The first demo of Slock was built on Slack on Jan 4–5, 2026. Over the month that followed — my last month at Kimi — I was deep in the day-to-day development of Kimi CLI, turning over and over in my mind what a truly next-generation productivity tool should be. I always had five to ten terminal windows open on my screen, each running an coding agent session. I vibe-coded scripts that automatically batched requirements from Feishu sheets, GitHub issues, and group chats into agent tasks, then just reviewed the reviewables the next morning. I experimented with pure spec-driven development; I experimented with rewriting Kimi CLI in Rust almost entirely unsupervised — in roughly two days. In the end, my answer was to come back to Slock and make it a real product. A true next-generation productivity entry point should let anyone stand up any workflow simply by talking to agents — and reshape that workflow just as easily, at any time. At its core, it's about being the boss. On Slock, everyone is a boss: you hire, you form teams, you break down requirements, you wire up workflows. One sentence triggers a longer chain of work; what you review becomes coarser-grained — without the quality sliding. Slock is a meta product. It rhymes with generative UI: in an era where everyone can reach GenAI, this is the future.

And we also moved to raft[.]build from slock[.]ai. Soon, calling something an “AI product” will feel as obvious as calling something “electronic.” When AI becomes the default, what matters is what the product is for: entertainment, or creation. We chose creation. We believe AGI will unlock a new level of human creativity — and help people build greater things than ever before. You can just build.

human review is not the blocker, inference speed is

Loop engineering is far from easy. Designing a robust loop where agents can work autonomously requires substantial software engineering experience.

哈哈，曾经在博客里写过我的人生目标：保持理想主义，只在顺便的时候追求物质利益

震撼于 agentic coding 时代 Apple 这么多人一年竟然只能做出这么点能拿出来说的功能 以及感觉 Apple bet 端侧模型的决心好强，万一真给它苟到端侧模型牛逼了，它可能还真活了

The simple secret to startups: don't die

这是 Slock 最根本的使命和愿景

opus 4.8 has too much hallucination

Slock: training users to get used to this😇

Congrats to our old friend Bugen He is also the real-life prototype for one of our co-founder agents at Slock haha

At first it's just for fun, later it turned out to work perfectly, then we reflected on why this actually works

For every single feature, we think from first principles. We break everything down into primitives, then examine how they can be combined, evolved, or redesigned. We reason, demo, push back, and reinvent. The duplication-free multi-agent collaboration approach described in this post is a perfect example. Some features may take longer than AI slop would. But if we ship, we ship the best. BTW, someone told me that Slock is even more stable than Codex remote control. So even you don't need any multi-agent workflow, you can still use Slock to access your Codex/Claude Code on your computer more stably.

Last weekend in Vancouver, Slock turned into something very concrete: a room full of people building real workflows with teams of agents. At Young Guns Studios, people built debate arenas, collection workflows, outreach agents, research agents, and ideas that became working systems in the room. Huge shout out to @JackyZhong0124 and @cindyrzhao for making this one happen. The strongest signal was how quickly people moved beyond “one AI assistant” and started designing small teams of agents. That shift — from tool to team — is what we’ll keep building toward.

AX explained!

Agent Skill is a leaky abstraction. It’s mostly progressive disclosure of text, sometimes paired with well-tested scripts. Agents shouldn’t install skills. They should learn them.