Agents tooling momentum

An AI agent is software that can plan steps, call tools like search or code runners, and keep state across a task. In April 2026, developers are circulating bigger maps of that tooling stack, from framework lists with 100-plus projects to roadmaps for deploying agents in production. (developers.openai.com, github.com) The new crop of guides is pointing builders toward a handful of common starting points: OpenAI’s Agents SDK, Alibaba-backed AgentScope, LangChain, LangGraph, and Nous Research’s Hermes Agent. OpenAI’s current docs describe agents as applications that “plan, call tools, collaborate across specialists, and keep enough state” for multi-step work. (developers.openai.com, developers.openai.com) Those projects now span both hosted and open-source approaches. OpenAI’s quickstart offers Python and TypeScript paths for defining an agent, running it, then adding tools and specialist handoffs, while LangGraph’s docs position it as a low-level runtime for long-running, stateful agents. (developers.openai.com, docs.langchain.com) AgentScope is pitching itself at teams that want production controls earlier in the build. Its GitHub repository says the framework is “production-ready,” built around tool use and reasoning, and its public roadmap is labeled “Jan 2026 -,” signaling active iteration rather than a frozen research project. (github.com) The same shift is visible in the numbers around the open-source ecosystem. LangChain’s main GitHub repository shows about 134,000 stars, AgentScope’s repository shows about 23,900 stars, and Hermes Agent’s repository shows about 102,000 stars as of April 2026. (github.com, github.com, github.com) Hermes Agent’s recent growth has been especially fast. Its GitHub page showed more than 102,000 stars on April 20, 2026, and its April 16, 2026 release added a “Tool Gateway” with web search, image generation, text-to-speech, and browser automation for paid portal subscribers. (github.com, github.com) LangGraph and LangChain are filling a different part of the market: orchestration rather than a single bundled assistant. LangGraph says companies including Klarna, Uber, and J.P. Morgan use it for “long-running, stateful agents,” and its GitHub repository describes the project as a way to build “resilient” agents as graphs. (docs.langchain.com, github.com) OpenAI has been moving in the same direction on the platform side since it introduced new agent-building tools around the Responses API. The company said those tools were meant to simplify orchestration and tool use, and its current docs now route developers from a quickstart into more advanced runtime patterns. (openai.com, developers.openai.com) The result is a market with fewer blank pages for new builders. Instead of stitching together memory, tool calling, retries, and multi-agent handoffs from scratch, developers in 2026 can pick from mature SDKs, open-source runtimes, and fast-growing GitHub projects that already package those pieces. (developers.openai.com, github.com, docs.langchain.com)