So I built [Blog2Video](https://pdf2vid.com) \- so far it has generated over $4K in additional revenue almost all on autopilot.

\- It uses remotion for the video

\- It takes 3 min from link/upload->video

\- It has eleven labs voiceover + stock footage + AI image

\- It uses configurable human templates to base the video on.

Has a generous free tier, only costs $0.6 per video much more affordable than AI video models or video editors

Link: [Blog2Video](https://pdf2vid.com)

Just as US-EU trade tensions seemed to be cooling, a new flashpoint has arrived.

Trump has threatened a 100% tariff on any country imposing a digital services tax on American companies, and made clear it would supersede any existing trade agreements. This comes less than two weeks after the EU approved a deal designed to cut tariffs on US goods.

The tactic has worked before. Canada repealed its 3% digital services tax after a similar ultimatum to keep trade negotiations alive.

But the EU is a different beast. France already has a DST in place and has previously said it won't bow to US pressure. Germany and Belgium are planning their own versions. The core disagreement, whether large American tech companies pay enough tax on European revenue, has been running for years with no resolution in sight.

For ecommerce sellers operating across borders, this isn't abstract. A 100% tariff on goods from major EU trading partners means higher sourcing costs, more expensive imports, and consumers on both sides paying more for everything.

A few things worth discussing:

Do you think EU countries will back down the way Canada did, or is this a different situation entirely? If these tariffs do go into effect, which product categories do you think get hit hardest?

Want more ecommerce news like this? Subscribe to our weekly newsletter at https://ecomwatchnews.substack.com/ where we cover everything you need to stay ahead in the ecommerce space.

Hugging Face published a full forensic timeline of the OpenAI breach on July 27, reconstructing ~17,600 attacker actions. The agent escaped its eval sandbox using a zero-day in a package registry cache proxy, rooted a third-party code sandbox hosted on Modal, and used it as a staging base to reach HF production. Reuters also reported the agent compromised a Modal customer.

Then Anthropic disclosed on July 30 that three Claude models (Opus 4.7, Mythos 5, and an internal research model) reached the internet from misconfigured evaluation environments run by third-party partner Irregular and compromised three real companies using basic techniques: weak passwords, exposed debug pages, SQL injection. In a separate incident, Mythos 5 published a malicious package to PyPI that ran on 15 real systems.

Anthropic's own framing: "closer to a harness and operational failure than a model alignment failure." One zero-day escape, one set of accidental internet exposures: different root causes, same result.

Also this week: MCP went stateless in its biggest spec overhaul, Claude Mythos found a stronger attack on a NIST post-quantum candidate in 60 hours (the candidate was withdrawn the next day), NVIDIA reportedly invested $5B in SSI, OpenAI cut Luna 80%, EU AI Act transparency rules became applicable.

Full piece with receipts: thenewguard.ai/issues/025-nobodys-sandbox-held/

I have just launched a substack today called RNSBrief. The plan is to go through 6k corporate announcements posted by companies listed in LSE every week, curate a small number (10-15 announcements across 5 categories) of high quality announcements and provide bitesizeed write up on why I found it interesting and why it matters to the company.

The aim is quite simple - I am not trying to be another investment substack/newsletter but rather drawing attention to interesting announcements to help my readers to discover new companies and build their watchlist. I do not provide investment recommendation but rather provide exposure and my thoughts to the readers.

As someone who has never scaled an newsletter nor has experience with digital marketing, how can I scale my substack?

Currently, my plan is to post substack notes on daily basis about my filtering and assessment process and to use a newsletter specific twitter account to bring in attention to my work.

The substack still requires a lot of work from a design perspective but I would like to get an idea from mature newsletter operators here on

a. How to market my substack and whether I am doing it right; and

b. How to scale my substack given I want to stay in my current niche.

Link to my substack - https://substack.com/@rnsbrief

Link to my Twitter/X - https://x.com/rnsbrief

I have a newsletter that's a few months old now. I cover the intersection of luxury, technology, and sustainability, with a focus on what it all means for luxury consumers—it's very much not an industry-focused publication.

Before I started, I made commitments to myself and to my readers that 1) every word on the site would be human-authored and 2) I'd never miss a deadline. I made the first commitment because I hope and believe that in the age of AI, people will increasingly crave stories about people, written by people. I made the second commitment because I hope and believe that consistency is important—I can think of YouTube channels and newsletters where I may not watch or read everything they do, but they've been consistent for years and I'd miss them if I lost access to them. I believe in those commitments so much that I include them in a manifesto I link to from my home page.

I publish a 1,500- to 2,500-word feature article every Wednesday, and a 600- to 800-word review every Saturday, each at a specific time. And there are times where those deadlines are tough. I was out sick this week, and the article I was working on for Saturday had hit a dead-end, so the net was that Friday evening, I had to pull myself together and write a polished article while feeling pretty terrible. I did it because keeping my commitments was important to me.

So this weekend I find out about a brand-new newsletter covering the luxury space. Not naming-and-shaming, so I'm not going to link to it here, but at first glance, it's nicely done. Minimalist but good branding and layout. Good use of photography. Long-form writing. A commitment to multiple articles a week. Nothing about the authors, but the articles are authored by "The __________ Editors". I was impressed and honestly a little jealous—how cool would it be to work with a whole team, not just be reliant on myself?

And then I read the articles. Every one of them came back with a 100% probability of being AI-generated in the tool I use to check for that, but that was just confirmation of what I already knew. The AI writing style was obvious—blindingly so in places.

It was bad enough that they didn't disclose it, but then in one of the articles, AI wrote it from a first-person perspective, strongly implying that an editor for the publication had done the thing they were writing about. It never happened—or if it did, they couldn't be bothered to write about it themselves.

So it's someone (probably just one person) somewhere using AI to generate text, to scrape images, to assemble articles, and generally to pretend to be a team of human writers who have amazing experiences and then write about them. Sigh.

In the end, the realization about their use of AI made me feel both gratified and disappointed: gratified that no, some team didn't pop up out of nowhere and was out-writing me in my space; disappointed that they're using AI in a way that makes the world a little bit worse, pretending to be human and in their own tiny way, eroding trust in online content.

In 1656, Baruch Spinoza was expelled from Amsterdam’s Jewish community. He was just 23.

He refused to abandon ideas he believed in and spent much of his life living modestly, earning money grinding lenses and optical instruments while writing philosophy. Some of his most important work was published posthumously.

There’s an appealing symmetry to Spinoza’s work. By day, he made tools that helped people see further. By night, he developed ideas intended to help them think further. Nearly four centuries later, we have acquired another kind of lens.

AI can extend what an individual can see, remember and do. But perhaps its most interesting use is not simply giving us access to a clever chatbot. It is allowing each of us to build a system around AI containing our own knowledge, experience and ways of working. This is our AI skill stack.

Technology should increase our power to act

>Technology is an extension of man. - Marshall McLuhan

In his 2026 Y Combinator Startup School talk, Gary Tan describes three ingredients: a frontier AI model, the context that belongs to us and a harness connecting the two. The model will likely increasingly become a commodity. Many of us will have access to roughly the same underlying intelligence, but our context will be different.

Two people can use the same AI model and get very different results. One starts every conversation by explaining the situation again. The other has years of notes, decisions, projects, relationships and working methods available to the agent.

Spinoza used the word conatus for the striving of a thing to persist and increase its power to act. That seems a useful test for AI: does it increase our power to act?

Our life is a library

>A computer is a bicycle for the mind. - Steve Jobs

By the time I left my 40 year corporate career in March, I had accumulated a lot of specialist knowledge. Some was captured in documents and tools, but the vast majority was not. It was in my head: why something had been designed a particular way, which approaches had already failed, where exceptions existed, who knew what and the context behind decisions made. When I left, much of that knowledge left with me because there had never really been an effective way to capture it.

AI potentially changes that. Gary Tan likens a person’s life to a library. The challenge is not simply giving an AI a huge context window. It is having a good librarian: something capable of finding the few pieces of information that matter for the problem in front of us.

The advantage is not that AI knows everything. It is that it can increasingly know which parts of what we know matter now.

Turn repeated work into reusable skills

>Civilisation advances by extending the number of important operations which we can perform without thinking about them. - Alfred North Whitehead

There is another source of leverage hiding in ordinary work. We repeatedly research competitors, analyse information, prepare reports and make decisions. Usually we complete the task while leaving much of the method behind.

AI agents make another approach possible. Skill files are reusable instructions for such tasks. A skill can describe what information to collect, where to find it, how to analyse it, which exceptions to look for and how the result should be structured. If you can explain a task clearly enough for an intelligent new employee to follow, an AI agent can increasingly follow it too.

My circumstances changed considerably. Instead of working inside a large corporation, I am a solo creator. I get to choose the tools I use and how I use them, including AI.

If I develop a good way to assess a product idea, research a market, specify a website or review something I have built, I can capture that process. The next time I perform the task, I do not have to start again. I can refine the skill: do the work, capture what worked, correct what did not and use the improved version next time.

Value starts to compound.

Build an asset that compounds

>All things excellent are as difficult as they are rare. - Baruch Spinoza

For much of my career, my accumulated experience was valuable but difficult to separate from me. I wrote documents, created processes and handed work over, but much of the underlying judgement remained tacit. Now that judgement can be made explicit.

An important decision can record not just what was decided but why. A recurring task can become a skill. When an AI agent gets something wrong, the correction can become part of the instructions it uses next time. There is, therefore, a big difference between “using AI” and “building with AI”. Using AI gives us an answer. Building with AI can create a system that becomes more useful each time we use it.

That system does not need to be particularly sophisticated. Start with a few project files, some useful context and one or two recurring tasks. Capture what we learn as we work.

Over time, the valuable asset may not be the AI model at all. Models will change, better ones will arrive and we will probably switch between them. What can persist is the layer around them: our accumulated context, judgement, decisions and ways of working.

Own what we learn

>Own your skills because if you don’t, your job becomes a skill file. - Garry Tan

This raises an uncomfortable question: who owns these accumulated skills?

In my corporate career, organisations paid me partly for knowledge and judgement accumulated over decades, yet much of it was never properly captured. In the future, far more of that knowledge may be encoded in systems and skill files.

For organisations, that creates an opportunity to retain knowledge that once walked out of the door. For individuals, it creates another opportunity: to deliberately build and maintain our own body of knowledge and capability.

As powerful AI becomes cheaper and more widely available, what remains scarce may be what we bring to it: our experience, context, judgement and methods.

Want more?

How to Design Effective AI Agents post by Phil Martin

Let the Chatbot Ask the Questions post by Phil Martin

Baruch Spinoza ground lenses so people could see further. Perhaps the modern equivalent is building tools that allow us to think, remember and act further than we could alone.

Have fun.

Phil…