Gen AI and agentic AI development tools

Even though AI, gen AI and agentic AI are evolving quickly, and seemingly gaining skills daily, there are some essential tools for developing and implementing them. Here are some of the best known and reportedly most widely applied so far:

Claude Code is an AI-based, agentic, command-line interface (CLI) tool from Anthropic that runs in local software terminals, and uses natural language to interact, understand, construct and debug codebases. It functions as an autonomous, interactive assistant that can navigate repositories, adjust files, perform tests, and managing Git operations, which are programming instructions and actions used by Git distributed version control systems.

Industrial information interoperability eXchange (i3x) is a standard, manufacturing-information API that delivers provide a common interface for accessing, contextualizing and sharing production data. Developed by the Clean Energy Smart Manufacturing Innovation Institute (CESMII), it seek to pierce data silos, and encourage portable applications that can operate across platforms and settings.

Large language models (LLM) access huge volumes of text, which is used for training by machine learning (ML) that’s self-supervised. They typically generate natural language and perform other language-processing tasks. Generative, pre-trained transformers (GPT) are the biggest, most capable LLMs. They can be tailored for particular jobs or guided by prompts. They gain predictive capabilities by using human syntax, semantics and ontologies, but also acquire preexisting mistakes and prejudices in their training data. The best-known LLMs include ChatGPT, Gemini LLM, Claude and Llama.

Get your subscription to Control's tri-weekly newsletter.

Model context protocol (MCP) is an open-source standard, also developed by Anthropic, which lets AI-based applications and models like ChatGPT and Claude link securely to outside information sources like local files and databases, software programs, and tools like search engines and calculators. It serves as a common bridge, like a USB port for AI, which lets LLMs  access files, databases and application program interfaces (API) without specialized integrations for each device.

MCP servers are programs that introduce functions to AI applications via standardized protocol interfaces, and deliver functions via three building blocks: tools, resources and prompts. They use several server types, including file system for document access, database for data queries, GitHub for code management, Slack for team communication, and calendar for scheduling.

Retrieval-augmented generation (RAG) lets LLMs bring back and integrate new information from outside sources. The means LLMs don’t have to answer queries until they reference a particular collection of texts, which are added to data from their existing training information. This permits LLMs to employ domain-specific and/or up-to-date content that isn’t available in the training data.

Software agents are behavior-based programs that can act on behalf of a human user or another software program. They’re reportedly able to start themselves in response to contextual conditions, activate other functions such as initiating communications, and don’t need to interact with users to perform tasks.

Unified namespace (UNS) allows users to collect data, incorporate meaning and context, and convert it into a format that other users and systems can understand. UNS accomplishes this by separating content computing from task computing, and setting up a central repository for contest and information. In this location, other users can consume or disseminate information needed to perform tasks. UNS usually partners with Message Queuing Telemetry Transport (MQTT) publish-subscribe protocol and Sparkplug B framework on top, and together they provide a platform for digitalized, scalable processes.