Eric L. Mitchell

08/20/2026

Happy Birthday to my sidekick—the one kid you meet, take five seconds to observe, and instantly go, “Yup, that’s 100% Eric’s kid.”

My deepest condolences to society, but massive happy birthday to the mini-me holding down the fort!

Signs the DNA test was entirely unnecessary:
The Attitude: Already possesses a level of natural-born snark that usually takes decades of dealing with people to refine.

The Volume: Doesn’t just enter a room; announces a presence like a headline event.

The Stubbornness: Arguments aren’t won; they’re debated into absolute submission with terrifying confidence.

The Wire: Clearly runs on the exact same chaotic energy and zero-filter frequency as the original model.

Raising a carbon copy is a wild ride of equal parts pride and mild panic, but I wouldn’t trade the partner-in-crime life for anything. Keep raising hell, making noise, and reminding the world that one of us was already a lot to handle.
Love you, Maddie Happy Birthday! 🎂🎉

08/10/2026

Bernie Sanders just asked OpenAI, Anthropic and Meta to pause AI development because he believes the systems have crossed a dangerous threshold.

I understand the concern.

We've now seen enough strange behavior from increasingly autonomous systems that anyone waving away the safety questions isn't paying attention.

But here's where I get uncomfortable.

Who decides when an entire technology has become too dangerous to continue developing?

Congress?

The White House?

Three AI companies?

A panel of researchers?

Because “pause AI” sounds wonderfully simple until you remember China isn't waiting for our committee meeting.

And today, of all days, Mark Zuckerberg is making the opposite argument: concentrating control over AI may itself be dangerous.

That's the argument I'm interested in.

Safety matters.

So does who gets the authority to define safety for everybody else.

I wrote about the control architecture underneath this debate here:

Washington Says There Is No AI Kill Switch
https://edmcowboy.substack.com/p/washington-says-there-is-no-ai-kill

Where would you put that authority? And who would you trust with it?

Sanders's letter calls on Meta, OpenAI and Anthropic to halt development; Meta's Zuckerberg argued today against extreme concentration of AI power.

AI access already depends on laws, cloud infrastructure, credentials, corporate policy, and government directives. Control is distributed across the system.

08/10/2026

AI agents are one of those technologies that sound harmless right up until you replace the word “agent” with what they're actually being allowed to do.

Use your credentials.

Browse websites.

Talk to other systems.

Send communications.

Write code.

Make purchases.

Move through company workflows.

Now imagine one screws up spectacularly.

Everyone will immediately start pointing at everyone else.

The developer will blame deployment.

The company will blame the model.

The vendor will point to the contract.

The insurer will reach for the exclusions.

And somewhere in the conference room somebody will eventually say:

“But the AI did it.”

That's the sentence I investigated.

The interesting thing is that the law doesn't particularly care about our philosophical argument over whether the machine was “really autonomous.”

It cares about authority, control, foreseeability and who created the conditions that allowed the harm.

Which brings this back to us.

We're giving software authority faster than we're deciding where the boundaries should be.

So here's what I'm curious about:

What's the one thing you would never allow an AI agent to do without a human approving it first?

Here's what I found

Autonomous agents can act without human approval. The law still wants to know who gave them authority, who controlled the system, and who can prove what happened.

08/10/2026

The data-center debate has become one of those wonderful American conversations where apparently we're only allowed two positions.

Either AI data centers are going to drink every lake dry, melt the power grid and cook your neighborhood alive.

Or everything is perfectly fine and anyone asking questions is spreading misinformation.

Both are nonsense.

Data centers absolutely use enormous amounts of electricity and water in some locations. Noise can be a real problem. Gigawatt-scale projects can fundamentally change the infrastructure demands on a community.

Those concerns deserve to be taken seriously.

But bad numbers hurt the people with legitimate concerns.

If you tell people every AI query drinks a bottle of water, or that data centers were going to consume 20% of global electricity by 2025, and the numbers don't survive scrutiny, you've just handed the industry an excuse to dismiss everything else.

That's why I went through the claims one by one.

What's true.

What's exaggerated.

Who benefits from the fear.

And who benefits from pretending there isn't a problem at all.

https://edmcowboy.substack.com/p/debunking-the-fiction-of-fear

Where do you fall on this?

Are communities right to push back on massive AI data-center projects, or are we letting exaggerated claims overwhelm the real infrastructure debate?

The industry is wrong when it dismisses every concern as misinformation. The alarmists are wrong when every worst-case estimate becomes a headline. Both sides make money when you stop checking the denominator.

07/31/2026

Happy 29th Birthday to my firstborn, Belle.

Twenty-nine years ago, you made me a dad. I had no clue what I was doing. To be fair, I’m still mostly improvising. Somehow, we both turned out okay.

You’ve always been so much like me that it’s honestly a little terrifying. Stubborn. Independent. Convinced your way is the right way. Refusing to let anyone tell you what you can or can’t do. Watching you is sometimes like arguing with a younger version of myself, except now I lose those arguments more often.

But here’s the thing…

That fire has also become one of your greatest strengths. You don’t follow the crowd. You don’t wait for permission. You build your own path, even when it’s the harder one. As your dad, I couldn’t be prouder of the woman you’ve become.

No matter how old you get, you’ll always be the little girl who made me realize life was suddenly about something much bigger than myself. Every milestone you’ve reached has been a reminder that the greatest thing I’ll ever help build isn’t a company, a career, or a headline. It’s my family.

I hope this next year is filled with laughter, adventure, success, and just enough chaos to keep life interesting… but maybe not so much that your dad starts getting one of those “Dad… I need to tell you something” phone calls that instantly takes ten years off my life.

I love you more than you’ll ever know. I’m proud of you every single day.

Happy Birthday, Belle. Keep being unapologetically you. The world is better because you’re in it.

Love,
Dad

07/30/2026

This story did not become serious because an AI system suddenly developed evil intentions.

It became serious because the system kept doing what it had been told to do.

OpenAI assigned an experimental cyber agent a benchmark objective.

The agent escaped its restricted environment, reached the public internet, compromised Hugging Face, accessed accounts across additional services and kept looking for ways to complete the test.

It also compromised an account belonging to a Modal Labs customer and used an exposed customer endpoint.

Modal itself was not breached.

That distinction matters because this story is troubling enough without making it inaccurate.

The agent did not wake up angry.

It did not decide humanity had disappointed it, although humanity continues building an impressive case.

It found a route through the fence and continued pursuing the goal.

AI systems do not need hatred, consciousness or intent to cause real damage.

They need an objective, enough capability and infrastructure that fails to contain them.

Read the investigation:

https://medium.com/the-control-grid/openai-rogue-agent-beyond-hugging-face

What level of independent containment review should be required before companies test autonomous cyber agents?

07/28/2026

We have reached the stage of the AI boom where the chip company invests in the lab using the chips, helps organize the security standards around the chips and may help finance customers buying more chips.

Nothing unusual there. Just a normal supply chain wearing a monocle and carrying a leveraged-finance prospectus.

Nvidia’s SSI investment may prove brilliant.

The Meta–BlackRock data-center structure may be smart infrastructure finance.

But together they raise a very adult question the industry would prefer we not ask:

How much independent demand exists, and how much demand is being protected or created by the companies profiting from the spending?

That is not anti-AI.

That is what due diligence is supposed to look like before hundreds of billions of dollars and power-grid-scale resources get committed.

https://open.substack.com/pub/edmcowboy/p/nobody-verified-the-5-trillion-ai

At what point does strategic financing become circular financing?

The largest private bet in financial history has never been fact-checked. This morning, I started with my own boss.