The frameworks people use to build agents hit a run of stable milestones in mid-2026. Google’s Agent Development Kit reached 2.0, Pydantic AI shipped its 2.0, and CrewAI reached 1.15, all within weeks. The pattern is a layer maturing: after two years of churn, several agent frameworks now offer stable, versioned APIs you can build a product on rather than chase.

What happened

Google ADK 2.0.0 reached general availability on 19 May 2026, centered on a graph-based multi-agent engine for defining and running coordinated agents. Pydantic AI v2.0.0 shipped on 23 June 2026 as a stable release, bringing its type-safe, Python-native approach to a 2.0 API. CrewAI 1.15.0 (25 June 2026) advanced its declarative “flows” for orchestrating role-based agents.

Two absences are as informative as the releases. LlamaIndex did not ship a 1.0, staying on its 0.14.x line, and the OpenAI Agents SDK remained a 0.x with incremental updates. So the “stable 1.0/2.0” milestone is real for some frameworks but not universal; the layer is maturing unevenly.

Why it matters for builders

A stable major version is a signal you can plan around: fewer breaking changes, clearer support, and an API worth learning deeply. If you are choosing an agent framework in 2026, the release status is now a real input, a 2.0 GA (ADK, Pydantic AI) implies more stability than a pre-1.0 line (LlamaIndex, OpenAI Agents SDK), though maturity is not the only axis.

The deeper trend is that these frameworks are converging on the same primitives: define agents, give them tools, coordinate them in a graph or flow, and observe the result. That convergence, plus open protocols like MCP and the Agent Client Protocol, means the multi-agent orchestration patterns are becoming portable across frameworks. Pick one that fits your language and stability needs; see CrewAI vs LangGraph and LangChain vs LlamaIndex for the trade-offs.

Sources

Further reading