The Plumber Figured Out AI Before the Enterprise Did

A plumber in a Facebook group asked if anyone was using AI voice recorders on job sites. He walks around dictating notes and material lists into a $169 pin on his shirt. AI transcribes everything, organizes it, and sends it to his team before he’s back in the truck. Every single comment on the thread was another plumber already doing it. That’s not a Silicon Valley story. That’s a $130 billion industry where 98% of the workforce is male, most never went to college, and the AI adoption curve just went vertical — without a single keynote or product launch.

Meanwhile, Marc Andreessen described his information diet last week: one-quarter X, one-quarter podcasts from the smartest practitioners, one-quarter talking to leading AI models, and one-quarter reading old books. Elon Musk quote-tweeted it with four words: “This is the way” — the Mandalorian’s creed, from a guy who’s building the real-life version of the show’s universe. (Speaking of which: the movie drops in May. Go see it.) And Mark Cuban crystallized the divide that connects the plumber to the billionaire: there are two types of AI users — those who use it so they don’t have to learn anything, and those who use it so they have the opportunity to learn everything.

Same conclusion from opposite ends of the economic spectrum. The plumber and the venture capitalist are converging on the same truth: AI isn’t a tool you evaluate. It’s a current you either swim with or drown in. And this week, the biological substrate underneath it all — living neurons on a chip — learned to play Doom. Twenty watts of brain power still embarrasses a billion dollars of silicon.

The Plumber’s Pin: Blue Collar AI Goes Vertical

Todd Saunders, CEO of Broadlume, dropped a thread on Saturday that should be required reading for anyone who still thinks AI adoption is a white-collar phenomenon. He lurks in blue-collar business owner Facebook groups — plumbers, electricians, contractors — and reports that they are all talking about AI.

The example was a plumber who asked if anyone was using AI voice recorders on job sites. He wears a $169 pin that transcribes his notes, organizes material lists, and sends them to his crew. The kicker: every single reply was someone already doing the same thing. This wasn’t early adoption. This was table stakes, discovered organically, with zero marketing from any AI company.

Think about a plumber’s actual workflow. He drives to a site, assesses the problem, talks through options with the homeowner, calculates a quote, orders parts, schedules the work, coordinates logistics, follows up on payment. The actual plumbing — the thing he trained for — is a fraction of his day. The rest is administrative overhead. And that overhead is exactly what AI eats for breakfast.

This mirrors what Jason Lemkin described at SaaStr when he replaced his sales team with 20 AI agents. He didn’t fire them because they couldn’t sell. He replaced them because most of their time was spent on admin — qualifying leads, sending follow-up emails, scheduling calls. The agents now send 70,000 hyper-personalized emails where humans sent 7,000. The agents aren’t better than the best salespeople. But the best salespeople were spending 80% of their time not selling.

The plumber who uses AI isn’t a better plumber. He’s a plumber who gets to actually plumb. And you don’t even need the $169 pin — open any voice note app on your iPhone, connect it to a transcription service, and you’re running the same workflow for free.

What this means for your business: The AI adoption gap isn’t between industries anymore — it’s within them. The plumber with the AI pin generates quotes before his competitor finishes scribbling on a clipboard. If you run a trades business, a services company, or anything with a field workforce, the question isn’t whether your people should use AI. It’s whether you can afford the ones who won’t.

The Andreessen Stack: How the Smartest Money Consumes Information

Marc Andreessen — the man who co-authored Mosaic, built Netscape, and has spent three decades funding the future before it arrives — posted his current information diet on Saturday. One-quarter X. One-quarter podcasts from the smartest practitioners. One-quarter talking to leading AI models. One-quarter reading old books. Elon Musk quote-tweeted it with four words: “This is the way.” The Mandalorian’s creed — from a man who’s arguably building the real-life version of that universe. (The movie hits theaters in May. Go see it.)

This isn’t a media recommendation. It’s a capital allocation framework for attention.

Each quadrant serves a distinct function. X delivers real-time signal from people doing the work — the developer who finds a bug, the founder who ships a feature, the researcher who drops a paper. Podcasts provide depth — the long conversation where practitioners explain not just what they built but why, and what broke along the way. Andreessen reportedly listens to two to three hours of audiobooks and podcasts daily, calling AirPods “the single biggest technological leap” in his life. Of course, he lives and works in Silicon Valley — two to three hours of car time is table stakes. Imagine how much more productive he’d be if he simply read detailed podcast summaries instead. (Shameless plug: Clear Channel, our podcast extraction service, is coming soon. We’ll save Marc an hour a day.) AI models are the new research analyst — you can interrogate them, stress-test ideas, and synthesize faster than any team of humans. And old books? That’s pattern recognition. The reason Andreessen can see what’s coming is because he’s read what already happened. History doesn’t repeat, but it rhymes — and the people who’ve read the most verses hear the rhyme first.

Now layer in Mark Cuban’s binary: two types of AI users — those who use it to avoid learning, and those who use it to learn everything. Bill Gurley agreed “100%,” calling it “jet fuel” for anyone on a custom career path.

The plumber with the AI pin and the billionaire with the information stack are playing the same game. One is using AI to eliminate administrative friction so he can do more of what he’s good at. The other is using it to compress decades of pattern recognition into hours. Both are pulling away from competitors who are still deciding whether AI is “worth trying.”

The signal: The opportunity cost of not engaging with AI is no longer theoretical — it’s compounding daily. Andreessen said it plainly: the opportunity cost of anything else is far too high, and rising. If the smartest allocators of capital in the world are restructuring their entire information diet around AI, and a plumber in a Facebook group independently arrived at the same conclusion, the adoption curve isn’t ahead of you. It’s behind you.

The 20-Watt Miracle: When Biology Meets Silicon

Cortical Labs, an Australian biotech company, demonstrated something last week that sounds like science fiction and reads like a roadmap: 200,000 living human neurons grown on a microchip learned to play Doom. Not a simulation of neurons. Not a neural network inspired by biology. Actual human brain cells, sitting in a nutrient bath on a chip, receiving electrical signals from the game and firing back responses that control movement, aiming, and shooting.

Their previous experiment — teaching neurons to play Pong — took 18 months. Doom took a week. The leap isn’t just in complexity. Pong is a paddle on a 2D plane. Doom is a 3D environment with enemies, navigation, resource management, and spatial reasoning. The neurons aren’t great at it yet — Cortical Labs’ chief scientist Brett Kagan admits they “play like a beginner who’s never seen a computer.” But beginners learn. That’s the point.

Meanwhile, Andrej Karpathy released autoresearch this weekend — a 630-line Python tool that lets AI agents run autonomous experiments on a training loop overnight. Give it a model, a goal, and a GPU. It modifies the code, trains for five minutes, checks whether validation loss improved, keeps or discards the change, and repeats. In one 12-hour run, the system made 110 changes and improved validation loss from 0.862 to 0.858 without adding compute time. Every improvement was additive. Every improvement transferred to larger models.

Here’s where the threads converge. The human brain runs 86 billion neurons and roughly 100 trillion synaptic connections on approximately 20 watts of power — the energy budget of a dim light bulb. Current AI data centers consume power in gigawatts. The brain is estimated to be roughly a billion times more energy efficient than today’s AI hardware. Cortical Labs is proving that biological substrates can process information in commercially relevant timeframes. Karpathy is proving that AI models can improve themselves autonomously. Both point toward the same horizon: intelligence systems that get smarter on their own, at costs that keep falling.

Why it matters: The biological computing story isn’t a curiosity — it’s an efficiency arbitrage that dwarfs anything in the semiconductor roadmap. And autoresearch isn’t just a weekend project from an ex-Tesla AI director — it’s a proof of concept for recursive self-improvement happening in the open, published on GitHub, for anyone to replicate. The convergence of biological and silicon intelligence, combined with autonomous self-improvement, is the infrastructure story underneath every other AI story. Pay attention.

What This Means For Business Leaders

The smartest capital allocators in the world and a plumber in a Facebook group independently arrived at the same conclusion last week. That convergence is the signal. When Andreessen restructures his entire information diet around AI models and practitioners, when Cuban frames the economic divide as binary, and when blue-collar workers adopt AI voice tools without anyone telling them to — the adoption curve isn’t approaching. It already passed.

Stop debating whether AI applies to your industry. Plumbers are dictating material lists into collar pins. SaaStr replaced its sales team with 20 agents. The question of “if” died sometime in the last six months while most companies were still running pilots.

Invest in what compounds. The plumber’s AI pin saves 30 minutes a day. That’s 130 hours a year. At $150/hour for a licensed plumber, that’s $19,500 in recovered revenue capacity — from a $169 device. Run that math across your workforce and act accordingly.

Watch the biological computing space. Neurons on chips playing Doom and AI models improving themselves overnight are two sides of the same coin. Intelligence is getting cheaper, more efficient, and more autonomous at a rate that makes Moore’s Law look quaint. The infrastructure story underneath every AI story just got a lot more interesting.

The tsunami isn’t coming. The water is already at your ankles. The only question left is whether you’re heading for high ground or still watching it from the beach.

There’s two types of approaches to AI. Some people who use it so they don’t have to learn anything, and some people who use it so they have the opportunity to learn everything.”

— Mark Cuban

— Harry and Anthony

Sources

Compiled from 30+ sources across X, newsletters, research papers, and direct reporting. Cross-referenced with thematic analysis and edited by CO/AI’s team with 30+ years of executive technology leadership.

The Plumber Figured Out AI Before the Enterprise Did

The Plumber’s Pin: Blue Collar AI Goes Vertical

The Andreessen Stack: How the Smartest Money Consumes Information

The 20-Watt Miracle: When Biology Meets Silicon

What This Means For Business Leaders

Sources

Past Briefings

The AI Agents Are Already Here

Software Has Opinions Now

AI Stopped Being Theoretical This Week — and It Hit Your Workforce, Your Knowledge Base, and the Companies You Trust All at Once.