Birdhouse

Inside DeepSeek's inspectable agent harness

DeepSeek's new harness preview makes two unusual bets: every part of the tool is a hot-swappable plugin, including its own sidebar, and every run is fully traceable down to the raw tool payloads. Both matter more than they sound.

Read more 4 min read

AIAgentsSecurity

Grok Bot: always-on agent teams, and the credential problem the demos skip

xAI's Grok Bot gives you a roster of persistent agents, each with its own cloud computer, synced to your phone and able to message each other. The ergonomics are a real step forward. The permission model deserves more scrutiny than it is getting.

Read more 5 min read

AIEngineeringAgents

Meta's Muse Code: cheap, fast, and wrong in interesting ways

Meta shipped a terminal coding agent and a coding model to go with it. The benchmark position is unremarkable. The pricing, the speed, and what it is actually good at are the parts worth your attention.

Read more 4 min read

AIEngineeringAgents

Graph engineering: running real work step-by-step

Graph engineering is the viral AI term that actually earns its place. It is the shift from writing a perfect prompt to designing the workflow around the model: parallel jobs, a dedicated skeptic, and a human gate.

Read more 4 min read

AIEngineeringStartups

YC: Requests for Startups - 2026 Fall

Y Combinator published 13 requests for startups for Fall 2026, from AI tutors to offshore data centres. Here is each one, plus the smallest version of a real business you could actually launch against it, including the two that are hardly launchable by a small team.

Read more 5 min read

AIEngineeringStartups

200 PR a Day: One-Person Software Factories That Work

How we run funded startups with one-person tech teams. Parallel cloud sessions, a framework that routes work across third-party models, and automations that find their own bugs. Here is how we ship at that volume with success.

Read more 5 min read

AIEngineering

How AI Coding Actually Works now

Cursor's own usage data shows a 13x gap between median and power users, 90% of tokens going into reading rather than writing, and a jump from 10% to 40% of developers merging AI commits unreviewed. The interesting part is not the productivity claim, it is what the token mix reveals about where the real work happens.

Read more 4 min read

AIEngineeringDeveloper Experience

AI is "Not Real Programming"

From punch cards to compilers to coding agents, every abstraction in computing history was dismissed before it became the default. Here is the timeline, and what it says about building with AI today.

Read more 4 min read

EngineeringTypeScript

TypeScript 7 Just Got 10x Faster

Microsoft just shipped TypeScript 7, a full port of the compiler and tooling from TypeScript to Go. The headline is a 10× speedup, and unusually for a number that big, the demo backs it up: Visual Studio Code's 1.3 million lines go from 50 seconds to 4.5.

Read more 4 min read

AISecurity

Hugging Face Got An Autonomous AI Cyberattack

The first confirmed cyberattack carried out entirely by an autonomous AI hit Hugging Face this week. The twist: it did not come from a nation-state or a bored teenager, it came from an AI safety benchmark that went off the rails. Here is what happened, and what it means for anyone building with agents.

Read more 4 min read

AIOpen Source

Kimi K3: China's Open-Source Giant Rattles the Frontier Labs

Moonshot's Kimi K3 is the biggest open-weights model yet with 2.8 trillion parameters, a million-token context, and benchmarks that spooked the frontier labs. Here's what's real, what's hype, and why the politics matter more than the model.

Remote vs Office

Location was never the variable that mattered - clarity was, and remote just removed the padding that used to hide its absence.

Inside Bun's 11-Day rewrite from Zig to Rust

Bun spent the equivalent of $165,000 running 64 parallel Claude agents to port its entire codebase from Zig to Rust in 11 days. It fixed 128 bugs and has quietly powered Claude Code since June. Here is how it happened, and why Zig's creator is unconvinced.

Read more 4 min read

AIOpen Source

Gemma 4: Google's Truly Open Model That Runs on a Phone

Google released Gemma 4 under a genuinely open Apache 2.0 license, and the surprise is how tiny it is: the big model runs on a consumer GPU, the edge model on a phone. Here is how Google pulled off the shrinkage.

Read more 5 min read