framework
Google's open-source Agent Development Kit (ADK) is a Python/Java framework for building, orchestrating, and deploying multi-agent AI systems.
No evidence-backed relationships are recorded.
kagent-agentevals keeps 8 of 14 ADK events from a real kagent session when it builds an agentevals trajectory for regression tests. Its author found the role field crediting some agent tool calls to the user, so the converter reads part types and drops runtime adk_ tools.
Reality
In apowerb the module the runtime imports is generated and disposable, and every instruction, tool and model string is read out of a database row at load time. The cost is a filesystem holding derived state.
Reality
The company says 88 percent of its AI proofs-of-concept never reach widescale deployment, and it blames the identity, session and evaluation layers each team was rebuilding, so APEX builds them once on Amazon Bedrock AgentCore.
Reality
A step-by-step build of the same Apache Iceberg data agent on Google ADK, AWS Strands and Microsoft Agent Framework measures what a cloud move costs in code, in latency and in storage wiring.
Reality
The framework never intersects a child's tool list with its caller's, so a supervisor holding one tool can spawn a writer that searches the web. The intersection is a middleware you have to write yourself.
Reality
The catalog serves its single-page app shell for malformed requests and for throttling alike, so the status line tells you nothing. The only gate that holds under load is checking the content type before you parse.
Reality
A dev.to walkthrough puts deterministic tool tests at the base of a four-layer agent pyramid and saves live Gemini runs for the top. It works only if you own the adapter that turns framework events into a contract.
Reality
Shipping meeting assistants mark coverage the moment a topic comes up. Junwei Lai's intake adjudicates each open item in its own call and defaults to insufficient, which pushes the missing answer back into the room while the patient is still there.
Reality
An eldercare agent published at a live URL puts its two hard rules in Python instead of prompts, and the demo tests them by calling the send function directly, which is the only version of that claim a reader can check.
Reality
A manager agent can only read the one dependency edge somebody wrote for a machine. That is why partitioning went back to a hand-maintained file, with a diff check standing between the engineers and the bookkeeping.
Reality
One escalation loop got the wording right but sent it to the wrong person. No unit test could have caught that, because who gets paged and what stops the paging are both resolved outside the function under test.
Reality
Sonjomon computes each action's autonomy from confidence and blast radius, which leaves a risk registry in code doing the load-bearing work while the confidence half rests on a number the model reports about itself.
Reality
Google ADK's output_key writes into whichever agent's session declared it, and in-process that session is shared. A branch-coverage gate that only ever runs the single-process topology cannot reach the failing path.
Reality
A discovery-only comparison of A2A agents on three clouds, measured in August 2026, found one protocol producing three auth models and one card whose sole declared interface URL is the container's bind address.
Reality
Bedrock AgentCore Evaluations promises to score agents built on LangGraph, LlamaIndex, OpenAI's SDK, Google ADK or the Claude Agent SDK. The catch sits in the instrumentation.
Reality
A grant of InvokeAgentRuntime will not fetch an A2A agent card on Bedrock AgentCore. And the card a Strands agent publishes there runs 2,109 bytes, most of it the agent's own instructions.
Reality
One line of Google ADK turns an agent into an A2A server. Deployed to Cloud Run and called from AWS and Azure, it advertised the address it binds instead of the address anyone can dial.
Reality
A Bedrock AgentCore agent serving Google and Azure callers over A2A pushes the vendor-specific work into the container contract rather than the agent. The credential claim is the part not yet shown.
Reality
Six paired evaluations on a Tokyo transit server: one better route, one worse, four identical. The guidance file changed the agent's decision path without breaking anything.
Reality
A paper scores five agent interoperability protocols against six governance dimensions and finds they coordinate tasks without expressing who may approve, how dissent survives, or when a human is called.
Reality
Build · August 20, 2026 · 1 publisher
Build · August 17, 2026 · 1 publisher