“This is your ambient architecture for your company. Agents are the orchestrators that define the processing power, and then skills and tools and memory are the resources that the agents use to perform that particular output.” — Lou
This Week in 30 Seconds
- The Agentic Company — Lou’s updated ambient intelligence architecture: your business as a functional folder organization — Chief of Staff, Writing Team, Research Department — each an autonomous agent with its own skills, memory, and tools. Now portable across LLMs via adapters.
- AI Voice That Can’t Be Faked — The missing ingredient in AI slop is lived experience. The fix: let AI participate in your actual work, then explicitly tell it why you did what you did. Month by month, the output becomes indistinguishable from yours.
- Focus Before Exploring — Lou’s discipline: 1-3 core tools, operationalize revenue-producing processes first. “Who’s running the business?” is the question that cuts through AI distraction.
- Get Coached Now or Get Fired Later — Jamie’s career coaching pivot: AI displacement creates the urgency generic coaching can’t. Short course + AI vulnerability diagnostic = a lead magnet built for this moment.
- Don’s Big Launches — Two wins: a 7-sub-project deep research project produces a no-fault institutional sales framework; and 16-person group coaching cohort launches with AI-generated personalized profiles that had participants emailing the morning after: “Oh my gosh, how insightful.”
Topic 1: The Agentic Company — Your Business as a Folder of Folders
The centerpiece of the session was Lou’s updated ambient intelligence architecture — now significantly evolved after a week of studying OpenClaw, Hermes, and Sam Woods’ “cognitive architecture” work.
The shift: ambient intelligence started as a way to make knowledge folders live. It’s now a complete organizational model. The structure Lou presented:
- A
Company/folder containing functional agent folders:Chief-of-Staff/,Writing-Team/,Research/,Copy-Team/ - Each agent folder has its own
SOUL.md(identity),skills/,memory/, andtools/ - The Chief of Staff has a heartbeat — an operational loop that monitors what needs to be done and orchestrates the right agents
- The Writing Team has a
team-lead.mdagent that calls Researcher, Writer, Editor, and Audience Avatar sub-agents dynamically
The key design distinction: agents are orchestrators, skills are functions. The agent decides which skills to invoke and when. Skills do the atomic work. Tools give agents access to the outside world — Bash, MCP, Python, computer use, browser — with scope you control explicitly.
What makes the architecture portable: Lou added an adapter layer with files for Claude, OpenAI, Gemini, and open-source LLMs. Nothing intelligent lives in the adapters. If the platform changes, the intelligence travels with the folder.
On security: Lou’s position is that he’s not ready to give AI control of his desktop, but the architecture is designed to grow at his pace. “You can introduce that at your own pace whenever you want.” The security approach: air-gap the data you can’t afford to lose, not the AI — give the agent access to what you can afford to lose, keep financial/sensitive data on a VPN-only device.
Lou noted: by end of day, a scaffold would be available to copy-paste. Goal: members can eventually copy a folder, say “great, now I’ve got an ambiently intelligent folder,” and start adding skills.
Kasimir observed: “If you guys haven’t gone through those [commands in the member repo], really go and see what we have available already. It’s really a treasure trove.”
Donald’s observation: “So the heartbeat is like the system clock.” Lou: exactly — it defines the agent’s operating rhythm.
→ Deep Dive: Insight - The Functional Agent Organization — Your Business as a Folder of Folders → Deep Dive: Insight - Distributed Agent Memory — Scope Memory to the Function, Not the Platform
Topic 2: The Full Content Pipeline — From Chat to Published Hub
Running alongside the architectural discussion: Lou’s live content pipeline, which turns any working AI conversation into multiple published assets automatically.
The pipeline Lou described running before the call:
- Chat session — brainstorm, problem-solve, or research with Claude
- “Process this conversation” — triggers the skill
- Recap → posted into the AIMM Living Knowledge Vault
- Teaching block → a publishable Medium-style article with the problem, solution, and process; includes graphics and excerpts; stored in the vault
- AIM Knowledge Hub generation → maps the content into the domain ontology, creates pages (FAQ, article, etc.)
- Schema injection + GitHub sync → as soon as it’s on GitHub, it’s on the web
Lou: “Anytime we want to make that process part of what we educate the market upon, we click that skill and the whole process begins.”
The Gears side of this: YouTube channel scraping and podcast ingestion now work. Give it a channel name, get all content integrated into the knowledge base. Next goal: make it standalone so members can point it at 5 podcasts they follow and get daily research summaries.
Lou’s timeline: “I think by the end of the month, I should have everything pretty much put together.”
Don’s observation in chat: “Gee, it comes all the way back to identifying the problem as the high-value piece.” The cognitive work, not the execution, is what justifies the rate.
Topic 3: Authentic AI Voice — The Lived Experience Ingredient
The session surfaced a precise diagnosis of why AI content fails — and why Lou’s content increasingly doesn’t.
The diagnosis: AI slop has no lived experience. Style prompts give Claude a template. They don’t give it anything to write from. The output sounds like you but has nothing personal to say.
The mechanism: Let AI participate in your actual work sessions. Then, explicitly: “Pay attention to what I just did. Pay attention to why I might have done it that way.” Repeat for months. The model builds up not preferences but a model of your reasoning — your instincts, your analogies, your way of framing things.
Lou: “There was no lived experience. But now, it’s literally writing from an experience I just had. Because it was in that experience with me, and it remembers it… when it has to come up with an analogy or a metaphor or a story to support a point I’m making, it has all of that. It doesn’t make up ‘let’s say this consultant named Sarah.’ It says, yeah, like I remember a time when I was having this problem, here’s how I solved it.”
Don’s question prompted a useful distinction: Claude’s automatic memory is passive — a note-taker that records what happened. The identity/soul file is active — encoding how you reason, applied in every context. Most people only have the first. The practice Lou describes builds the second.
→ Deep Dive: Insight - Authentic AI Voice Is Built on Lived Experience, Not Style Prompts
Topic 4: Focus Discipline — “Who’s Running the Business?”
Lou’s direct challenge to the group, particularly in response to Scott’s data analytics work and the general tendency to chase new tools:
“I’m encouraging us to stay focused on one or two or three key tools, and really double down on getting those tools to do what we need it to do to operationalize. And especially if we can operationalize anything that generates leads or revenues.”
The two lens to use when auditing your AI work:
- Repetitive/procedural — “a smart intern could do 80% of this and leave me 20%.” Those are automation candidates.
- Revenue-producing — not automation for its own sake, but multiplication: more sales cycles, more follow-up, more proposals, without proportional time increase.
Advice to Jamie on the fire-hose feeling: “Pick one thing that’s relevant to what you do, because putting it into action and getting that experience is going to have you handle more of the fire hose. Give yourself a project. When you learn something cool, ask yourself: where does this belong in my business?”
→ Deep Dive: Insight - AI Focus Discipline — Operationalize Revenue Processes Before Exploring New Tools
Topic 5: Jamie — “Get Coached Now or Get Fired Later”
Jamie shared a business pivot: she’s created a short course targeting career change clients, framed around AI’s impact on every career.
The positioning: “Get coached now or get fired later.” Simple, direct, time-sensitive.
The insight: AI displacement creates an entirely new market for career coaching — people who know something is changing but haven’t yet acted on the urgency. The short course serves as a diagnostic lead magnet: where are you in the spectrum of potentially getting AI’d out?
Jamie also shared her use of Perplexity Computer for building domain-specific dashboards. A dashboard covering wage trends and AI career displacement patterns — information that previously required synthesizing 10 professional publications — was built in 6-7 minutes and updates live. “That is GOLD advice” (Elizabeth in chat).
Lou: “It’s a little bit expensive [credits-wise], but yeah, super useful.”
→ Deep Dive: Insight - Get Coached Now or Get Fired Later — Positioning Coaching for the AI Displacement Era
Topic 6: Local AI, Security, and the Open Source Trajectory
Scott’s note about considering a Mac Mini for local models (MLX on Apple Silicon) prompted a broader security discussion:
Lou’s air-gap approach: Rather than isolating the AI, air-gap the data you can’t afford to lose. “I’m trying to air-gap the information I don’t want the agent to see.” Use a VPN-only device for financial and sensitive work; give AI access to the rest.
On open source trajectory: Lou noted Gemma 4, TurboQuant (single-bit/4-bit encodings), and the general trend toward edge-device efficiency. Scott: “They’re gonna be more efficient in 2 years than they are now.” Lou: if you have a good context manager, memory manager, and skill specification, frontier models may not be necessary for most knowledge-worker tasks.
On OpenClaw/Hermes: Lou is cautious. “I’m not ready to outsource control of my desktop to AI.” The locally-controlled ambient architecture gives 80-90% of the same capability without the trust surface area of a third-party agent framework.
Scott: MLX performance on Apple Silicon has doubled open-source model speed. A Mac Mini as a dedicated AI workstation, remote-controlled from the MacBook Pro, may be a reasonable compartmentalization strategy. Elizabeth asked what “air-gap” means; Donald’s answer won the session: “A digital thermos.”
Topic 7: Don — The No-Fault Gap Protocol
Don shared a 4-day deep research project: 7 sub-projects tracing academic culture from 1809 Berlin forward. The purpose: understand the research university system from the inside so he can approach institutional clients without accusation.
The core problem he was solving: his coaching niche (PhDs navigating career transitions) requires institutions to recognize a gap in how they treat early-career researchers — what Don called “the great crime.” But you can’t lead with that framing. The system’s participants see it as normal. Moral accusation triggers defensiveness; no gap opens.
His solution: use AI to do the deep historical research that produces a no-fault framing — the system evolved this way for understandable reasons; the question is where to go from here.
The sequencing principle Don articulated: “Only when you open the gap can you continue to move down to ‘here’s the problem’ — and you cannot introduce a solution until you’ve got a problem, because the solution can’t exist in the absence of a problem.”
He collapsed the research into a white paper — portable context that travels into every future institutional conversation.
The proposal skill comes next: after a discovery conversation, the skill will combine the white paper backdrop, the conversation transcript, and Don’s problem-solving patterns to produce a near-final proposal in minutes. Lou: “You’re going to have closer and closer to a final proposal ready within minutes after your conversation.”
→ Deep Dive: Insight - The No-Fault Gap Protocol — You Cannot Introduce a Solution Until You’ve Opened the Gap
Topic 8: Don — The Onboarding Intelligence Pipeline Goes Live
Don’s second major share: 16 participants onboarded into his new group coaching program, launching the next day.
The assessment battery: MBTI, OCEAN (Big Five), Career Claimer Index (his proprietary instrument), and a standardized 6-question 15-minute interview.
What the skill produced:
- Detailed prescribed coaching notes per participant — not a summary, a coaching plan
- Personalized profile insights per participant — “upbeat, futuristic-looking” — with how the program will help them specifically
These went out individually the day before the first session. The morning-after response: participants emailing “Oh my gosh, how insightful this is.”
The double yield: the same data that produces coaching notes and participant profiles is also VOC and avatar research. Don is building a repository of client language and psychology from every cohort that informs future content and messaging.
Scott observed in chat: “Feeds into training materials as well.” Don took that observation and wove it into how the first session is structured: feeding back participants’ own profile data so they feel seen, heard, recognized — and then moving them from there.
→ Deep Dive: Insight - The Onboarding Intelligence Pipeline — From Multi-Assessment to Personalized Coaching Assets
Community Corner
Don Back launched his 16-person group coaching program — first session the following day at noon. The AI-generated participant profiles hit inboxes before a single session began, and participants responded immediately. A strong signal for what professional onboarding looks like in the AI era.
Jamie W published a short course built around career change + AI displacement. The “Get Coached Now or Get Fired Later” framing was the session’s most quotable moment — Lou named it “Quote of the day” in chat.
Bally Binning helped a client in British Columbia create an AI-powered housekeeping manual — a skill that generates and updates the manual automatically, including what’s been done and what hasn’t. Client reaction: “Oh my gosh, this is gonna save me tons of time.” Bally noting potential for Airbnb hosts as a use case.
Kasimir Hedström has been through the entire member repo and called it a “treasure trove” — specifically the commands folder. Building an AI agent pipeline for LinkedIn content generation (15-20 specialized agents, fully AI-generated posts with human review). Ran out of Claude Pro credits mid-week and ran the equivalent on Gemini to keep momentum.
Donald Kihenja has his domain transfer from Cloudflare almost complete after last session’s discussion. Shared screenshots of his 77 Day Protocol site pages.
Links Shared in Chat
- Ambient AI: The Invisible Harness (Lou) — the 5-part mini-series on ambient intelligence architecture
- The 77 Day Protocol — Energy Audit (Donald Kihenja)
- The 77 Day Protocol — Assessment (Donald Kihenja)
⚡ Try This Before Next Session
Map your business as an ambient folder organization. Take 30 minutes: list your top 5-8 recurring work functions. Name each one as if it were a team (Writing Team, Research Department, Client Onboarding, etc.). For each, ask: if this function had its own agent, what would it need to know, what skills would it use, and what would it hand off? You don’t need to build anything yet. Just get the map on paper. When Lou releases the scaffold, you’ll know exactly what to put in it.
Open Threads
- Lou is finalizing the ambient agent scaffold (copy-paste folder structure) — target: end of month
- Ken collaboration on Gears pricing and access model for members — Lou working out the terms
- YouTube/podcast ingestion integration into Gears coming; tested yesterday, needs to be made standalone
- Scott contemplating Mac Mini + MLX local model setup for compartmentalized AI work — worth watching as a model for privacy-conscious members
- Don’s no-fault institutional proposal skill: first deployment after this discovery conversation
- Jamie’s short course launch — will share traction data in a future session
Next session: Thursday, May 7, 2026
[← Previous: 2026-04-23_Mastermind] · Sessions Index