Agents should maintain system-level and user-level memory blocks to separate application state and individual user preferences for more nuanced intera...Show more

You can attach any number of MCP servers (e.g., build, exa) as tools in Deep Agents to inject custom context pipelines into your agents.

The Deep Agents package supports both mcp and LangChain tools and lets you modify the system or research instructions directly (e.g., in research_agen...Show more

By annotating and marking up Agent Builder output in an evaluation-like workflow and feeding it back, you can iteratively refine both primary and sub-...Show more

Implement a cognitive architecture with dynamic sub-agents that spawn to fetch and validate data (e.g., public records) and then reintegrate results i...Show more

To embed a sequence of non-trivial tasks into a deep agent, begin by selecting a deep research agent framework and defining your task sequences for th...Show more

The Deep Agents UI presents task decomposition as a checklist with status updates, enabling iterative planning and action execution in one interface.

Cameron set up a command-line agent called Granola to mimic Agent Builder workflows for automating task flows outside of a UI.

They prototype a pipeline of specialized agents—product manager, researcher, stakeholder communicator, backlog manager, and feedback analyzer—to modul...Show more

The first iteration of their agent tooling uses a virtual in-memory file system without persisting data to disk, simplifying early development.

Enable agents to read from and write to a file system for persistent context and data handling during execution.

Use a detailed system prompt that includes complexity handling, failovers, and routing logic to enable robust long-running tasks.

Incorporate a planning tool based on a Claude-style checklist to break down tasks and guide agent execution.

An effective AI agent architecture includes a planning tool, a comprehensive system prompt for complexity and failovers, a file system for read/write,...Show more

Use the Deep Agents package to automatically generate a team of specialized sub-agents via natural language prompts, breaking down complex tasks and o...Show more

Tom Spencer highlights that AI agent failures are often caused by unspoken assumptions and urges developers to explicitly supply all relevant context—...Show more

Structure multi-agent pipelines as a loop of processing steps followed by context injection to prevent output degradation across chained AI operations...Show more

Combine a deterministic workflow builder with non-deterministic Langgraph agent nodes in one canvas to manage both structured sequences and variable i...Show more

Adopt a systematic visual mapping exercise for chained AI agents to manage triggers, data flows, and inter-agent dependencies as complexity grows.

Use AI-powered "vibe coded apps" to ingest actual product analytics and context to automatically generate a custom roadmap planning and KPI tracking t...Show more