PowerUp AI Mastermind — June 4, 2026
“Stop fixing the input. The model is not deterministic. Even great prompts produce slop on some runs. What if instead of monitoring that, we turn it into a number from 0 to 1?” — Lou
This Week in 30 Seconds
- Lou’s AIA Quality Gate — the week’s centerpiece: an external evaluation agent trained on 20–50 gold-standard examples, with an autoregressive improvement loop, that any skill can call before publishing. Integrates into the new harness.yaml library architecture.
- Harness architecture — solving the token-tax problem: one central AIA-lib, one harness.yaml per project, only the capabilities you declare get loaded.
- Gears software release — $2,500 perpetual license, source code + content hub tool (Astro + Cloudflare + Gears schema injection), updates through end of 2026.
- Donald’s Codex computer use — GoHighLevel has no API for what he needed. Codex’s computer-use feature navigated the UI anyway. He now has a full inventory of his GHL instance; it’s running a second task as he’s talking.
- Scott’s scraper upgrade — headless browser fallback added, reducing JavaScript failures from 20 to 4 out of 144 contacts; down from 17s to 9s per contact.
- Joanna’s content strategy — moving from daily manual posting into the void to batching with AI, using client VOC + competitive research as inputs.
- Kasimir’s avatar video pipeline — HeyGen API + script generation + auto-upload, working toward 20 serialized videos in a story arc.
- Don Back’s negotiation rehearsal — loaded the Act, bylaws, meeting transcript, and everyone’s position; ran a full negotiating rehearsal with Claude; the real meeting “rolled out exactly as it was role-played.”
- Closing reflection — “Attention is All You Need” as the E=MC² of AI; the gap between mathematics and intelligence is still the most miraculous thing in the room.
1 — The Output Quality Gate: Stop Fixing the Input
Lou opened with a full walkthrough of the quality gate system he’d been building. The starting reframe: we spend enormous effort optimizing prompts, model selection, and memory — all input-side improvements. But the model is non-deterministic. Even great prompts produce slop on some runs, and new model releases reshuffle the behavior you were relying on. The input can never fully solve an output problem.
The system Lou built starts from gold-standard examples: 20–50 pieces of the best content you’d want to produce — your own work, or anyone else’s you’d aspire to match. A command reads through these and derives a scoring rubric: not just grammar and structure, but substance — hook quality, angle, perspective, whether the piece earns its claim. The rubric encodes what makes these pieces worth publishing.
The gate runs downstream of any writing skill. It accepts content + a content-type label (LinkedIn post, newsletter, thought-leader article), scores the output against the matching rubric from 0–1, and returns what failed. It never edits. If the output scores below threshold, the writing skill reruns. The gate gets harder to fool over time — every use and every correction sharpens the discrimination.
Multiple rubrics for multiple content types. Each content type can have its own 20–50 examples and derived criteria. When the invoking skill passes the type, the evaluator uses the right rubric.
Lou tied this to the broader pipeline aspiration: focus on the conversations where ideas happen, automate everything else. The quality gate is the floor that makes automation safe to trust.
💡 What This Means for You
The rubric encodes your taste, not best-practice defaults. Collect 20 pieces of content you’d want to be known for. That collection is more instructive than any prompt about your voice.
Deep Dive: Insight - The Output Quality Gate — Train a Rubric, Not a Prompt
2 — The Harness Architecture: One Library, Per-Project Declaration
Connected to the quality gate, Lou described the AIA-lib and harness.yaml pattern he’s been developing to solve the token-tax problem. The issue: as a skills library grows, naïvely embedding everything in a global CLAUDE.md costs 10–15K tokens on every inference — paid whether or not the project needs those capabilities.
The solution: one central library (AIA-lib, version-controlled on GitHub) holds all skills, commands, resources, and agents. Nothing loads automatically. Each project folder contains a harness.yaml that declares only the capabilities it needs. The global CLAUDE.md holds one thing: a pointer to where the library lives.
Lou’s mental model: a USB hub — one cable in, ten ports on top, plug in only what this task requires.
The eval loop agent itself lives in AIA-lib as an ambient folder with its own intelligence — an agent that any skill can inherit. The harness.yaml in a writing project says “I need brand-voice, I need the eval gate.” It doesn’t pay for the coding skills, the intake skills, or anything else.
💡 What This Means for You
The token cost stays flat regardless of how large your library grows. Every improvement to a shared skill in the library is immediately available to all projects that declare it.
Deep Dive: Insight - The Harness Architecture — Declare Only What You Need, Import the Rest
3 — Gears Software Release
Lou announced the Gears source code release. After working through the licensing questions — respecting Ken’s IP, keeping the product useful, avoiding royalty complexity — he settled on a straightforward arrangement: $2,500, perpetual license, no resell restriction on usage (only no redistribution of the code itself), copyright stays with Lou. Everything on GitHub, updated through end of 2026.
The offer includes more than Gears: a content hub tool built on Astro + Cloudflare Pages, connected to Gears for automatic schema injection. The pipeline: idea → Claude conversation → articles extracted → formatted in HTML → auto-published to Cloudflare site, with Gears schema woven in automatically. “You won’t have to worry about that” was Lou’s summary for anyone who’s been putting off the schema work.
Members who are interested can message Lou directly for the GitHub access link.
4 — Donald’s Codex Computer Use: When There’s No API, Click Around
Donald opened with a discovery that landed clean: he doesn’t need APIs anymore.
GoHighLevel doesn’t expose the API endpoint he needed. Claude told him that, and normally that would be the end of it. But Donald had watched a video about Codex’s computer-use feature — a capability where the model opens its own Chrome session, navigates to the platform, and clicks through menus as a human would. He handed the task to Codex and watched.
“It’s clicking around, even the deepest menus. I felt like: now I really have an assistant.”
His current task while on the call: a full inventory of his GHL instance — all tags, all URLs, all APIs. He can’t remember what he’s built over the years, and Claude kept hitting dead ends without that context. Codex is building the map.
The parallel execution model: While Codex navigates GHL, Donald runs Claude for the next planning step. Neither model burns idle while the other works. He described it as feeling like a team of three.
Lou’s note on Claude’s equivalent: Claude Desktop has computer use too, enabled in settings. He’d seen it try to “open Claude to find a conversation” — which led to Claude trying to use itself — “and it kept saying, I can’t see the application. I said, no, you are the application.” Funny, but the capability is real. His hesitation, post-malware, is about giving anything broad access to his screen. His resolution: a dedicated sandbox machine for computer-use experiments.
Dirk asked whether Donald had trained Codex on GHL. He hadn’t. “It just knows. OpenAI bought a company that was designing this feature for Macs — they added the technology. Almost shocking.”
💡 What This Means for You
“Any excuses about WordPress, GoHighLevel — it will figure it all out, click around, even the deepest menus.” — Donald
Deep Dive: Insight - Computer Use as API Replacement — Navigate Any Platform Without Code
5 — Scott’s Contact Scraper v2: Headless Browser as Fallback
Scott upgraded the contact verification system he built two sessions ago. The original problem: many university websites are heavy JavaScript, which his Python/Mistral 7B scraper couldn’t parse. Results were failing at roughly 20 out of 144 contacts; each failure took 17 seconds.
The fix: a tiered approach. If a standard scrape works, done. If it fails, fall back to a headless browser (Playwright-style), which downloads and renders the page before extracting. Down to 4 unresolvable failures (sites that block everything but human-verified browsers), and processing time dropped to 9 seconds per contact.
The script already ran for June 1st — a hire rolled over at one of the Ontario universities, and it caught it. The admin’s workload: review 4 flagged contacts manually. Previously: 4 days of annual work.
“This is automating the thing that nobody has the time to do. It’s a stupid waste of an executive assistant’s time.”
Scott’s pattern: use cloud-based AI (Claude, Gemini) to develop the Python, then run production work locally with Mistral 7B. Budget-friendly at a not-for-profit where variable API costs are a real constraint.
Lou: “Imagine you did that once a week. By the end of the year, 52 automations — you could save yourself an entire year.”
6 — Joanna’s Wellness Coaching: From Daily Posts to Batched VOC-Driven Content
Joanna is wrapping up her trial — one more session to come. She’s had a significant breakthrough in her in-person work: a client said “you systemize joy” and it has stuck as a frame. She’s leading wellness experiences, building an organic social following, and getting invited back.
Her challenge: posting manually every day is torturous, and she doesn’t feel her hooks are landing. She wants to shift to batching one week’s content at a time using AI.
Lou’s framework: the hook has to come from the audience, not from you. “You systemize joy” is already a hook — it came directly from a client. Feed your transcripts, client calls, and sales calls into Claude, ask it to identify what your people actually say about the problem, and generate hooks from that language. That’s VOC-driven content, and it outperforms anything crafted without it.
Competitive research as a supplement: if your own client library is thin, have Claude research the top performers in your space — their highest-performing YouTube videos, Twitter threads, LinkedIn articles — and use that as a signal for what angles are resonating right now. Combine it with your avatar and what you want to teach, brainstorm 10–20 ideas, pick the ones that feel true, and execute from there.
Lou’s most important tactical note: “If your content isn’t seen by people, it doesn’t matter how many you put out.” His recommendation: one or two pieces per week, put $5–10 behind each to amplify to a targeted audience profile. Facebook as the networking vehicle to find prospects; YouTube as the office where you have the deeper conversation. Don’t build your library on a platform you don’t own.
Don Back’s tip from the chat: “Ask Claude to analyze your videos and give feedback toward your intended purpose and objectives.” Scott’s tip: “Try Claude Haiku first for this — see what it feeds back before going to Opus.”
7 — Don Back: The Most Mind-Blowing AI Feature Is That It Helps You Think
Joanna’s closing question — “what has been the most groundbreaking, breathtaking AI innovation you’ve seen?” — produced the session’s best mini-thread.
Donald: “It’s not a feature. What AI enables me to do is think. People think AI makes you not think. For me, it makes me think more and wider. Thinking companion. Stretches your mind.”
Don Back turned it into a story. He’s co-chair of a not-for-profit board facing a governance issue that has been festering for three years — covered by the Act, covered by the bylaws, should be settled. He recorded the last meeting where everyone stated their position. He loaded the transcript, the Act, the bylaws, his objective, his negotiating strategy, and seven proposed actions (only one of which he actually wanted), and ran the whole thing through Claude in back-and-forth debate mode — testing every position, auditing every argument, tuning the strategy.
Result: he walked into the executive committee meeting last night, and “it rolled out exactly as it was role-played by Claude.” He’ll never tell the other board members how he prepared.
“It wasn’t telling me the truth. But it was enabling me to think it through and get better and better at honing at what I wanted to achieve.”
Kasimir: the default AI mode is search-and-summarize. The higher use is as a thinking partner — but only if you break the sycophancy first. His technique: give AI personas instructions that they will not change their position without data. “If you are not critical, you are the next Leonardo da Vinci, Marconi, and someone else all put together.” The data-gate forces you to actually reason rather than assert.
Lou: for him, the miracle is architectural. All the reasoning, the creativity, the problem-solving — it reduces to matrices and mathematics trying to predict the next word. “How that translates into solving mathematical problems, reasoning, deep thinking, creative thinking, strategizing — to me that is just magic.” Attention Is All You Need, now twelve years old, is the E=MC² of AI. And it’s serving tens of millions of users simultaneously, each getting a considered answer in seconds.
💡 What This Means for You
Use AI more as a thinking partner. Give it the context of your situation, not just the question. Let it push back. The thinking that emerges from friction is the kind you wouldn’t have done alone.
Deep Dives:
- Insight - AI as Negotiation Partner — Role-Play the Deal Before It Happens
- Insight - Data-Gated Pushback — Break AI Sycophancy by Requiring Evidence
8 — Kasimir’s Avatar Video Pipeline
Kasimir has been experimenting with HeyGen’s API to build a serialized avatar video production pipeline. He’s created 3 custom characters, written 20 scripts forming a story arc, and built a workflow: generate scripts → upload to HeyGen agent → the agent handles pauses, emphasis, and captions automatically → download the rendered video → publish to YouTube when ready.
The plan: pay for a HeyGen account for one month, batch-produce all 20 videos, download them all, then release them on a schedule. The AI does the repetitive production; he does the story design and narrative arc. “Without AI, it would have been very difficult to create that.”
He noted this connects to his existing content pipeline: the same workflow that generates blog posts and LinkedIn ideas can be extended to feed scripts to HeyGen. He also has a quality gate — a virtual board of advisors scores each post, and when it clears the threshold, he reviews it manually before publishing.
Lou: “If you ever put that into action, I’d love it if you’d share the workflow. I’ve seen people do this — I haven’t tried it myself.”
9 — Bally’s Tony Robbins Project + Local LLM Tool
Bally has been working with Jarek Robbins (Tony Robbins’ son) in Puerto Rico, leading a project involving 200+ AI agents, developers, and business leaders. She described discussions around MCP linking and being able to quickly orient because of this community’s conversations. “It didn’t take me long to just ask Claude and give me all the steps.”
She shared a tool from an Australian developer she met at the event: 123chat.ai — a free local LLM platform that runs on device, saves tokens, and lets coaches work with client data without sending it to cloud servers. Useful for anonymizing sensitive conversations while still getting AI analysis.
Congratulations from the group to Bally on the work with the Robbins organization.
10 — Security Note: Malware and Computer Use
Lou shared two personal security incidents this week. First: someone approached his 90-year-old mother’s front yard, distracted her, and stole a gold necklace she’d owned for decades. “I don’t know who has the sociopathic tendency to do that to a 90-year-old lady, but apparently it’s a thing.” Warning to everyone: if a car pulls up and someone gestures you to come closer, maintain your distance.
Second: malware discovered on his machine — a comet-icon process that had been accessing all his drives and network for an unknown number of days. Cleaned, hopefully resolved, but the lesson for everyone: keep virus detection installed and up to date. “With AI now, everybody’s accessible — quickly and immediately.”
His planned response: an air-gap machine dedicated only to financial work, with isolated email accounts, all on two-factor authentication.
Scott’s helpful note: VLANs on a managed switch can isolate machines on the same physical network if you want compartmentalization without separate hardware.
Community Corner
- Joanna J is wrapping up her trial but continuing next week. Has had a momentum shift in her in-person work — being invited to lead wellness experiences, landing her first client organically from social. “You systemize joy” — the positioning is coming from the field.
- Donald Kihenja is running Codex computer-use tasks in parallel with Claude while on the call. Literally doing two things at once.
- Bally Binning was named as lead on a project with Jarek Robbins (Tony Robbins’ son) — 200+ AI agents in Puerto Rico.
- Don Back dropped the negotiation story quietly, almost as an aside, and it was the session’s most grounded AI application story.
- Mazie Zdanowicz is 2 degrees from Tony Robbins through Bally (Bally knows Jarek directly).
- Elizabeth Stief noted she’s fallen behind on the last few sessions — “things shown seem so complicated and time-intensive to try out.” Scott’s reassurance: it starts with one specific task that needed doing anyway. His scraper was just that.
Links Shared in Chat
- here.now (Lou’s asset repository): https://brave-quiche-rax4.here.now/
- 123chat.ai (Bally — local LLM tool, free, privacy-preserving): https://123chat.ai/
Try This Before Next Session
Pick one AI interaction you repeat weekly — something you could theoretically automate but haven’t. It doesn’t have to be technical. It could be:
- A content type you produce regularly (a proposal, a report, a follow-up email)
- A research task you run on a schedule
- A review pass over AI-generated output before you send it
Then: collect 5–10 examples of that output you’re proud of. Ask Claude: “Read these examples. What distinguishes the best ones? Generate a scoring rubric I can use to evaluate future versions.” That rubric is the seed of your quality gate.
Open Threads
- Gears release — members interested in the $2,500 source code offer should reach out to Lou directly. Updates through end of December 2026 via GitHub.
- Donald’s Codex computer-use demo — Dirk and Scott both asked to see a live walkthrough in a future session. Donald and Scott both agreed to demo their respective workflows.
- Scott’s scraper walkthrough — Scott confirmed he can walk through his headless browser upgrade in a future session.
- Kasimir’s avatar video pipeline — Lou asked for a demo when it’s running. Kasimir will share when the first batch is done.
- 123chat.ai — Bally will share more from her conversation with the Australian developer.
- AI Focus Discipline — Lou’s post-malware air-gap philosophy and his thinking on when to stay in your current ecosystem vs. explore new tools remained an open thread.
Next session: 2026-06-11