Tools & Protocols
Microsoft brings Claude’s tool connections into Foundry
Foundry’s August Claude integration shows how tool discovery and connection management are becoming platform capabilities.
The lead / Enterprise Agents
Genspark’s Gen-1 Slides makes finished work the agent’s job
Genspark’s September 2026 specialist model ties autonomy to the creation, inspection, and revision of a usable presentation.
Read the analysis ↗
Across the publication
On our radar
Multi-Agent Systems
Anthropic’s compiler experiment exposes the real work of agent handoffs
The 16-agent compiler prototype and Claude Research used different coordination mechanisms because they had different work to finish.
Reliability & Oversight
Agent reliability starts with the result the system leaves behind
Anthropic’s evaluation and infrastructure studies show why a fluent final answer and a leaderboard score can both miss the real outcome.
The subject desks
View all ↗Enterprise Agents
Autonomous work inside real business processes.
Enterprise Agents
Genspark’s Gen-1 Slides makes finished work the agent’s job
Genspark’s September 2026 specialist model ties autonomy to the creation, inspection, and revision of a usable presentation.
Enterprise Agents
Cursor’s new bots follow code beyond the pull request
The September launch of Rollouts and Security Reviewer extends Cursor’s agents from generating patches to monitoring their consequences.
Enterprise Agents
An office agent’s hardest handoff is the editable file
GenOffice and Gen-1 Slides put the handoff to a human editor at the center of the office-agent promise.
The subject desks
View all ↗Tools & Protocols
The connections and contracts that let agents act.
Tools & Protocols
Microsoft brings Claude’s tool connections into Foundry
Foundry’s August Claude integration shows how tool discovery and connection management are becoming platform capabilities.
Tools & Protocols
Claude’s tool-use push makes interface design a reliability issue
Tool search and usage examples expose a practical limit on agent autonomy: the model must understand what each interface actually does.
Tools & Protocols
Persistent projects make agent memory a software-state problem
Cursor Projects and Anthropic’s long-running experiments show why remembering the conversation is only part of preserving progress.
The subject desks
View all ↗Multi-Agent Systems
Coordination, delegation, and the costs of collaboration.
Multi-Agent Systems
Anthropic’s compiler experiment exposes the real work of agent handoffs
The 16-agent compiler prototype and Claude Research used different coordination mechanisms because they had different work to finish.
Multi-Agent Systems
More agents do not automatically produce independent judgment
A compiler reference and a deliberately isolated permission checker illustrate why multiple agreeing models are not enough.
Multi-Agent Systems
Managed Agents moves orchestration into the infrastructure layer
Anthropic’s April architecture separates durable sessions, execution, and the changing logic that directs long-running work.
The subject desks
View all ↗Reliability & Oversight
Repeated success, useful review, and bounded authority.
Reliability & Oversight
Agent reliability starts with the result the system leaves behind
Anthropic’s evaluation and infrastructure studies show why a fluent final answer and a leaderboard score can both miss the real outcome.
Reliability & Oversight
Claude’s auto mode separates dangerous actions from persuasive explanations
The March permission design and September enterprise safeguards address different layers of the authority problem.
Reliability & Oversight
Self-correcting agents need evidence outside their own narrative
Cursor’s deployment monitoring and Anthropic’s testing reports ground self-correction in observations that can contradict the model.
The subject desks
View all ↗Agent Economics
The cost of completed work, not just generated tokens.
Agent Economics
Cursor’s efficiency work puts the agent harness on the cost ledger
The September efficiency release shows how prompt structure, tool loading, and caching affect the price of autonomous work.
Agent Economics
Cursor Router makes model selection part of the agent’s work
The July launch and August engineering account explain the signals behind automated model choice—and what those signals cannot prove.
Agent Economics
Long-running agents make context an operating expense
Cursor’s efficiency changes and Anthropic’s context architecture show why useful memory must be selective, inspectable, and refreshable.