Founder, Sigil Systems | AI Product Risk Review & Runtime Control | Creator of FSP & Anubis Review | Builder - Sigil One | Former USAF Bioenvironmental Engineer
United States
Joined April 2017
- Tweets4.1K
- Following1.1K
- Followers2.5K
- Likes15.3K
Pinned Tweet
Been heads down rebuilding Sigil One, and this week it finally clicked into place.
What started as a set of financial tools now feels like one connected system for household + business finances, assets, evidence, decisions, history, collaboration, and Merlin.
Rebuilt the interior. Fixed the latency. Made the workflows actually connect. Shipped a full product demo into the landing page.
Still early, but this is the first version that really feels like the product I’ve been trying to build.
ALT Sigil One landing page showing a dark financial dashboard. Large headline reads “One clearer picture. Sharper money decisions.” A Financial Pulse panel summarizes household cash, crypto, stocks, metals, and real estate, with a Merlin insight flagging a cash gap. Below are free workspace/report offerings and a product demo preview.
It takes real scar tissue to look back at your own mindset from a year ago and say,
“Holy shit, I didn't even know what I didn't know." 🤣
This week I had to debug the work, the tools doing the work, and what those tools meant by “finished.” 😂
Building software. Repairing workflows. Reviewing AI-written code. And some unexpected conversations that connected my old field experience with the problems I’m working on now.
One thing industrial hygiene teaches you: a reading only helps if you understand what you measured.
I keep coming back to that with AI. The code runs. A test passes. Another model agrees. You can have all three and still be missing evidence that the thing does what you asked.
I’m interested in ambitious ideas. I also want small, inspectable pieces that let us find out whether they hold up.
Frustrating week in places. But a few connections clicked that I wasn’t expecting, and I’m heading into the weekend with more to think about than I started with.
And you know, there was some swearing involved.
Going back through requirements, checking the actual work, and figuring out where “done” got ahead of what had been demonstrated.
Something worth trying on your own project: write down what would convince you it’s wrong. Then make sure your test gives you a way to see that. If every possible result becomes a reason the idea works, you’ve made it very difficult to learn anything.
Workflow data is only part of the agent problem.
You also need to know what changed during the work.
Which assumptions still hold?
Which approvals are still valid?
Which paths already failed?
What evidence now counts as completion?
The next action depends on that state.
One AI problem I keep running into:
A failed step does not mean every piece of prior work became useless.
The hard part is knowing what remains valid, what needs repair, what needs to be recomputed, and what should stay historical without affecting the next move.
Two things One million Thanks to everyone following. ❤️😁
What actions are you carrying out today to change your life?
I don’t need AI to be perfect.
I need it to know when it does not have enough evidence to claim the job is finished.
That sounds like a small distinction.
In a long-running workflow, it can be the difference between useful autonomy and a confident mess.
He says, "You take the red pill and you stay in Wonderland and I show you how deep the rabbit-hole goes".