Leverage Jupyter Notebooks for iterative development and exploration of AI code rather than expecting one-click deployment from the outset.

Use an Agents MD file—a markdown-based specification—to declare your AI agent’s tools, workflows, and execution pointers in a human-readable yet machi...Show more

When building agents with the OpenAI API, define and register functions in a standardized JSON schema to drive uniform execution and response handling...Show more

"Claude skills could really come in handy."

OpenAI’s ChatGPT Pro offers advanced code-generation and data analysis features, including support for matplotlib visualization and the ability to pac...Show more

Anthropic’s Claude Skills provides a script-based skill definition and config engine for LLMs to enforce consistent behavior across invocations.

Tom recounts using ChatGPT Pro for data analysis with matplotlib, noting highly detailed graphs and clear outputs that exceeded expectations.

Develop extensions to OpenAI’s Assistants SDK that mimic Claude skills, allowing developers to register and invoke custom LLM skills programmatically.

Build a marketplace or internal catalog of shareable LLM skills that teams can reuse, validate, and extend, improving collaboration and reproducibilit...Show more

Frame LLM-driven systems as an operating-system engine for enterprises that embeds consistent processes, guardrails, and reusable skills to streamline...Show more

Adopt a shared config-file approach (inspired by Claude skills) to enforce consistent LLM prompt structures and processing pipelines across your devel...Show more

Use LLM “skills” as modular scripts with guardrails to ensure consistent code generation across repeated queries by converting ad-hoc prompts into reu...Show more

Built a temporary LLM-driven interface with fixed prompts to avoid exposing a full chatbot, balancing safety and functionality during development.

A SaaS offering that ingests financial API flat files, preprocesses them into optimized formats, and provides containerized endpoints to AI apps for h...Show more

When compute resources are abundant, raw streaming and brute-force data handling may suffice, but efficiency matters for constrained environments.

Utilizes ChatGPT's built-in code interpreter to generate and execute Python code for on-the-fly analysis within a testing server environment.

Building a scoped tool like 'MCP Central' that exposes specific function calls with fixed one-shot prompts helps safely test and debug LLM-driven work...Show more

Downloading and storing API flat-file data like tickers CSV in a container allows pre-processing and improves query performance over streaming raw API...Show more

Cameron recounts initially failing to configure the remote server connector in ChatGPT until Tom guided him through the connectors UI.

Tom observes that the current market data API was not designed for MCP tool calls, indicating opportunity to redesign endpoints for better ChatGPT int...Show more