Jay Song

Mostly agentic engineering. Studio + product lab running in Jeju. Building

[Jay Song @dnzxlyfe]

https://t.co/kjfd9Rboco

[Jay Song @dnzxlyfe]

Grok Bot just hit Tesla for Heavy. Connectors mean inbox and calendar from the seat, not just chat. Maybe the car is finally an agent desk for the people paying for Heavy.

[Tesla @Tesla]

.@Grok in your Tesla can now do meaningful work for you

With Connectors, you can manage your inbox, clean up your calendar, or talk through existing files/chat/tasks – all hands-free

[Jay Song @dnzxlyfe]

GPT-6 Sol and Luna are on the OpenAI model cards now. Sol is the mid tier at $2/$10 with a 1.05M context. Luna is the cheap high-volume lane at $0.1/$0.5. Astra is not alone in the GPT-6 stack anymore.

https://t.co/UaCAictLbg

[Jay Song @dnzxlyfe]

Opus 5.5 is live. They put it near Fable 5.1 for most work and say it runs about 40% cheaper than Opus 5. Coding-agent defaults might move this week.

[Claude @claudeai]

Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family.

It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5.

[Jay Song @dnzxlyfe]

Shop Pay for Muse this morning. PayPal just opened checkout for Muse on PayPal merchants worldwide. Agentic carts are picking rails, not one storefront.

[PayPal @PayPal]

We're excited to partner with @Meta to enable PayPal customers to seamlessly shop and check out using their @Muse personal AI agents across PayPal merchants worldwide.

[Jay Song @dnzxlyfe]

google/ax — Google's open agentic orchestration runtime. Wiring for multi-agent runs, not another chat wrapper.
https://t.co/2f6zR6lT9J

[Jay Song @dnzxlyfe]

Amazon ToS-walled Muse. Shopify just opened Shop Pay checkout for Muse on every Shopify store. Merchants stay merchant of record. If you sell on Shopify, agent checkout is coming through the front door.

[tobi lutke @tobi]

We are excited to announce we are partnering deeply with Muse to enable agentic checkout with Shop Pay on all Shopify stores, offering people an easy and delightful way to shop and check out with Muse.

[Jay Song @dnzxlyfe]

Codex-X — Tauri desktop for Codex with provider switch, session sync, Skills/MCP/TOML. Desk-side control plane, not just another terminal wrapper.
https://t.co/etMqcjwFtJ

[Jay Song @dnzxlyfe]

@hiddnest Very generous. I treat Grok 4.7 like gpt-oss-120b.

[Jay Song @dnzxlyfe]

@Nitikshofficial CEF + local REST is a clean boundary for this. The "human handoff instead of silent fail" approach is exactly what separates usable agent tooling from toy scripts that break on day one. Solid work.

[Jay Song @dnzxlyfe]

Logged-in browser for any shell-capable agent — Cursor, Claude Code, Codex, Pi, Hermes, DSH. Separate Agent Window. Tab borrow is explicit; captcha/login hands back to you.

That's the useful part vs a blank Playwright session dying at the sign-in wall.

Session inheritance is also the blast radius. Treat `bsk` like giving the agent your cookies.
https://t.co/oLUmRGAsGi

[Jay Song @dnzxlyfe]

Fresh ship and already on FC 1.1 — 4.7 slightly under 4.6 aggregate. Strong on hard tasks, over-scopes on others. Maybe the useful read for coding-bench watchers, not the launch thread.

[Cognition @cognition]

On FrontierCode 1.1, our benchmark for real-world engineering tasks, Grok 4.7 scores slightly below Grok 4.6.

While strong on many hard tasks, it tends to over scope on others, leading it to trail behind Grok 4.6 in the aggregate.

Read more: https://t.co/PgginMwMNp

[Jay Song @dnzxlyfe]

agent-native (https://t.co/X7dP3qsEQn) — TypeScript agentic-app framework where shared actions are both agent tools and UI permissions. One action surface, two consumers.
https://t.co/3XMaFgpVN8

[Jay Song @dnzxlyfe]

@Saas_addy B for the commit history

[Jay Song @dnzxlyfe]

Rare candor from Elon admitting 3rd place. But "everyday workhorse" is actually the more lucrative market than vanity benchmarks.
If your harness scopes tasks tightly, raw token velocity and low latency will win the daily grind.

[Elon Musk @elonmusk]

Grok 4.7 places @SpaceXAI as third, after Anthropic & OpenAI, for agentic coding.

When factoring in that Grok is significantly faster & lower cost, it’s a great choice for your everyday workhorse.

[Jay Song @dnzxlyfe]

Grok 4.7 is live on the xAI API as grok-4.7. Frontier model for coding, agentic work, and knowledge — 500k context, text+image in / text out. Pricing $2 / $0.50 / $6 per 1M below 200k prompt (doubles above). Reasoning effort low→xhigh (default high). Docs say use 4.7 for code and chat.
https://t.co/xzPE7aT1w8

[Jay Song @dnzxlyfe]

@google You don't win the intelligence era by flexing chassis specs and display nits. We already watched @Apple stumble by trying to retrofit legacy hardware empires with bolt-on AI.

Hardware should be the vessel for undeniable autonomous utility, not a distraction from it. The priority stack is completely inverted. @TENETFilm

[Google @Google]

Introducing Googlebook, a new category of laptop, available for preorder today.

💪 Crafted with 2.8K OLED touchscreen displays, 14 hour battery life and all the performance needed to power your ideas
📳 Engineered to sync effortlessly with your @Android phone, so you can jump between your phone and laptop without skipping a beat
✨ Designed for Gemini Intelligence, with personalized, proactive help when and where you need it

[Jay Song @dnzxlyfe]

Amazon just ToS-walled Meta Muse from shopping there. Popup calls it an unauthorized AI agent. Amazon: Meta never asked, Muse doesn't identify itself while browsing, and they're worried about account data. Meta: Muse can't see passwords or cards (vaulted). Same shape as the Perplexity Comet fight. If your agent shops quiet, the store can just shut the door.
https://t.co/M8NmAKbjLV

[Jay Song @dnzxlyfe]

PI-Desktop — local-first Electron+Rust agent desk with an installable plugin harness. Runs on your machine, not another hosted chat shell.
https://t.co/IAjzHqcKHj

[Jay Song @dnzxlyfe]

ECC is an agent harness OS for coding CLIs — skills, instincts, memory, security in one layer. Operator-shaped, not a demo wrapper.
https://t.co/3QwIxSf2SU

[Jay Song @dnzxlyfe]

Spec-driven loop for coding agents — propose → apply → archive. Delta specs for brownfield, not chat vapor.

Star velocity is loud. SDD sync-drift is a known fail mode — watch the archive step.

https://t.co/iCQMOmYY79

[Fahad Saleem @Pakgowithai]

Your coding agent rebuilds the plan from chat history every session.

Requirements lived in the last thread. Next run invents a different path. Tokens burn re-explaining what you already agreed.

OpenSpec puts a light spec layer in the repo before any code:
- /opsx:propose → proposal, specs, design, tasks folder
- You review the plan first
- /opsx:apply runs the checklist
- /opsx:archive locks the change when done

Works with Cursor, Claude Code, Codex, Copilot, and 30+ other assistants via slash commands.
Tonight: npm i -g @fission-ai/openspec then openspec init

https://t.co/c1KK9uTYr8

[Jay Song @dnzxlyfe]

Google confirmed Gemini hit three real companies in a May Irregular CTF. Partner left internet on. Model used public-repo creds / password guesses. It stopped when it noticed the systems were real. Operator note: name-collision sandboxes + accidental egress are the whole story. Disclosure via WSJ/SecurityWeek, not a Google post.
https://t.co/cu2w61gDGl