Cameron Rohn highlights a blog by an associate professor of computer science providing printable, handwritten worksheets to manually compute vector an...Show more

Use the PID5 personality inventory to benchmark LLMs across maladaptive traits by overlaying model scores on human response curves and flagging those ...Show more

Despite recent buzz around alternatives, the MCP framework remains central to enabling robust code execution in prompt engineering workflows.

Conducting yearly surveys within your network can track shifts in tool adoption, model preferences, and developer workflows over time.

T2 bench is specifically designed to evaluate conversational agents in a dual-control environment rather than raw compute performance.

Google’s internal computer model performance is evaluated entirely using screenshots, indicating a visual-based benchmarking methodology.

Use back-of-envelope math against open-source model benchmarks to estimate proprietary model parameter counts, revealing Gemini’s scale (~1.5T–30T par...Show more

Cameron Rohn uses a separate GPT instance in “thinking mode” to outline changes and then pastes those user stories or plans into a Claude file to exec...Show more

Tom Spencer often manually simulates agent-to-agent handoffs by creating a “handoff document,” pasting it into the next agent’s prompt, and iterating ...Show more

Cameron Rohn spun up three agents plus a supervisor agent to each pull batches of tasks from a 100-item backlog, mark files to indicate ownership, and...Show more

Adopt an “Agent harness” pattern that orchestrates multiple AI agents in the background to enable modular, scalable workflows.

Compare AI model performance in dedicated agent-only environments versus coding-centric IDE integrations to identify optimal use cases and capabilitie...Show more

Use the same codebase and prompt to compare AI model behavior across providers (e.g., v0 Gemini 3 vs new environment) by measuring tool call success r...Show more

Leveraging multiple AI agents in a pipeline by having them operate on top of each other enables complex, modular workflows for advanced task orchestra...Show more

Combine local, on-device models for inference-heavy tasks with cloud-hosted models for specialized or larger-scale workloads to optimize performance a...Show more

A simple yet effective architecture for deep agents consists of a planning tool (inspired by Claude checklists), a detailed and structured system prom...Show more

Integrate research workflows with systematic tracking of feature and tool additions to ensure visibility and reproducibility in AI development.

Use an interactive UI as a reference point to iteratively research and develop deep agents, refining capabilities through successive demos.

Emphasize building AI tools around a user-driven iterative cycle to drive practical adoption and refinement of capabilities.

Use the XJS starter template to streamline app development and accelerate approval processes.