👋 Hey, I’m Jaryd. Each week, I share one stealable product idea or growth play. With evidence from winning companies who shipped it, and proof from me trying it in our build-in-public lab. This is our classic 5-Bit Friday series—the 5-minute Friday AM roundup with 5 important ideas to noodle on over the weekend.
Happy Friday guys — most of you will know all about 5-Bit Fridays.
I used to write them at How They Grow. They were a hit alongside the main series, so I’m brining back a fan favorite.
No need for an intro you know what to expect: 5 quick ideas and why they matter to you.
Some bit-size info to think about as you head into Friday and the weekend.
No need to dilly dally, let’s get going.
—Jaryd
"Better product" isn't a strategy anymore
I’ve seen great products get no users, and shitty AI-slop products make tons of money. The idea of product as a moat is weaker than ever, because distribution beats craft when everyone can ship craft.
Why it matters: Monthly new paid app launches went from ~2,000 in Jan 2022 to 14,700+ by Jan 2026. A 7x supply jump. Obviously revenue hasn't followed, and the products launched before 2020 still generate 70% of all subscription revenue. Apps launched in the "vibe coding era"—like mine— account for just 3%. Only 17.3% of newly launched apps reach $1K MRR within two years. The standing players have compounding advantages: audience, brand, retention data, acquisition infrastructure. If you're shipping a new product this year, you need a genuine distribution edge or an underserved category, not just cleaner UI or better product.
Applying it: Before your next launch, write down what room (channel, forum, literal place) the potential people you’d like to sell are in, and why you going to be able to get them to try you with one insane offer they can’t say no to.
Go deeper: source
RevenueCat - I use them to power purchases on my apps and run tests. They’re also the app revenue engine for OpenAI & Notion.
Longer trials convert 70% better, but apps keep shortening them anyway
Trial-to-paid conversion for trials of 17–32 days: 42.5%.
For trials of ≤4 days: 25.5%.
That’s a 70% lift from running a longer trial. If you're running short trials to avoid passive cancellations, you're solving the wrong problem.
Why it matters: Longer trials crush short ones on conversion, yet short trials (less than 4 days) grew from 42% to 46.5% of all trials year-over-year. Operators are shortening trials to reduce passive churn from users who forget to cancel, but trading conversion quality for billing cleanliness is brutal math. 55.4% of cancellations on a 3-day trial happen on Day 0 — so a longer trial doesn't actually expose you to more cancellation risk, it just gives the people who want the product more time to feel it.
Applying it: Run a split test on your trial length. Take whatever you're at today and double it for half your cohort. Measure trial-to-paid at D30. If the data says what RevenueCat's 115K-app dataset says, you'll find money on the table.
Go deeper: source
beehiiv - my all-in-one powerhouse for running a $10k/m newsletter. I moved from Substack for brand tools/ways to grow/ad network/MCP/and more. FWIW, a paid newsletter is the ultimate side hustle.
Anthropic just split Claude into two wallets and it's a preview of how every AI product will price
Interactive (you chatting to it) and agentic usage (APIs, runs, automations ) are two different products entirely. Price them like it, and be prepared for your personal costs changing.
Why it matters: On June 15, any Claude subscriber who runs automated workflows through the Agent SDK, scripts claude -p commands, or depends on Claude Code in a continuous integration pipeline will wake up to a fundamentally different billing reality. Starting June 15, 2026, those workloads will no longer draw from the same subscription pool as interactive chat. In other words: interactive chat stays flat-rate, automated/agent usage moves to metered credits Agents munch through compute at a rate that flat-rate subscriptions were never designed to sustain, and companies giving us the compute are feeling it. If you've shipped any feature that loops, plans, or runs in the background, your pricing math is already broken. You just might not know it yet.
Applying it: Audit your top 10 users this weekend. If your power users are running anything that loops (background jobs, scheduled runs, agents), separate that meter from your base plan before they cost you more than they pay you.
Go deeper: source
Athyna - use them to hire top talent worldwide and get work done with humans. I’m using them to help me run sales and subscriber growth
Your pricing page needs a markdown twin or agents won't shortlist you
Check this pricing page out from Resend, and see the markdown file. They claim it’s helping agents become less confused about how their pricing works. This signals a big shift in how companies are thinking about who they are product marketing too—the customer who never visits their site.
Why it matters: Pricing pages were built to seduce humans. Pretty cards, charm pricing, "contact sales.". Agents don't care and are not longer tricked by behavioral pricing moves like charm pricing ($9.99) or the deal effect. Their evaluation will likely be structured and rational based on a set of decision making criteria. Vercel and Resend are already shipping clean, structured markdown versions of their pricing that is designed to be agent-readable, no JS, no marketing fluff. Some gold from Kyle Poyar, “your "contact sales CTA means you're dropped from the agent's shortlist of products altogether”. The robots will absolutely judge your /pricing page.
Applying it: This weekend: publish a plain-markdown version of your pricing page. Try it with https://pricing.md/ with the actual numbers, what each tier includes, what happens at overage, and an FAQ. Link it in your sitemap.
Go deeper: source
100+ Prompt Guide - check out this deep library of great Claude prompts
Supabase just raised $500M, with most users agents, not humans
We’re shifting beyond the need to simply product market for agents with markdown files about ourselves. The agent is becoming, if not already is, the new power user in a lot of apps and we need to think about designing for that.
Why it matters: Supabase closed a $500M Series F at a $10.5B valuation this week (200% + since October) and databases on the platform are up 600% YoY, and Claude Code is the single biggest contributor. Their on-app user isn't a developer anymore. It's an agent acting on behalf of a developer through MCPs. I use Jira and PostHog now more through the MCP than the actual apps themself. Who is the real buyer now?
Applying it: Pick one step in your product and ask: could Claude or Codex do this without a human ever seeing a UI? If yes, start planning the path to exposing it as an API or MCP server. If no, that step is now a liability.
Go deeper: source
ICYMI - here’s the latest edition of the main series from Tuesday, which I’m know considering calling a Byte. Thoughts? 👀
Sweet. That's the five. Reply and tell me which one you liked this week—I read every response and would love to hear your thoughts on what you’d want more or less of.
And if you enjoyed it, consider forwarding this email to one person who's building something. See you next week for the next main idea/byte!
Thanks again for being here and reading today’s Diff.
— Jaryd
