Evan Tarver

🚨Confession time: I’ve been writing for years. I founded and sold a content business (now 100k+ subscribers), ran growth for B2B SaaS, even published books for fun. I knew the importance of a consistent newsletter first-hand, but even I found it impossible to keep one going. For years, my newsletter sat at the top of my priority list. Scheduled in my calendar with a specific time block. Every week I moved it. Told myself I’d find time the next day (I wouldn’t). So, like anyone does these days, I tried to fix it...with AI. I spent months crafting prompts to write in my voice. Capture the essence of my brand. Got pretty good at it, too. But ChatGPT and Claude are clunky for this kind of work. Multi-prompt nightmares at best. And the apps marketing themselves as AI writing tools? Built for marketers running ad copy through a generator. None of them captured my voice the way the prompts I’d crafted myself did. So I built Flyletter. It’s the AI content team I wish I had when I was running newsletters. A specialist for the strategy, a researcher for the topic, an editor who knows you better than ChatGPT ever will. All guaranteed to capture your brand voice. You’re still the author. The team just does the parts you don’t have time for. If you have a newsletter (or keep meaning to start one), Flyletter is for you. Try out your brand voice for free at Flyletter (link in bio). Would love feedback if you give it a spin!

Do people realize that for every AI writing tell, there is an entire pattern family to avoid? Because I don't see a lot of "Here's the thing:" anymore, but I see a lot of: - Picture this: - Here's the part that matters: - Here's what they're not telling you: - Etc... Any declarative statement (known as throat clearing) with a colon that proceeds it is, in fact, an AI giveaway. Sorry to those who actually use it naturally in their writing (I still miss the em dash tbh)

I built an automated, AI-intelligent CRM with Notion and Claude Cowork, and so can you. It logs leads, moves them through my pipeline, and hands me a daily follow-up report with personalized messages included (based on template sequences and the contact's LinkedIn profile).  It does this all automatically. All you have to do is chat with Claude Cowork. You never have to see a CRM interface again. Total cost? Whatever you're already paying for Claude (I figured out a workaround to do it in Notion for free). This is the process I followed to set it up: > Step 1. Build Your Database: Create a lean Notion database with six fields (Name, Company, Status, Last Contact, Next Action, Source). > Step 2. Define What "Success" Means: Write one blunt sentence describing the exact outcome that counts as a win. Mine is a free trial signup. > Step 3. Map Your Pipeline Stages: List every stage a lead passes through to get to your desired outcome. > Step 4. Set a Follow-Up Cadence: Give Claude actual numbers (mine is 2/3/5 days), because "follow up soon" isn't a rule an automation can run on. > Step 5. Write the Operating Manual: Consolidate your win, stages, cadence, and messaging into one Notion page that Cowork reads as its instructions. > Step 6. Build a Roster: A simple Notion doc listing every lead with a unique ID, so Cowork can query your full database on Notion's free tier. > Step 7. Create a Scheduled Task: A standing order that runs every weekday morning to create a follow-up report based on your cadence. > Step 8. Update by Chatting: Tell Cowork what happened after each interaction, and it updates the CRM for you. Bonus points: Connect Cowork to your meeting notetaker and it can automatically log call results. > Step 9. Run the Weekly Rhythm: Read the morning report, run the outreach, report back to Claude. Once a week, sanity-check the manual and tell Cowork what to change. That's it. You'll never forget to follow up with a lead, and you'll never have to look at a CRM dashboard again. Full step-by-step on how it set it up (Cowork prompts included): evantarver.substack.com/p/ai-crm-setup…

PSA: "Picture this:" is just as bad an AI writing tell as "Here's the thing:"

10/ Building a simple MCP connector using Claude to walk you through the process is easy. The architectural decisions that come after are hard. I created a complete guide on how to build a Claude MCP connector that's actually good, including the pitfalls I had to overcome so you don't have to: evantarver.substack.com/p/build-claude…

9/ Part 6: Verify. Add a custom connector in Claude, paste your MCP URL (add /mcp at the end, easy to miss), and your existing auth prompts the login. Same URL for every user. Then watch a query come back scoped to exactly one user's data.

1/ A customer just used your app without ever opening it. They asked their agent instead. 🧵

For me, I've spent most of my life in writing and content marketing.  Even thought anyone can create a Claude skill or a series of prompts to help them write, almost no one has gone through the decades-long journey of understanding the essence of good writing, or the countless hours I've spent learning how to build AI writing agents. This is my zone of genius, and that's why I built Flyletter.

For example, you could build your own real estate investing app that gives you insights that improve your returns. But a 20 year, multi-million dollar broker with years of domain expertise could build a better product that encodes years of real deals and hard-won mistakes you can't vibe-code in a weekend. Why not pay for access to their product instead of trying to build your own? That's the moat. It's the expertise you've packaged into something an agent can plug into, that would require too much time and effort for a user to do themselves.

Everyone is saying that with AI, distribution is the new moat. But that's not true. It's expertise, and here's why 👇