Subscribe
Share
Share
Embed
Interview me about this project one question at a time. When we finish, write a one page summary that explains the result I want, the format I need, the source material you should use, the people the work is for, the rules you must follow, and the actions that require my approval.
Your experience and years on the job is not a liability when it comes to AI. It's your biggest advantage.
'AI After 50' is the practical podcast for professionals over 50 who want to learn the skills that matter, to build smarter and faster.
Real lessons from the field, from developing and teaching inside a medical manufacturer training facility, to helping businesses across Northeast Ohio work smarter, to building a live media brand from scratch, with all the mistakes in-between.
No coding required. No tech background needed. Just honest, practical skills.
New episodes every Tuesday.
Welcome to Episode #11 of AI After 50. This is the podcast for people who are over 50 and want to understand AI without the hype, without the tech speak, and without feeling like it's already too late to figure it out. It's not too late. We'll prove that today, and we do it with every episode. This week, it's the technical episode, but stay with me.
I promise to keep it simple, and I'll talk about how I use tools and working aids within Claude Cowork to push out and distribute the Lake and Grape Weekly every Friday morning, the files it reads, the instructions it follows, and the connections it uses to reach the outside world. I'll use real examples, including all the mistakes, so stick around for that.
Everything we discuss here on the podcast is available in our show notes, and the link to grab those show notes is down below within the episode description down there. But we also offer what we call premium show notes, and those are show notes that include things we don't cover on the podcast but are very relevant.
So this week, that means a plain English glossary of today's terms, so you can come back to it when one of them slips your mind. And please subscribe to this podcast from whatever platform you're listening on, so you just don't miss an episode. Again, thanks for being here. I really appreciate it wherever you are, and I would like to know where you are and where you're listening from. Email me, mark@marktennant.net, or just leave a comment below. Well, let's start with the big change. So until recently, Claude had three front doors.
There was the chat, that's where you ask a question and get an answer. We're all used to that. There's Claude Code, which is exactly what you think it would be. It writes the code for any project or tool or whatever it is you're building. And then there's Cowork. That's where you hand Claude a job, you step away, and you come back to a finished product or finished work.
I talked about Cowork, uh, actually in the last several episodes, and essentially, Claude now is all three wrapped into one desktop app since Anthropic merged them into a single conversation. There's no more mode to pick. You ask what you need, and if it's a quick question, you, you get a quick answer. If it's a bigger job like research or a report or a spreadsheet or a presentation, Claude takes it on and hands back the finished files.
And here's what I like. You don't have to guess, and you don't have to know which tool to use anymore. Claude just figures that out for you. And the web search feature, there used to be a switch on there where you said, "Okay, search the web." Well, now it doesn't do that. It just decides when search would help and when it would best decide when to use the search feature.
Finally, permissions. Now there's a setting in the message box that controls how much Claude checks with you, and there's two choices. There's auto and manual. Auto means Claude keeps working without stopping at every step, and automated safety checks are run before it acts. Manual, which I have it set to, means Claude asks you before it executes anything every time, and you just tell it yes or no.
The Anthropic page itself says that for work with real consequences like sending messages or changing important files, stay close and review what Claude does.
And I agree with that. I keep a standing instruction that Claude shows me what it's going to change before it edits, deletes, or renames anything. And by default, Claude asks before permanently deleting files.
Now we're gonna get a little bit technical on you. So there's a bunch of files that Claude reads. The first one we're gonna talk about is an MD file, and that's really just a simple markdown file. The .md just stands for markdown. It's an old format. It's simply formatted text that the AI can quickly read and apply to any task.
Here's all it is. It's a plain text file, words. It's the kind of thing you open in Notepad. That's it. No fancy software, no code. So why do MD, Markdown files, why do they matter? Well, because Claude needs to be told things, and it needs somewhere to read them. So think about the last time a new hire came on board, wherever it is you work, maybe your business.
On day one, you didn't really expect them to know how the business runs. You kind of hand them a binder or different things so they learn about it, you know, who the customers are, you know what you do, how you do things, um, what never to do. And the Markdown file is that binder. I also have a Markdown file called the project brief, and the first line says, in effect, "Read this at the start of every session."
Well, what's a project brief? It is what the Lake and Grape Weekly is. It's who it's for, the area we cover, how the newsletter is laid out, and every file, where it lives, and what I've decided and why. So when I start a new conversation, Claude reads the brief, and it's caught up, and I don't have to re-explain a thing.
Then I have reference files, the mastery wine list, every winery in the area with its website and Facebook page, the entertainment list, which is all the musicians who play the wineries with their web and social pages, and a monthly file where I pull events out of several local sources. And here's a story about why all of that matters.
When I made the very first issue, it took about an hour and a half, and it looked great, but a lot of links were wrong, and some were guessed, and you know how AI could be. Some were outdated, and I had to check each one. That's when I created the winery master list, which essentially was just a spreadsheet I gave Claude.
I verified each link once, and now Claude uses the list instead of guessing. And there's even a rule in there, never make up a winery link. Use the ones and only the ones on the list. When AI gets something wrong, just don't fix it once. Fix it in the Markdown file so it's right every time and you don't have to go back to it, and you don't have to worry about it.
Now, you might be thinking, "Doesn't Claude have a memory?" Well, it does, but it can remember things from your chats, and that carries into the bigger jobs too. And it's great, for general things about you. But for exact things like a web address or a rule that must be followed every time, I want it written down in that Markdown file, so I can open it and read it myself and edit it myself if I need to.
I can fix it. Okay, one more thing. When Claude writes you a markdown file, you can edit any piece of it and click Edit with Claude, type what you want changed, and Claude edits it for you right there on the spot. You don't have to describe what each paragraph means. You just have to describe what you want it to do
Okay, the next thing we want to talk about are skills, the rules that are written down. So if the markdown file is the binder, the background, the facts, then the skill is different. The skill is the how. The skill is a saved set of instructions for one kind of a job. You teach Claude how you like something done, and it follow those steps every single time that job comes up.
So if a markdown file is the binder, a skill is the recipe card, if you will, for the dish. And I have two skills for The Lake and Grape Weekly. One is for the newsletter, and one is for my Facebook page and posts.
Let me tell you what's in the newsletter skill because it shows how it works.
It knows who the reader is, and it's a busy woman, twenty-five to sixty-five, who likes wine and plans the weekends for her group or her family. She's always on her phone. She scans before she reads. It knows the mission. We're not here to tell folks about every wine or history of every place. We're here to make their weekend.
It knows the layout, five sections in the same order every week, four to five hundred words total, a three-minute read. It knows the voice, and it's, you know, it writes like a well-connected local. That's me, not a tourist guide. And it has a list of phrases and words it never uses and, like stunning views and don't miss, those types of things.
It has a checklist that gets run before every issue. Now, here's the part I really want you to hear. Nearly every rule in that skill exists because something went wrong. That's true. In the first five issues, the emails went out plain white instead of my navy and cream design that we had went over together.
A setting in Beehive, which is my email provider, was overriding the template it was supposed to go to. So now I have a rule, always use the template. So recently, when Cowork was curating content, an event from Cleveland slipped in. Now, Cleveland is in Northeast Ohio, but it's nowhere near our area of coverage.
So now there's a hard rule, nothing outside of our three counties ever. Another time, the same winery showed up twice in one issue as the top pick and again in the events list. Twice, and it did it two weeks in a row. So now there's a rule, and it's on the checklist that it won't happen again. So you see the pattern here.
A mistake happens. I figure out why it happened, and I just tell Claude, "Write a rule so it doesn't happen again," and it never does. So here's my rule of thumb. If you must correct Claude twice on the same thing, it belongs in a skill, right? One more thing. You don't have to write a skill from scratch. You can just describe what you want, and Claude will help you build it.
Then there are connections for a particular kind of a job. So now let's talk about connections and MCPs. Oh, yikes, MCPs.
Okay, so far everything we've discussed lives inside the conversation: files Claude reads, instructions that it follows. But the real work happens in other places: your email, your calendar, your Google Drive, your publish platform. That's where connectors come in. A connector links one of your apps to Claude, like Google Drive, Gmail, Microsoft 365, Slack.
There's a million different connectors out there. Once it's connected, Claude can even reach into that app while it works on a job on the Claude side. So you might say, "Find last quarter's board deck in my Drive and tell me what changed." Claude goes in and gets it. You don't download anything. You don't upload anything.
You set these up under Customize, then Connections when you're in Claude, and then you decide which one Claude is allowed to use. And another term you'll hear is MCP. Ooh, that stands for Model Context Protocol, but don't let the name scare you. Think of a plug on a lamp. Every lamp, every wall outlet, it's the same shape, and that's what a standard does.
MCP is a standard way for an AI tool and an app to plug into each other. So when you hear someone say an MCP, they usually mean one of the pieces that connects an app to Claude. And you'll see connectors on screen and the MCP technology is underneath, which you don't see. It just, just runs. So if you remember one of the two, remember connectors.
So what do I use? Honestly, I don't use as many as you'd think. I do use the Google Drive connector now and then, but, but not every week. The bigger story is Beehiiv, and that's the platform I publish the newsletter on. And I try to have Claude send the finished newsletter straight into Beehiiv through its API, which is just another way, but don't worry about that.
And it didn't work. So what's the workaround? So Claude uses my Chrome browser through something called Claude in Chrome. It's a plugin, almost like a connector, but it's a plugin in the browser. It opens Beehiiv the same way I would, and it starts and fires up the Lake and Grape template
And then it fills in each section from the draft file with the bold text and the links in place. It's clicking and typing like a person would. It's, it's really fascinating to watch. Every week it blows me away. And once the initial draft is done, then I take over. I review the draft in Beehiiv, and I schedule it, and the newsletter goes out at 6:00 AM on Friday.
That's the pattern. If an app has a connector, Claude can use it. If it doesn't or the connector can't do what you need, Claude can often use a browser and do it the way you would. Now, two cautions. One, only connect things you trust. So Anthropic, the makers of Claude, says, "Carefully assess how much you trust a connector to a website before you give it access." Number two, treat passwords and secret keys like keys to your house. Just don't leave them lying around in places and files that you share with other people. And again, and I can't say this enough, for anything with real consequences, like sending a message as you or publishing, stay close and check the work.
I never let the newsletter go out without my eyes on it and without my final approval.
So let me put all the pieces together for you now. Here's what a week actually looks like. So step one is the research, and on Wednesday, I have Claude look for what's happening in the area for the coming weekend, and we cover Thursday through Monday.
And it checks the multiple websites and dozens of event calendars all over the place, and it checks the individual winery websites and Facebook pages also for the artists as well. And here's a rule I like: Every event must be confirmed from a live source. Claude logs the web address for each one in the notes at the bottom of the draft, and that way, if something looks off, I can check for where it came, you know, I can check for where it came from.
Step two, the overflow file. Claude always finds more than the newsletter can hold. The newsletter's only 4 to 500 words.
So everything that's verified but didn't make the cut goes into this overflow file. Nothing researched gets wasted. I use that file for my Facebook posts during the week.
So step three is the draft. Claude reads the skill, it reads the winery list and the entertainment list, and it writes the issue as a markdown file, a plain text markdown file in the same five sections in the s- every week the same way.
Then it runs the checklist, and is the weekend pick also in the winery list? Is anything outside of our counties? Does every item have a link, and does the link work?
Step four, Beehiiv, my email provider. That's the browser step I told you about earlier. Claude fills in the template, and my editor's notes, the behind-the-scenes sections, it gets left out.
It's just something for me on the side as a guide as I go through and I'm reviewing the newsletter for each week. They never show up in the published issue.
Step five, me. I read the draft in Beehive, and if something doesn't match what I know about the, you know, this corner of Ohio, it's, that's my call and I check it.
And once everything's done, I schedule it and it goes out on Friday at 6:00 AM.
Step six, the overflow goes to work. All week long, I turn that overflow file into Facebook posts. There's three types I do. There's standard posts, event posts, and community spotlights. I tag all the wineries that didn't make the newsletter so they see it and they stay engaged, and then they start knowing who we are.
That's the loop.
So what you can try this week. You don't have to build a newsletter to use any of this. Here are five small things you can try this week with Claude Cowork. Look at Claude in your desktop app or your browser. Do you still see Chat and Cowork? If you do, you're on the old version, and, and that's fine.
And if you don't see them, then you have the new version. Either way, you know now. Two, brief Claude like you'd brief a colleague. Tell it three things. Tell it the outcome you want, the format you want it in, and what it should work from, the files, the links, the apps, the outcome, the format, the inputs.
That alone will improve your results.
Three, write your own binder. Open a conversation and say, "Interview me about my work one question at a time, and then write a one-page summary I can keep." And save that. Next time, hand it to Claude at the start, and you've just made your first markdown file.
Number four, the next time you connect Claude on the same thing twice, stop. Stop and say, "Add this to my instructions so you do it right every time." That's the start of a skill.
Five, try one connector. Google Drive, it's a good place to start, or Microsoft 365. Set Claude to ask before it acts so you can see what it does. Then ask it to find one document and then summarize that document. Small steps at a time. You're not building a system. You're learning how these pieces work together.
And just like I said in episode 10, you don't need the perfect plan. Start with one thing you care about and work out the next question as you go.
So coming up in the next episode, we're gonna move away from the Lake and Grape Weekly, and we're gonna use Microsoft Copilot to solve a real-world business problem I took on at work at my full-time job at Phillips. It's not glamorous, but it's solving a real problem, augmenting and automating the manual steps and processes, and in the end, it will save one of our admins approximately 10 to 12 hours per month.
So 10 hours per month times 12 is 120 hours saved in a year, and that's significant. And we're gonna tell you about that and how we did it and what it was in the next episode, so please don't miss that.
If you've been listening to AI After 50, thanks so much for coming back. If you're new here, subscribe where you're listening from. I really would appreciate it. It would mean the world, and leave a review if you feel like it. Please do.
And if you have a question, email me at mark@marktennant.net. I read and I respond to all of the emails. By the way, come and find me at AIAfter50.com for show notes and resources, and link to the premium show notes in the episode description and subscribe. So until next time, keep building, keep learning, and don't let anybody tell you it's too late to figure this stuff out, 'cause it's not.
Take care.