Saturday, April 18, 2026
Daily picks
22
articles scored
#1 GOLDAnnouncementAnthropic News
Introducing Claude Design by Anthropic Labs
- Anthropic launched Claude Design — a new Anthropic Labs product that lets you collaborate with Claude to create visual work: prototypes, slides, wireframes, landing pages, and one-pagers, all without leaving the chat
- During onboarding, Claude reads your codebase and design files to build a design system (colors, typography, components) that carries into every future project automatically — no more pasting hex codes into prompts
- You can start from anything: a text description, uploaded images, a PPTX/DOCX file, your live website (via web capture), or your codebase — Claude pulls your actual product's visual language
- Iterate inline: comment on specific elements, edit text directly, or use live parameter knobs Claude builds for you (spacing, color, layout) — then apply changes across the whole design at once
- Export to standalone HTML, PDF, PPTX, or a shareable internal URL; collaboration is organization-scoped with view-only or full-edit access — available now in research preview for Pro, Max, Team, and Enterprise
#2 SILVERReleaseGitHub anthropics/claude-agent-sdk-typescript
[Release] anthropics/claude-agent-sdk-typescript: v0.2.113
- New `sessionStore` option (alpha) on `query()` lets you mirror full session transcripts to external storage — comes with an `InMemorySessionStore` reference implementation and `importSessionToStore()` for migrating existing sessions
- `deleteSession()` method now lets you remove a session from disk or from a custom store — useful for cleanup in automated pipelines
- OpenTelemetry distributed tracing added: the caller's active trace context is forwarded to the CLI subprocess so agent spans appear as children in your existing distributed trace, not orphaned events
- Breaking change: `options.env` now replaces `process.env` entirely (not overlays it) — to add individual vars, use `env: { ...process.env, MY_VAR: 'x' }`
#3 BRONZEGuideDev.to Claude
Opus 4.7 Uses 35% More Tokens Than 4.6. Here's What I'm Doing About It.
- Opus 4.7's new tokenizer uses 33-50% more tokens on the same prompts as 4.6 — prose-heavy content gets hit hardest (up to 47% inflation), code tokenization barely changes
- Pin Claude Code to 4.6 as default with `export ANTHROPIC_DEFAULT_OPUS_MODEL=claude-opus-4-6`, then switch to 4.7 temporarily with `/model opus` only for complex reasoning tasks