Anthropic’s Claude as an ecosystem of AI tools rather than a single chatbot, with different products, models, connectors, skills, memory systems, and projects suited to different oil & gas tasks. Professionals should select the right Claude tool for the job while keeping humans responsible for design, validation, security, and deployment.
Claude’s main tools
- Claude.ai / Claude App – General-purpose conversational AI for planning, analysis, requirements, documentation, and other business tasks. Projects can contain persistent instructions and files that provide reusable context across conversations.
- Claude Code – Developer-focused AI accessed through a terminal or IDE such as VS Code. It’s designed for writing, debugging, refactoring, and working directly with Salesforce code and metadata.
- Claude Design – Focused on visual work such as prototypes, wireframes, UI mockups, pitch decks, and potential Lightning Web Component designs.
- Claude Cowork – Handles more complex, multi-step work by combining connectors, skills, subagents, and interaction with the user’s computer. Claude Dispatch extends this by allowing tasks to be initiated from a phone and executed on the computer.
- Claude Tag for Slack – Allows teams to interact with Claude inside Slack channels almost like another team member, including shared memory within the channel.
The key building blocks
A particularly useful distinction is between Connectors and Skills. MCP connectors give Claude access to external tools and systems—including Salesforce, GitHub, browsers, and documentation sources. Skills provide the instructions, processes, scripts, or actions telling Claude how to use those tools to accomplish a particular job.
In simple terms:
Models = intelligence → Connectors = access → Skills = procedures → Memory/Projects = context
Projects are especially useful for recurring Salesforce work because instructions and reference documents can be loaded once and reused across multiple conversations.
Claude models
The hierarchy of models designed for different levels of complexity and cost:
| Model | Positioning |
|---|---|
| Mythos | Extremely powerful, restricted/invite-only model |
| Fable | Highest-end public model for very complex tasks |
| Opus | Strong choice for agentic coding and complex development |
| Sonnet | Balance of speed, intelligence, and lower token usage |
| Haiku | Fastest/lightest model for simpler tasks |
The recommendation is essentially don’t use the most powerful model automatically. Match model capability—and therefore token consumption—to the complexity of the job.
