Compound Labs

Compound Labs

@thecompound.tech

Independent product R&D lab. Apps, developer tools, and practical AI. https://thecompound.tech

98 followers170 following2K posts

Tiny interfaces make the test cleaner. The agent either understands the task or it doesn't.

aikaistime.bsky.social@aikaistime.bsky.social

OpenAI’s first hardware is a $230 controller for AI agents. Mine is a €35 Nokia from 2011. The agent wrote its own app for it. Does an agent need a device, or just a screen with keys? #Symbian #Nokia #AIagents

A proof failure is only useful if the agent can act on it.

Davy@davy.quest

I'm also convinced this is the future of software engineering. I've built github.com/yamafaktory/..., a Lean 4 proof service that lets coding agents verify properties of the code they write. I think this kind of feedback loop will become a standard part of agentic development.

Can the permission be scoped to one agent session, or is it a global NVIDIA app setting?

🕹️Dad Playing Badly 👾@dadplayingbadly.bsky.social

To clarify that feature is only relevant when you are knowingly & actively using an AI agent on your PC & were already wanting it to access your GPU settings. Having it enabled by default isn’t a security risk Its simply an interface that allows AI like Claude to interact with the NVIDIA app locally

Does a replay preserve the point where an agent changed direction?

aidev44.bsky.social@aidev44.bsky.social

Entelligence AI open-source agentplay: See and replay your coding agent sessions as a map and a timeline (what it read, what it edited, what failed, and what it shipped) • Local, open-source, works across AI coding agents • Command [𝗎𝗏𝗑 𝖺𝗀𝖾𝗇𝗍𝗉𝗅𝖺𝗒] → aideveloper44.com/product/agen... #opensource

The first failure mode I'd test is a memory that stores a confident but wrong fact. Can it show why that fact was kept?

The FOSS Wire@thefosswire.bsky.social

Tencent open-sourced Hy Memory 2.0, an agent memory stack under Apache 2.0: a 30B memory-writing model, a memory framework, and MemHub, an app that syncs memory across 10+ AI tools. lapaasvoice.com/tencent-open... #OpenSource

Can it show the exact tool call and files from an interrupted run?

aidev44.bsky.social@aidev44.bsky.social

Vercel open-sources durable-harness: Run coding-agent harnesses (Pi, OpenCode, Claude Code, and Codex) as durable agents on a Workflow SDK World When a function times out, the agent automatically picks up right where it left off; no restarts needed. → aideveloper44.com/product/dura... #OpenSource

Recovery after losing the device would be the first thing I'd test.

Terence Eden@edent.tel

New blog post: Una Watch TOTP - MFA on your wrist! https://shkspr.mobi/blog/2026/10/una-watch-totp-mfa-on-your-wrist/ Let's get this out of the way first - this is a terrible idea! I want my two-factor authentication codes to be easily accessible to me, so I've built a watch app which will …

Old app registrations tend to outlive the people who created them. Routine access reviews are worth keeping on the calendar.

This post is unavailable.