Building effective agents
Anthropic Engineering
A durable architecture reference for deciding when workflows are enough and when agent autonomy is actually warranted.
A curated external reading layer for AI systems work. Selection is intentionally sparse: each signal earns its place by changing a question, sharpening an architecture decision, or challenging an assumption.
Anthropic Engineering
A durable architecture reference for deciding when workflows are enough and when agent autonomy is actually warranted.
OpenAI
Useful framing for turning fuzzy quality expectations into explicit, measurable evaluation loops.
Anthropic Research
Connects agent autonomy to human control, transparency, security, and operational boundaries.
OpenAI
A current production reference for agent harnesses that combine tools, memory, controlled sandbox execution, durability, and explicit separation between orchestration and compute.
Claude by Anthropic
Extends the agent question beyond autonomy into coordination: how humans and agents divide work while retaining shared context and control.
OpenAI
A concrete example of turning intended model behavior into scenario-based evaluation rather than relying on prose policy alone.
Signals is not a breaking-news feed. Recency can make something visible; relevance is what keeps it here. Sources are selected for their usefulness to the lab's research, frameworks, systems, and build decisions.