Agentic Systems Intelligence
Build better agents. Know what holds up.
Research-backed decisions for engineers and technical leads shipping AI agents: architectures, evaluations, memory and operating trade-offs.
What are you building?
Choose an agent architecture
Start with the agent types, then compare how they use tools and coordinate.
Read the architecture guide →Evaluate before you ship
Explore benchmarks, failure modes and evidence for deployment decisions.
Explore evaluation and safety →Design memory and retrieval
Understand the trade-offs in RAG, context engineering and persistent memory.
Explore reasoning and memory →Categories
Six research verticals covering the full AI agent stack.
Agent Design
Architectures, tool use, and frameworks for building agents.
- ComplexMCP Finds Tool-State Gaps
- ToolGrad Turns Tool Failures Into Data
- Can Next-Turn Metrics Test Workflows?
Swarm Systems
Multi-agent coordination, swarm intelligence, and collective behavior.
- BenchAgent Tests Multi-Agent Lift
- Tool-Using Agents Can Hide Messages
- Multi-Agent Finance Workflows Need Cost Curves, Not More Agents
Reasoning & Memory
Reasoning tokens, RAG, context engineering, and memory systems.
- Can Memory Systems Track a Group?
- When Tools Get in the Way Cuts Accuracy
- Can SentinelBench Measure Waiting?
Safety & Governance
Red teaming, bias, interpretability, and benchmarks.
- OpenShell Moves Agent Safety Below the Model
- Cross-Substrate Authority Moves to Runtime
- Claude Misuse Moves From Chat to Campaigns
Models & Frontiers
Model comparisons, training data, open source, and research frontiers.
- Android Bench Tests Long Coding Work
- Progress Hints Cut Tool-Call Waits
- Should MoE Experts Stay on the GPU?
Real-World AI
Enterprise deployment, workforce impact, and developer tools.
- KNOWS Tests Browser Work After Search
- OSWorld-Pro Exposes Click Failures
- SciAgentArena Maps Science Agent Limits
Latest
Most recent articles across all categories.
ComplexMCP Finds Tool-State Gaps
ComplexMCP is a useful benchmark because it tests the part of tool use that simple function-calling tasks avoid: one action changes the state that the...
Android Bench Tests Long Coding Work
Android Bench turns mobile coding evaluation from small patch repair into long-horizon Android work. Google says the updated benchmark adds multi-day...
Can Memory Systems Track a Group?
GroupMemBench reframes agent memory as a group problem, not a private chat log. The paper argues that assistants used inside teams need to remember who...
KNOWS Tests Browser Work After Search
The KNOWS paper turns web-agent evaluation from finding information into producing a usable document, spreadsheet or slide deck...
OpenShell Moves Agent Safety Below the Model
NVIDIA's Open Agent Safety Platform is worth reading as an infrastructure claim, not as a new prompt rule [NVIDIA...
When Tools Get in the Way Cuts Accuracy
The paper *When Tools Get in the Way* isolates a failure mode that is easy to miss in agent design: a model can answer worse when an unnecessary but...