This is so nice to play - tried to re route my paragliding flight's preview that I did today 🪂 Finally Google Earth supports uploading flight records like KML, GeoJSON. I've used several tracking apps in the past but none has such view so you can have irl like previews with best earth model.

City of love and blue, doing its show again

Same

Now exploring @Google's agentic commerce protocol: AP2 (part of UCP) and buying things using @WalletConnect Pay. This time using Gemini CLI as my agent and asking to buy stuff. Flow works very similar to my previous MPP example — our sample e-com app stays as a Checkout Mandate (has an endpoint with the AP2 standard where the agent can explore products), and on the payment side our WC Pay MCP drives the WalletConnect signing flow. Making WC Pay one of the default Payment Handlers under UCP would allow hundreds of wallets like @zerion @rainbowdotme @TrustWallet etc. to onboard into agentic payments directly. Going to explore further to make it possible.

New WWDC vid: watch later New WC vid: OMG

Missing context in my tweet; MPP is co-authored by @tempo and @stripe. Tempo's also been added as a supported chain on WalletConnect Network - curious to see how the WC wallet ecosystem and Tempo connect from here 🤝

I've been looking for some frontend skills on skills.sh - most of them unfortunately AI written and I'm seeing that Inter font has been banned to use. Is this how the 'taste' looks like? Idk man

context-mode v1.0.103 just shipped. 📊 105K users · 10.8K GitHub stars · 745 forks · 14 platforms Your AI coding sessions now have analytics. `ctx insight` — a personal dashboard that reads your session history and tells you what's working, what's not, and what to fix. What it tracks: → 23 event categories. Every file edit, git commit, error, subagent delegation, plan approval, blocker, and CLAUDE.md load — captured automatically. No config needed. → 37 insight patterns. Not vanity metrics. "Your session ended with 7 errors and zero commits — all effort was lost." "CLAUDE.md loaded → error resolution rate jumped to 85%." "Same file edited 12 times — write a spec first." → 4 composite scores (0-100). Productivity, Quality, Delegation, Context Health. One number per dimension. Track weekly. → Error Intelligence. Resolution rate, retry storm detection, P95 latency by tool, top error sources. Not just "you had errors" — "here's which tool fails most and how long it takes to recover." → Delegation analytics. Agent completion rate, parallel burst detection, time saved. "You launched 45 agents, 38 completed, 12 parallel bursts — saved ~76 minutes." → CLAUDE.md correlation. The killer insight: sessions with rules loaded show measurably higher error resolution. Sessions without → errors pile up. This single metric convinced 3 teams to adopt CLAUDE.md this week. The numbers from 29 sessions: Files tracked 391 ██████████████████████████████ Git operations 98 ████████░░░░░░░░░░░░░░░░░░░░░░ Rejected approaches 83 ██████░░░░░░░░░░░░░░░░░░░░░░░░ Errors caught 66 █████░░░░░░░░░░░░░░░░░░░░░░░░░ Latency events 52 ████░░░░░░░░░░░░░░░░░░░░░░░░░░ Decisions tracked 21 ██░░░░░░░░░░░░░░░░░░░░░░░░░░░░ Skills used 15 █░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 1,100 events remembered across 29 sessions — searchable after compact & restart. Also in this release: → Windows hook fixes. `process.execPath` replaces bare `node` — no more MSYS path mangling on D: drives, no more Git Bash PATH failures. All 8 adapters updated. → Node v24 compatibility. `ensure-deps` no longer skips `better-sqlite3` install on modern runtimes. The SIGSEGV-prone probe is skipped — the install isn't. Every insight runs locally. Your data never leaves your machine. Open source. Free tier. No account required. github.com/mksglu/context-mo…

Like to hear thoughts from @lgrammel @nicoalbanese10 if there is any way for mid stream messages

[Meta Ads] Generative UI components are so fun to build. Added support for promoting existing post from Facebook or Instagram. This type of ads are the most common way businesses do because it's easy for them but I think many forget that without right targeting it's waste of money. Now it'll be more efficient with AI to adjust the ad with details like target users, following metrics and make updates for better performance. Showcased this to some companies and hearing great feedback. I think there are a lot to build in the digital marketing area. Will share more soon

This might go out soon

[Meta Ads] Creating AI based ad campaigns achieved internally Been spending time with ads APIs, they are completely cursed - there are thousands of variations of a campaign object, bad/missing documentation, things getting easily out dated. Same for both Google Ads and Meta Ads. Hopefully with Claude Code made deep dive on each APIs and found out basic campaign creation. I love the interactive components in chat, so easy with @aisdk - User confirmations - Interactive data entry (image and account selection) - Remembers choice across chat - Success/error handling - Beautiful by its root; @shadcn components Each component designed as tool to @aisdk for it's purpose. Next step I want to try json-render by @ctatedev, curious how AI come up with dynamic components for different variations of an ad/campaign.

Meta Ads with AI just got enabled for everyone 🔓 Didn't know it was that hard to get API approval from Meta. You literally need to create a video for each permission with detailed instructions, why/where/how you use them etc. Wait around a week for results. So annoying but after a couple of rejections, finally I got it

Focus is on WalletCon today