We survived the fax machine, the Rolodex, and dial-up internet — we’re not scared of AI. But we are short on time. And that’s exactly why ByteDance just handed us something that matters. They open-sourced DeerFlow, the AI agent system they’ve been running internally at TikTok, and released it free under MIT license. Seventy-two thousand GitHub stars in the first week. I’m not saying this is the move that changes everything, but I’m also not not saying it.

Why This Keeps Coming Up

Every couple months there’s a new “AI breakthrough” that turns out to be a marketing department’s fever dream. I get it. We’re tired. But this is different because it’s not a tool trying to do everything — it’s a framework that does one thing really well, which is break down big messy problems into parallel tasks and let multiple AI agents handle them at the same time. That’s the thing Elon’s team used for some of their stuff. That’s the architecture behind the AI systems people are actually making money with right now.

The reason it matters for us specifically is that we’re the generation that figured out how to make money in the gaps. We did side hustles before there was a word for them. We learned to code in our spare time. We built email lists when nobody was paying attention. DeerFlow is built for exactly that — you set it and it works while you’re sleeping or doing your actual job.

What I Actually Found Out

I spent the better part of a week setting this up and running test cases because I needed to know if the hype was real or just another GitHub star grab. Here’s what I actually found: you give DeerFlow one goal — let’s say “create a competitive analysis for my niche” or “draft a three-month content calendar” or “research and summarize the top fifty industry blogs” — and it breaks that into subtasks automatically. It doesn’t ask you to micromanage. It doesn’t need you to explain every detail.

Then it spins up multiple AI agents running parallel processes. Each one handles a piece of the problem simultaneously instead of sequentially, which means the work gets done in a fraction of the time. I ran a competitor research task that would’ve taken me a solid day of manual work, back-to-back clicking and reading. DeerFlow had a comprehensive report ready in about three hours while I was out. The output wasn’t perfect — you still need to read it and validate — but it was eighty percent there, which is the threshold where it stops being a toy and starts being useful.

The barrier to entry is real but not insurmountable. You need basic command line comfort, you need to set up API keys with whatever LLM you’re routing through (Claude, GPT, whatever), and you need to understand what you’re actually asking for. But if you’ve ever managed a Zapier automation or fiddled with IFTTT, you can figure this out.

What You Can Do With This Today

Start with something that’s costing you real time right now. For me it was competitive research. For someone I know it was building case studies from customer interviews. For another person it’s daily market monitoring. DeerFlow works best on tasks that need parallel processing — things that would normally require you to juggle multiple browser tabs or hire someone to do the legwork.

Clone the repo, follow the setup guide (they’re clear about it), and run a small test first. Don’t start with your biggest, most important project. Use it on something that matters but isn’t make-or-break. That way you figure out the quirks without consequences. Once you’ve got one workflow running reliably, you add another.

The real play here isn’t that it’s free — it’s that you own it. You’re not depending on some company’s API pricing or their terms of service changing next quarter. It runs on your infrastructure. That’s GenX as hell.

I’ll be honest: I don’t think this is the last piece of the puzzle. But it’s a real piece. And right now, that’s enough.

Watch the Full Video

I covered all of this in a short video too — sometimes it’s easier to watch than read.

Watch on TikTok →

I turned 50 last October and I’ve been thinking a lot about the 30 years I spent calling specialists. Someone needed to build a database? Call a database engineer. Need compliance documentation? Hire a lawyer or a compliance officer. Marketing copy? Freelancer or agency. This was just how the world worked, and we accepted the overhead, the wait times, the invoices.

Then I ran into Claude Skills. It’s a GitHub library someone put together—380 of them, just sitting there, free. I’m not usually the guy who gets excited about libraries, but this one actually changed how I’m thinking about what solo makers can do in 2026.

Why This Keeps Coming Up

We’re at a weird moment. AI is everywhere now, but most people still treat it like a search engine with better sentences. You ask it a question, you get an answer, you move on. The stuff that’s actually shifting the game board is when you stop asking and start installing. When AI becomes something that just knows how to do your job.

For 30 years, that required hiring. You wanted someone on your team who understood marketing? You hired a marketing person. You needed engineering standards? You hired an engineer. These weren’t optional—they were the only way to scale work beyond your own two hands. But that assumption is getting old fast.

Claude Skills is just a more obvious version of what’s already happening. Someone took entire job functions and packaged them into files. Engineering practices. Marketing frameworks. Compliance checking. Productivity optimization. Over 30 agents and 70 commands, all of it just sitting in a folder on GitHub waiting for you to copy-paste.

What I Actually Found Out

The thing that got me was how simple it is. I was expecting to learn some new prompting technique or spend a weekend reading documentation. Instead, the actual mechanic is ten seconds. You copy one file into your project folder. That’s it. Claude now knows how to do that entire job function because it’s built into the context.

That’s different from what we’ve been doing with AI for the last couple years. We’ve been asking it questions and hoping it remembers context. This is the opposite—you’re loading the entire framework into memory so it never forgets what it’s supposed to do. The skill knows the patterns, the templates, the standards, everything. You just keep working.

What actually matters here is that this works because the library was built by people who already understand those jobs. Someone didn’t build a “marketing skill” by guessing—they built it by codifying actual marketing work into Claude’s system prompt. That’s why copy-paste works. The heavy thinking is already done.

What You Can Do With This Today

If you’re running anything solo or with a small team, this changes the conversation pretty fast. You want engineering standards applied to your code? Copy the engineering skill. Need your marketing copy checked for brand voice consistency? Copy the marketing skill. Trying to make sure you’re hitting compliance requirements? Copy that skill instead of calling a lawyer for a consult.

I’ve been testing this for a couple weeks and the pattern holds. Where you’d normally hire someone or outsource work or just accept that it won’t get done, you now have a working alternative that costs nothing and takes ten seconds. The skills aren’t replacing specialists entirely—I’m not naive about that—but they’re covering the gap where you’d normally just skip the work because it wasn’t worth hiring for.

The practical version looks like this: you’re building something, you think “we should have someone checking this for X,” and instead of that becoming a line item you never get to, you paste a skill and keep moving. It’s not perfect, but it’s there. It’s free. And it’s 10,000 times better than doing nothing.

This is what our 50s are getting for being patient with technology for 30 years. We get to skip the part where we hire for everything and jump straight to doing it ourselves with better tools. Makes the whole reinvention thing feel less theoretical and more like something we can actually pull off.

Watch the Full Video

I covered all of this in a short video too — sometimes it’s easier to watch than read.

Watch on TikTok →

Look, I’ve been using Obsidian for about three years now. I’ve got thousands of notes scattered across my vault—old client projects, ideas from 2am, random research rabbit holes, meeting notes I swear I’ll organize someday. The problem is exactly what you’re experiencing: I know the stuff is in there somewhere, but finding it? That’s a different story. Last week I heard that the CEO of Obsidian just released some GitHub skills specifically designed to teach Claude Code how to actually read your vault. Not chat about it. Actually read it, understand it, work with it. That changed everything I was thinking about for 2026.

Why This Keeps Coming Up

We’re the generation that actually kept journals. We built filing systems with real intention. We labeled folders, created hierarchies, tried to impose order on chaos. But here’s the honest part: we also did all of that and then the system became so complex that it defeated its own purpose. I’ve got notebooks within notebooks. My vault has subfolders that reference other subfolders. I know intellectually that something useful is in there, but I can’t retrieve it because I can’t remember if it’s under “Projects” or “Research” or “Clients” or some three-level-deep structure I created at 11pm on a Tuesday. Claude Code couldn’t help with this before because it couldn’t actually access my files in a meaningful way. It could talk about what might be in there, but it couldn’t do anything real.

What I Actually Found Out

The Obsidian CEO built five specific skills designed to let Claude Code actually understand your vault structure. I’m talking about Claude reading your actual files, understanding how they relate to each other, and then rewriting them, reorganizing them, pulling information from them. This isn’t theoretical. You can point Claude at your entire vault and it can start working through it like a real assistant who actually knows where everything is. The skills teach Claude how to navigate Obsidian’s system, how to find files based on content, how to understand your linking structure. It’s the difference between describing a filing cabinet to someone versus handing them the keys and a map.

What surprised me is how much this opens up. I started thinking about all those meeting notes from three years ago that probably contain context I need now. Or the random ideas I jotted down that could connect to something I’m building right now. Or the client research I did that I’m half-remembering but can’t quite locate. Claude can find all of that. Not just search for it—actually read it, understand it, surface the relevant pieces.

What You Can Do With This Today

Honestly, start by just setting it up. Go grab the skills from the GitHub repo the CEO built. Get Claude Code connected to your vault. Then give it a simple task—something like “read all my notes from 2024 and tell me what ideas keep appearing” or “find every note that mentions X topic and summarize what I’ve learned.” Let it do actual work on your files. Not organizing everything at once, because that’s paralyzing. Just one real task that shows you what this can actually do.

I’ve been running some experiments this week. Asked Claude to read through my entire project vault and suggest which old ideas might actually be worth revisiting right now. It found connections I’d completely forgotten about. Last night I had it clean up a folder of meeting notes from 2024, pulling out action items and decisions I’d buried in prose. This is the kind of actual grunt work I’ve been avoiding because it requires sitting down and manually reviewing everything. Claude just does it.

For someone our age who’s spent decades building knowledge systems, this feels like finally having someone who actually respects the vault you’ve built instead of treating it like clutter. You’re not starting from scratch with some new tool. Your Obsidian vault is still your vault. Claude just gets permission to actually help you use what’s already there.

Go set it up. You’ll figure out pretty quickly what you can do with it. I did, and I’m still discovering new angles.

Watch the Full Video

I covered all of this in a short video too — sometimes it’s easier to watch than read.

Watch on TikTok →

Two years ago, I convinced myself I needed to try everything. Every AI tool that launched, every beta, every new feature. I’ve got spreadsheets full of sign-ups, browser tabs permanently open, and a credit card that’s seen better days. Last month I counted: I’d paid for 47 different AI services. Forty-seven. Most of them I hadn’t opened in six months. So I did what any reasonable person who’s afraid of wasting money does—I deleted almost everything and kept five. And honestly? That’s when things actually started working.

Why This Keeps Coming Up

GenX guys like us grew up in the “more is better” economy. We got our first computers running Windows 95 and learned that you upgrade, you add features, you keep options open. That logic worked fine when you were buying VCRs. It’s killed my productivity with AI tools. I had writing assistants, research tools, design platforms, coding helpers, marketing automation, image generators, voice tools—the list went on. Every single one promised to save me time. Together, they just fragmented my attention into useless pieces.

The real problem wasn’t the tools themselves. It was the mental overhead of switching between five different logins, five different interfaces, five different ways of prompting the same basic question. I was spending time deciding which tool to use instead of actually using a tool. That’s the trap nobody warns you about when you’re evaluating AI.

What I Actually Found Out

Here’s the first one I kept: an AI writing assistant. I use it to turn professional experience—stuff I already know how to do—into paid work. I was already a decent writer before AI. I could already write consulting documents and freelance deliverables. The writing assistant doesn’t replace that skill. It speeds it up. I’m billing at twice the speed I used to, which means more money in less time. That’s the entire value proposition, and it works.

The second tool is an AI research and summarizer. I tested maybe twelve different research tools. Most of them buried me in information. This one gets me to the answer fast. I used to spend ninety minutes down internet rabbit holes looking for specific data. Now I spend fifteen minutes. That’s not hype—that’s actually what happened when I timed it.

Third is a content repurposing tool. One piece becomes five pieces. A blog post becomes social media posts, an email, a YouTube description, a newsletter section. I’m not hiring a content manager. I’m just making sure the work I already did gets used more than once.

Fourth and fifth are an image generator and a coding assistant. The image generator handles visual concepts I used to either pay someone else to create or go without. The coding assistant catches syntax errors and speeds up my development work by probably thirty percent. Neither one makes me a better coder or designer, but they both save me concrete time on concrete projects.

The pattern I noticed: I kept the tools that made me faster at something I was already doing. I deleted the ones that promised to replace a skill I didn’t have. That’s the distinction nobody makes when they’re selling you AI tools. Most of them are marketed as replacement technology. They’re actually productivity tools.

What You Can Do With This Today

Start with what you actually do for money. What skill generates income for you right now? Now ask whether AI can make you do that thing faster without replacing your ability to do it. If yes, you’ve got a candidate tool. If it’s promising to replace a skill set—to be your accountant or your designer or your programmer when you’re not one—delete the trial. You’ll just end up frustrated when the output isn’t what you wanted.

Pick one tool per skill category and stop looking. This is the hard part because new tools launch every week and they’re all improving. I had to make myself stop. Stick with what works for ninety days. Write it down. Track whether it actually saved you time or made you money. If it did, renew the subscription. If it didn’t, cancel it and stop feeling guilty about wasting the sign-up.

I’m not against AI. I use it every day. I just got tired of the mess of trying to use everything at once. Sometimes the best decision is to commit to less.

Watch the Full Video

I covered all of this in a short video too — sometimes it’s easier to watch than read.

Watch on TikTok →