Anthropic just released Anthropic Academy. And it's completely free. 13 self-paced courses covering everything from Claude 101 to building with the API, Model Context Protocol (MCP), and enterprise cloud deployments. Three learning tracks: AI Fluency - for beginners or anyone using AI day-to-day Product Training - hands-on with Claude's tools Developer Deep-Dives - API, MCP servers, Claude Code No credit card, trial period, or upsell (I didn't see one). You sign up with an email and get access to everything (including certificates you can add to LinkedIn for my AI agency homies). There's even a Claude Code course, so you can finally stop watching your 5th "Claude Code Full Course (X Hours)" video and get the how-to guide from the people who actually built it. Whether you're brand new to AI or you've been using Claude daily, this is worth your time. In my opinion, Claude is the #1 LLM on the market right now, and learning it from the source is a no-brainer. Check it out here: https://anthropic.skilljar.com/

It's cheaper to have AI build you your software rather than pay for it yourself. Let me explain. We live in a world where one of the things AI is best at right now is coding. It can't do a lot of the enterprise-grade stuff by default (security, data privacy, things like that tend to require more technical know-how), but it can most certainly set up basic scheduling software, booking systems, or CRMs. That could mean using vibe coding platforms like Lovable or Bolt, or going direct with OpenAI's Codex or Anthropic's Claude Code. The real implication here is that you no longer have to sit through the pain of broken software or a missing integration, because you can actually just build the integration yourself. Or build the custom piece of software to close that gap. For example, if you've ever thought "I wish I had a proposal generator that worked in this exact way based on how my contracts are written" but nothing like that exists, you can describe the exact feeling and function you're missing to an AI, and it will build it for you. You might need to tweak it a little to make sure everything works correctly (that's normal), but hopefully you get my point.

GPT 5.4 dropped last week, and it felt like the right time to talk about something I've been thinking about: how much more we can actually trust AI agents to finish a task now, especially with the newest models. The real power of these models right now isn't in a chat window. It's in setting them up inside your terminal or an IDE, where they get actual freedom to do things they couldn't do before (file access, running code, hitting APIs). That's where the magic happens. MCP server hype has died down, but I think it is a great time to talk about them again. If you don’t know what that is, it is basically a custom server an AI agent can search to access all of the API endpoints of specific software. Here's the thing, you don't really need one handed to you anymore. Writing custom instructions for how to use a specific software platform's API lets you basically build your own. You're just treating each platform like a tool the agent can reach for when it needs to. Now, safeguards still matter. You probably don't want your agent taking out a $5 million loan on your behalf (clearly). But there's a huge range of tasks where it can get to 90, maybe even 95% done without you touching anything. And that last 5-10%? That's where you step in for the finishing touches instead of doing the whole thing from scratch. TLDR: I am realizing I can use claude code for more shit.

When using AI, it's really important that we get good at prompt engineering, which is a fancy word for “giving instructions in a clear way”. The big issues I see with this are two things. One, we're actually pretty terrible at communicating. And two, we don't actually know what we want the agent to do. The first one is self-explanatory. We use big, vague words that don't clearly state what we want or what makes an output good versus bad. I'm guilty of this all the time, but honestly, it's a lot of work to write a detailed prompt (especially when all you want is to find the cheapest place to buy groceries). The second one is more important though. I've caught myself wanting AI to write good copy, but then I realize I've never actually defined what "good copy" means. I don't have enough experience to even understand the nuances of what makes a strong foundation for a good piece versus a bad one. I'm using copywriting as the example here because marketing and lead gen is where I'm struggling in business right now, but this applies to anything. If you're having trouble with writing and you try to get AI to write something, of course it's not going to understand what makes it sound human versus not human. You haven't taken the time to explain it. Same thing with AI-generated images and video. We know it can produce good results, but it also produces bad results a lot of the time because the defaults are things like waxy skin and unnatural body compositions. The issue is that we haven't told it the specifics (things like "use a light skin tone with some blemishes and natural imperfections"). These models are exceptional at what they can do. But a lot of the time when we don't get the output we want, it's user error, not a model problem. TLDR: Bad AI outputs are a skill issue.

A post I put up last week got me thinking about this. Too many people don't realize how much of what they do every day can be automated. AI is really good at a few things right now, and one of them is making yes or no decisions. That sounds simple, but I think it's more useful than most people give it credit for. For example, AI might struggle with something broad like "sell this course," but it's surprisingly good at something like "qualify this lead." The difference is that qualifying a lead can be broken down into a checklist of binary questions (does this person match the criteria, did they fill out the form, are they in the right location, etc.). Selling a course has way more gray area (nuances like timing, tone, etc.), and that's where AI still gets shaky. So here's a challenge: Think about something you do regularly that you haven't automated yet. Try breaking it down into a yes or no checklist. If you can get it to that point, an agent like GPT 5 or Claude can probably handle it. You'd be surprised how far it gets. PS. You can then turn the checklist into a custom "skill" for those of your who use Claude Code/equivalent.