The Commoner

Friday, October 2, 2026

Sup, freaks.

There is a lot of money being thrown at AI and still a lot of work to be done figuring out how to use it. Some encouraging developments on that front today, along with a rather disturbing look at what a convincing fake video call could mean for your bitcoin and your family.

WHAT TO EXPECT IN THIS NEWSLETTER

• Why I think people putting AI to work are still very early.
• Tavus's video demo and the bitcoin scams we've been warning about for years.
• Tommy Eastman on agents, Trump on inflating away debt, and France's identity crisis.
• Ben Affleck making a very good case for open models.

Marty's Bent

You're Still Early

Putting together pitches for prospective advertisers is one of the things I have dreaded doing at TFTC. We have the information about our audience and our shows. We know why a particular business would want to get in front of that audience. Getting all of that into a deck has been the frustrating part. As I explained in my conversation with Tommy Eastman, we've been using an agent to help put those pitches together and it has helped us close more ad deals. I find this incredibly encouraging because it is such an ordinary problem to have. There are a lot of people running businesses with work like this sitting in front of them.

As Tommy and I discussed, there are three legs to this stool: the model, the harness and the second brain. At TFTC, we use Hermes as the harness and a second brain we've built from flat files, semantic search and a knowledge graph. The biggest impact for us is coming from that harness and second-brain setup. Too much of the conversation about AI gets stuck on which model is best, while people overlook the systems that make it useful in their actual work.

a16z's latest State of Markets report gives some perspective on how far there is to go. Its Q2 chart shows 69% of S&P 500 companies disclosing a live AI deployment, 29% reporting quantified results and only 2% disclosing a metric they track over time. We are looking at what companies tell investors, which will leave out some of the work happening internally. But even among these very large businesses, a consistently reported measure of what AI is doing for them is still rare.

The amount of money being thrown at the infrastructure can certainly make it seem like the opportunity has already passed. The same report points out that demand for compute is strong enough that even older A100 chips are retaining rental demand as newer hardware arrives. People are already making use of these machines while the applications are still being worked out. If you're learning how to use them, you can benefit from the improvements without having to be the person financing the next data center.

The information about our audience, our shows and how we work is something we can keep building on in that second brain. Flat files give us a place to keep it, semantic search helps find the relevant information, and the knowledge graph connects related information. That gives the agent working through Hermes context it would otherwise be missing. I don't think most people playing around with AI even understand these concepts yet, let alone that they can implement them themselves. If you've already put a harness or a second brain to work, you're considerably further along than the endless stream of model announcements would lead you to believe.

Yesterday I wrote about the idea that AI will save the government from its debt. I remain skeptical. What we're doing with these tools at TFTC is a much more tangible reason for optimism. There is a lot more to using AI than picking a model, and we are still very early in figuring out what these systems can do.

SIGNAL

A familiar face is going to fool a lot of people

This Tavus video demonstration is impressive and disturbing. The expressions and the way the conversation moves are becoming very convincing. Tavus says 26 of 54 participants, told to expect another participant, thought its model was human after a one-minute call. Griffin-Lite is still a restricted research preview.

For those of us holding bitcoin, it is worth revisiting the scenario we discussed with Matt from Unchained in 2023. Somebody impersonates you and convinces your co-signers to authorize a transaction. Having multiple keys does very little for you if the people holding them are all fooled by the same call. Each signer needs to independently verify the authorization, amount and destination.

Then there is the fake family kidnapping call. Imagine hearing somebody you love pleading for help and now being able to see them too. Have the conversation with your family before that happens. Hang up and call a number you already know. A family code word can help, but it cannot be your only check.

Tommy Eastman on putting agents to work

My conversation with Tommy Eastman is a good place to start if today's Bent piqued your interest. We got into the work of giving agents useful information and getting them to do something with it. The model matters, but it doesn't arrive knowing how you run your business.

Tommy described using Hermes to pull up telemetry and build dashboards for his compute clusters. I went through the ad pitches we're building at TFTC. What makes those examples interesting to me is that they start with work somebody already knows needs to get done.

Trump is telling you how they want to deal with the debt

Bitcoin was around $86,425 at 8:07 a.m. Eastern, with Bitcoin Lab reporting $102.7 million of net inflows into U.S. spot ETFs on Thursday. Lawrence Lepard flagged something that makes me considerably more bullish than the morning price action.

In his September 28 interview with TIME, published yesterday, Trump said: “Certain levels of inflation will also pay off that debt very rapidly. Very rapidly.”

The president is openly discussing inflation as a way out of the debt problem. My read is that they would much rather melt up than default. Run the economy hot, debase the currency and hope nominal growth gets them out of the hole. The government gets to repay its debts in dollars that buy less, while the people who worked for those dollars absorb the loss of purchasing power. I think this bodes very well for bitcoin.

This is an identity crisis

Image via Visegrád 24.

A firefighter injured in a barricade fire in Bordeaux, surrounded by people filming. A man who came to put out a fire ends up in the hospital. What a depressing scene.

As I wrote last Friday, I believe the West is going through an existential identity crisis. Importing people with incompatible political and religious norms while refusing to insist on assimilation is a recipe for disaster. A country needs people who want to be part of it and respect its laws and customs. I don't see how this ends well if that expectation is treated as unreasonable.

The Camp of the Saints is becoming a rather uncomfortable book to think about. To my mind, the open-borders worldview promoted by Soros and his allies treats the national identity being lost as something we should be happy to get rid of. I think that is disastrous. I do not see how France is supposed to come out of this with its identity intact.

Ben Affleck gets it

I was very impressed by this clip of Ben Affleck explaining how his AI filmmaking company works. He is talking about unfreezing weights and training open video models on cinematic footage, then letting filmmakers train further on their own productions and keep the proprietary value of that work. He understands this stuff considerably better than a lot of people who spend their days talking about AI.

Netflix's announcement about acquiring InterPositive describes the footage his team shot on a controlled soundstage. They were working on things filmmakers actually need, including missing shots, backgrounds and lighting.

Affleck knows what is wrong with the footage and can explain how his team trained the models to improve it. Access to open models let them put their filmmaking knowledge into the training. They could work on those problems themselves instead of waiting for a closed provider to decide what was worth fixing.

⚡ FREEDOM TECH CORNER

DATA SNAPSHOT

Spot price and block height: October 2, 2026, 8:07 a.m. ET. Bitcoin Lab daily series: October 1, 2026. U.S. spot ETF flows: October 1, 2026.

Sources: Kraken spot quote, mempool.space block height, and Bitcoin Lab daily metrics and ETF flows. Daily values use the provider’s October 1 observation, not live intraday estimates.

The full a16z State of Markets II deck is here if you want to go through the charts this weekend.

Marty Bent · TFTC · Nostr

YouTube

Podcast