I built this from scratch — no co-founder, no template, no agency. The idea started from a real frustration with newsletter overload and turned into a working product in six weeks.
View Live Product →The Problem
That was the starting point. Not a market report — a real habit I noticed in myself and confirmed in almost everyone I talked to.
Subscribing to newsletters feels productive. Keeping up with them doesn't. They pile up, subject lines blur together, and eventually you stop opening them — even the good ones.
Existing tools don't solve it. RSS readers just add more sources to manage. Aggregators give you the same generic feed as everyone else. Neither adapts to what you actually care about this week.
People don't need more newsletters. They need one that was actually written for them. YouLetter flips the model — tell it what you care about, it handles everything else.
Research
Spoke with professionals in tech, finance, and healthcare about how they actually consume content. A few things came up in almost every conversation.
Nobody said "I want an AI newsletter." They said "I just want to open my inbox and find something worth reading." The AI is a detail — the value is the output. That realization shaped every UX decision I made.
I prototyped this first. The problem: users still have to manage sources. The burden just shifts from inbox to feed. Not a real fix.
Following what other people find interesting doesn't fix personalization. It creates another popularity contest with a different interface.
More subscriptions is the opposite of what users need. Discovery was never the bottleneck — overwhelm was.
Key Decisions
These are the calls I made during scoping. I've tried to write them the way I'd explain them in a room, not just list what I decided.
Scheduled sends need background workers, which adds infra complexity before I know if anyone wants the output. On-demand lets me answer the real question first: is the newsletter actually good? If users generate it three times without being prompted, then I build automation.
Claude alone doesn't have live news. A search API alone gives you raw results, not readable prose. Together they produce something that feels like a real newsletter — current, well-written, and on topic.
Newsletters live in email. Building a custom reader creates a new surface to maintain and asks users to change a habit they already have. SendGrid handles delivery and I stay focused on content quality.
Trust is the whole product. If users find out mid-use that they couldn't verify sources or didn't know it was AI-generated, I've lost them permanently. Transparency isn't a legal checkbox — it's a retention strategy.
If users are manually generating three or four times, that's the signal to automate. I don't want to build infrastructure for a habit that hasn't been validated yet.
Open rates and click data are meaningful at scale. With fewer than 100 beta users, qualitative feedback will tell me more, faster.
Willingness to pay is validated — that came up clearly in research. But building billing infrastructure before proving retention is backwards. I want 50 beta users with strong NPS first.
YouLetter is a personal utility, not a social platform. Adding sharing mechanics pulls the product toward a completely different set of incentives and design problems.
That's a B2B feature in a product targeting individual users. Adds UI complexity that doesn't serve anyone in the beta cohort.
MVP Scope
I scoped everything around one question: can a user go from signup to a newsletter in their inbox in under 30 seconds? Everything that didn't serve that got cut.
Pick topics. Choose tone. Hit generate. YouLetter pulls current news, writes a full newsletter, and sends it to your inbox. No manual curation. No template to fill out. Just relevant content, fast.
Topic selection, tone preference, and frequency — two-minute onboarding with smart defaults so users aren't starting from blank.
Claude searches for current news via Tavily and writes a 3–4 section newsletter with real sources and links, matched to the user's tone preference.
SendGrid sends a formatted HTML email with AI disclosure and source attribution built in. CAN-SPAM compliant from the start, not patched in later.
Users can update preferences, pause, or unsubscribe. Terms of Service and AI disclaimer are woven into the product, not buried in a footer.
How It's Built
I wasn't trying to build for scale on day one. I was trying to ship something real without betting on infrastructure I hadn't earned yet.
Claude runs about $0.03–0.05 per generation. Tavily adds ~$0.01. SendGrid covers the first 100 sends/day free. Total cost to deliver one newsletter: under $0.06. At $12/month with four newsletters per month, that's north of 98% gross margin before fixed costs. The unit economics work.
User interviews, competitive landscape, stack decision, database schema, API selection and testing.
Authentication, onboarding, topic and tone preference storage, subscription management.
Claude + Tavily integration, prompt engineering, response parsing, HTML newsletter templating.
SendGrid integration, email formatting, AI disclosure, unsubscribe flow, Terms of Service.
QA, beta outreach, user feedback loops, cost optimization, and roadmap updates based on what's actually being used.