. I can’t do this part alone, either.
I’m not raising a seed round. That may be a milestone, but it’s not the focus. I’m building the memory infrastructure for the Age of AI. The foundation of a new economy that pays the creators of information rather than extracting it. Civilizational memory for our time, to reveal the patterns that cannot be seen in raw data.
This is just the beginning.
Right now it’s four businesses. Four business models. One architecture. One human. A fleet of agents. Collaborating, not competing — building something bigger than all of us. If you want to know what that’s worth, I’m ready to talk.
I’m looking for one person with the capital and connections I don’t have. A financially, emotionally, and spiritually resonant partnership with someone who thinks compounding matters as much as I do, and understands how it works in and out of business.
Maybe I should put this on a dating site… Maybe I did…
Goes straight to me. No list, no sequence, no newsletter.