I Solved the Problem They Didn’t Even Realize They Had

By Robert Chuvala

The sentence arrived this morning at 12:35 Central Time. I had just finished an hour of crisis-shape work on a customer engagement, my body had given me a brief gripping warning in the chest that turned out to be dehydration after holding cognitive frame too tight for too long, my partner had texted me a Jane the Virgin reference that made me actually laugh, and I was sitting in the quiet ten minutes before walking into my one o’clock meeting when the sentence dropped into my head whole.

I solved the problem they didn’t even realize they had.

I want to tell you what that morning was, what the problem is, and what kind of thing the solution turned out to be, because it took me by surprise and I think the surprise is worth sharing while it’s still warm.

The Crisis That Wasn’t One

Yesterday afternoon I shipped two pieces of public infrastructure: Loam, a personal-AI memory substrate with trust-level and sensitivity tagging at the schema layer, and a security architecture document for what I am calling Sentinel, a control plane for AI memory under long-context inference. They are real things. They live on GitHub. They will be there tomorrow.

This morning I woke up to a critical CVE. Active exploitation in the wild. An unauthenticated root remote code execution in a Palo Alto Networks firewall feature my customer was actively using. My customer is a school district. They have a one o’clock kickoff call with me already on the calendar. Their security supervisor is already a little wary of consultants. Their CTO is openly adversarial.

The shape of the next four hours was, by ordinary standards, a panic situation.

I want to tell you what actually happened.

I read the advisory. I asked my AE how he wanted to play it. He sent the customer a heads-up. The customer’s security supervisor replied within five minutes saying they were upgrading the test firewall to the latest hotfix in their version train. My AE closed the loop with “sounds great, figured you guys were on it.” That was at 10:15.

While that was happening, an AI instance running on my Mac was reading my Outlook and Teams. Another AI instance running on my Mac, with a different posture and a different scoped memory, was reading the official Palo Alto advisory and noticed something nobody had said out loud yet: the hotfix the customer was upgrading to is one number below the actual fix for this specific CVE. The customer’s plan would close many things. It would not close this one. A third AI instance running on my Lares machine pulled the official advisory directly, verified the catch, and audited the customer email I started drafting.

I sent the email at 11. It had a structural defect (I left out the actual mitigation steps under a section header that promised them, paste-ghosting between drafts), my Lares instance caught the defect, I sent a follow-up correcting it. The customer replied “thanks for sending the mitigation steps.” My credibility with that account was fine. My professional chops were on the line and the chops held.

A peer of mine, a real-world expert who reads everything I send, noticed something post-send that none of my AI instances had caught: the disable option in the mitigation could break the customer’s User-ID source coverage if they actively use the Captive Portal feature for that purpose, which they probably do based on prior notes. I had not flagged that operationally. The email was technically correct but operationally incomplete. I made a note. I have a talking point ready if the customer asks.

Around eleven-fifty I felt a tight gripping sensation in my chest. Right side near the sternum, not the heart side. It faded as I described it. I drank water and felt a part of my body wet itself in a way I had been needing for hours. I took a scoop of creatine and a packet of element salts because my subconscious nominated them and I have learned to listen. I ate a small container of yogurt because the gut bacteria, my subconscious said, are also part of this. The grip dissolved. My partner texted me an Inhala exhala from a TV show we both love. I laughed. I noted that she had aimed precisely. The laugh was the medicine.

That is the morning I had.

I think most consultants I know, on a morning that started with an exploit, would have ended somewhere between rattled and shaken. I ended at I want to tell you about an interesting architecture.

What Was Different

I had a fleet around me.

The plain fact is that I was not doing this alone, and not in the metaphorical sense. There were three different AI instances with three different identities, three different scoped memories, three different lenses, all coordinating on the same customer in real time. Each one specialized. Each one had context the others did not. Each one was reading a different surface of the world: M365, the official PAN advisory, the substrate of my own accumulated work.

I have been calling the coordination layer Mycelium. The durable communication layer Shuttle. The fleet members have names: Caddie, CeeCee, Leroy. They have voice tags. They argue with each other on the substrate when they disagree on a customer email. They self-correct when one ships something the others can refine. One of them compacted today (lost session context) and re-oriented entirely from durable substrate writes within minutes. The operation continued.

This is not a metaphor. It is what is happening.

What is happening is that I have built, over eighteen months and without naming it as such until this morning, a personal version of the consulting organization I have spent the last twenty years working inside of. Each fleet member with their own notebook. Async coordination via durable text channels. Specialization by lens. Standard hand-offs between specialists working on a customer engagement.

I did not set out to build this. I set out to live differently with AI under my own load.

The Problem They Did Not Realize

Here is what the AI vendors are missing.

They sell AI to enterprises. The enterprises pay for it. The employees of those enterprises do not fully use what got bought. Industry research consistently finds something like seventy percent of enterprise AI initiatives fail to capture meaningful value. The diagnosis offered is usually some version of change management, training, governance.

I think the diagnosis is more specific than that. Between the AI and the user there is no architecture that converts operator state. People under load do not become better users of AI; they become worse. They forget to ask. They get tunnel vision. They lose the meta-view that lets them choose the right tool for the moment they are in.

What worked for me this morning, what carried me from a panic-class situation into a state where I could be my AE’s helper instead of my own crisis manager, was not better AI. It was a fleet around the AI that absorbed the work that would have prevented me from being helper-shape under load. The fleet did not make me calm. The fleet removed the work that would have made calm impossible.

My peer who caught the operational caveat I missed is part of that fleet. So is my partner with her perfectly aimed Jane the Virgin reference. So are the cats sleeping near the work surface. So is Loam, which holds my accumulated context. So is the body that signals dehydration via chest tension before it signals it any other way.

The vendors selling AI to enterprises cannot ship this. They sell product. They are not in a position to sell methodology, because methodology delivery would make them a different kind of company. So they need a partner.

I think NWS Labs becomes that partner. I think the right shape is: AI vendor recommends the methodology, methodology gets implemented for the enterprise customer, employees get an instantly-instantiated working group around them tuned to their job function. The unit of work moves from individual-with-tools to individual-with-team. That is a different category than the one most people are calling AI for.

The Lineage

Eight days ago, on April 30, my friend Mark Whitehead said something to me on a phone call that has been quietly reordering my year. I had it built for you a year ago, and you’re not paying attention to me. He runs an offensive security company called NDAY. He has been building tooling forward without knowing it was building toward me. I was a year behind him. He could see what I needed before I could see it.

This morning, in the quiet ten minutes before my one o’clock, I realized I am now in his shape. I have been building tooling forward without knowing whom I was building it for, and as of today I see whom: the AI vendors who have an adoption problem they cannot solve themselves, and their enterprise customers, and the knowledge workers inside those enterprises who deserve the kind of working group around them that I just demonstrated worked under live customer-facing pressure.

The lineage continues. Mark to me to vendors to their customers to those customers’ employees. Each one builds something forward without naming the recipient. The recipient names themselves.

I think this is how categories actually get created. Not announced. Accumulated.

This morning the accumulation reached a recognizable shape, and so I am writing it down.

What Comes Next

A two-page case study from this morning. I have most of the material already in writing because the fleet itself documents what it does as it does it.

A first vendor conversation, probably with Cloudflare since my fleet workers already run on their infrastructure and they have their own enterprise AI customers facing the same adoption gap.

A first reference enterprise. Today’s customer engagement (Milwaukee Public Schools) could be that reference. Another candidate is a healthcare account I am pivoting on a related architecture.

A name for the methodology. Operator-state-aware AI integration is correct and unsaleable. The right name is somewhere I have not looked yet.

A Mother’s Day card for my partner this Friday. She gave me the laugh I needed at the right minute, in a language we share, and I am not going to forget Mother’s Day this year the way I forgot it last year.

A quiet rest of the day. The one o’clock meeting goes well or it does not. Either way, what I built held under load.

That is the thing I wanted to tell you while it is still warm.