I get asked about the tools more than almost anything else.
Which makes sense. When you hear someone describe running a multi-stream operation with a small team, the first instinct is to try to reverse-engineer the machinery. What is this person actually using? What's doing the heavy lifting? And does any of it require skills I don't have?
I'll answer those questions as directly as I can. But I want to start with a frame that I think matters more than any specific tool list.
The biggest mistake I made in the first few months of building with AI was treating the tools as the strategy. I spent a lot of time learning about the tools — how they worked, what they could do, how to chain them together — without spending enough time asking what problem I was actually trying to solve. The tools were interesting. The tools were fun to learn. The tools were also, for a while, a very sophisticated form of procrastination.
The tools do not think strategically. You do. The tools are fast and capable and often impressive. They are also completely indifferent to whether they're helping you build something valuable or helping you spin your wheels very efficiently.
So the honest answer to "what tools are you using" is: the tools are secondary. The work is figuring out what to build and why. The tools are how you build it faster.
With that said — here's what we're actually running.
The core AI layer is Claude and ChatGPT, used for different tasks.
I use Claude (Anthropic's model) for most of my writing-adjacent work — long-form drafting, editorial feedback, research synthesis. The writing quality is high, the reasoning is solid, and it handles nuance better than any other model I've used for that type of work. If you want a model that sounds like it's thinking rather than just outputting, Claude is where I spend most of my time.
ChatGPT (GPT-4 and its variants) I use for structured tasks — data analysis, building out outlines, anything where I need speed on something relatively well-defined. It's also my primary interface when I'm testing something new and want a quick second opinion on whether my reasoning is sound.
I use both because they have genuinely different strengths. Pretending you have to pick one is like pretending you have to pick one tool in a toolbox.
For automation, the backbone is a combination of simple scripting and AI-assisted process design.
I'm not a software engineer. I have enough technical literacy to read code, modify things, and understand what an automation is doing — journalism and law will give you that, if you're paying attention. But I am not writing production software from scratch.
What I've found is that with AI assistance, you don't need to be. You need to be able to describe a process clearly, understand the output you want, and recognize when the result is wrong. The actual code generation and workflow design, I offload almost entirely to AI. The judgment about whether the result is what I actually want — that's mine.
For publishing and distribution: Beehiiv for the newsletter, standard web infrastructure for the directory and other web properties.
Beehiiv is the best newsletter platform I've found for an operation that wants both editorial control and monetization infrastructure without being a developer. The analytics are good, the deliverability is solid, and the upgrade path to paid subscriptions is straightforward when the time comes.
For image generation: Midjourney for creative work, DALL·E for quick functional assets.
I won't go deep on this here — there's a full guide in the free toolkit. The short version is that AI image generation has become genuinely usable for operational purposes. Not fine art. Usable images for newsletters, directories, and digital products without paying a designer for everything.
What I do not have: a large team, a complex proprietary tech stack, or anything that requires a significant technical background to replicate. The systems we're running are available to anyone reading this newsletter. The differentiation is not the tools. It's the judgment applied to the tools.
I want to be direct about that because I think there's a tendency in the AI productivity space to make the stack sound more complicated than it is, as if the complexity is itself a credential. It's not. Simple systems maintained consistently beat complex systems maintained badly, every time. A decade and a half of watching contracts get negotiated taught me that.
There's a full breakdown of the tools — including specific use cases, what we pay, and what we'd replace if we could — in the free toolkit.
This Week in AI: Anthropic and OpenAI both continue to iterate rapidly on their flagship models, with each release bringing meaningful improvements to reasoning and instruction-following. The pace of model improvement right now is fast enough that a tool evaluation you did six months ago may already be out of date. Worth revisiting periodically rather than assuming last year's take still holds.
The full tool stack — every app, what we pay, and how we use it — is in the free toolkit at start.tenstreamslab.com.