Five things that mattered this week. Read the full take on each via the linked articles. The newsletter version of this recap goes deeper on each item; subscribe at /newsletter if you want it in your inbox Sunday.

1. Anthropic shipped the Skills Marketplace

The story of the week. Anthropic’s ~/.claude/skills/ registry got a public marketplace (marketplace.anthropic.com/skills). About 600 skills at launch. Most are free; paid skills supported with a 15% Anthropic cut.

Founder-relevant: install three this week — pdf, customer-research, and one paid skill if you want to test the buy-side flow. The PDF skill alone removes a 4-6 hour weekly task for anyone shipping lead magnets.

Read: Anthropic Skills Marketplace just launched: what changes for $500K founders.

2. Claude Code 1.5 — persistent subagents + new Plan mode

Anthropic also shipped Claude Code 1.5 mid-week. Two changes that matter:

  • Persistent subagents. Subagents now keep state across calls within a session. The “build → test → fix” loop no longer re-discovers the codebase 3 times.
  • Plan mode redesign. The new Plan mode produces a structured plan.md that can be reviewed, edited, and approved before execution. Closer to “human-in-the-loop” than “agent-doing-stuff.”

Founder-relevant: if you’ve been hesitant to let Claude Code touch production systems, Plan mode + the new approval flow makes it materially safer. Try it on a small refactor this week.

3. Beehiiv Referrals 2.0

Beehiiv updated the referral program with tier-based rewards (3 / 10 / 25 referrals) and a redesigned referral landing page. Newsletters that activated it in the closed beta saw 20-40% lift on referral-driven sub growth in the first month.

Founder-relevant: if your newsletter is past 1K subs, turn it on this week and rebuild your welcome series email #2 around the referral CTA. Worst case: no lift. Best case: a few hundred extra subs/month at zero ad cost.

Read: Beehiiv review 2026: still the best newsletter platform under $50.

4. OpenAI o5 pricing leaked — and it went up

A pricing card for OpenAI’s upcoming o5 leaked on X mid-week, then got pulled. The numbers: $25/M input tokens, $75/M output tokens. That’s higher than o4-mini ($15/$60), reversing the recent trend of frontier prices dropping each generation.

Founder-relevant: plan budgets accordingly. If you’re per-token-billed on OpenAI, the next generation is more expensive, not less. This makes the case for flat-rate plans (Claude Code Max 5x, Cursor Pro) sharper. If you’re already on flat-rate, ignore.

5. New: the case study pattern across 8 of 10 founders

We published our deep dive on what 10 anonymous solo founders did to reach $500K. Eight share the same 5-step pattern. The single most-broken rule: adding a second product before $200K. Six of ten did it. All six lost 3-6 months.

Founder-relevant: if you’re under $200K and looking at a second product, this is the read of the week. Stop, read, then decide.

Read: From $0 to $500K solo: the pattern 8 of 10 founders share.

New on 500k.io this week

We published five articles this week. The three to read first if you only have 30 minutes:

  1. How to use Claude Code for SEO at scale — the 9-stage content factory pipeline.
  2. 10 best AI tools for solopreneurs in 2026 — refreshed list with two recent removals.
  3. Claude Code vs Cursor in 2026 — decision tree + pricing math.

One thing I’m trying this week

Rebuilding the $500K AI Stack PDF production around the new pdf skill. The current pipeline is a 90-line custom Puppeteer script + a Markdown source. The skill replaces the Puppeteer layer with one call. If it works, I’ll push the revised pipeline as a follow-up article next week.

I’ll report results in next Sunday’s recap. If the migration breaks something I didn’t anticipate, that’ll be its own article — those tend to be the most useful.

Get this recap in your inbox every Sunday. Subscribe →

FAQ

Why a weekly recap when you publish daily?

+

Most founders skim daily. Sunday recaps catch the 3-5 stories that actually matter and the one thing they should try this week. We send the full version via newsletter; this is the public mirror.

How do you pick what makes the recap?

+

Three filters: (1) does it change a workflow? (2) does it move a price? (3) does it expose a new opportunity? If yes to one of three, it gets the slot. Pure announcements that don't pass any filter get cut.