Engineering notes
Every article starts with a practical question, explains the necessary terminology, and links to primary sources. We separate external evidence, project experience, and conclusions — and state what cannot be inferred from them.
Sixty tool calls produce a history longer than any window, so something has to go. Summarising the oldest part discards the decisions and the failed attempts, which is exactly what stops an agent repeating itself. Kang et al. optimise what to compress for long-horizon agents. Our rule: observations compress, decisions and errors stay verbatim, and the reason is measurable.
On 22 October 2025, LangChain 1.0 and LangGraph 1.0 shipped with a no-breaking-changes commitment until 2.0. We examine create_agent and its middleware hooks, checkpointer-backed persistence, human-in-the-loop interrupts, and the mechanics of migrating legacy chains to langchain-classic — including what the stable APIs do not solve.
The technicians who most need the documentation work where there is no signal. What fits on a handheld device is a small model and a subset of the corpus, which forced a decision nobody wanted to make: which questions this thing is allowed to be asked. The scoping that made it useful, and why the honest version says out loud what it cannot see.
At three in the morning the hard part is orientation, not the fix: what changed, what is affected, what happened before. Xiang et al. combine cluster state with a model for root cause analysis. Ours is read-only, proposes hypotheses with the evidence attached and never a conclusion, and the rule that keeps it honest is that it may not say what to do.
Long-running agents cannot keep their state in process memory for hours or days. We define durable execution, compare event-history replay with checkpoint snapshots, survey the late-2025 engine landscape from Temporal to LangGraph 1.0 and Microsoft Agent Framework, and state plainly what durability does not solve: side effects, decision quality, and context growth.
We spent two years telling customers a knowledge graph was overkill, and then a corpus arrived where vector retrieval failed on nearly a third of questions for structural reasons. Ahmad et al. benchmark vector, graph and hybrid pipelines side by side. What made this corpus different, what the graph cost to build and keep current, and why the answer was hybrid rather than a replacement.
A self-improving assistant that adjusts its own instructions from corrections is the most requested feature we decline. Han et al. describe how self-evolution can push agents away from their intended behaviour over time. Our objection is narrower and harder to argue with: a system that rewrites itself has no version anyone can name, and every process around it assumes one exists.
At its DevDay on 6 October 2025, OpenAI introduced AgentKit — a visual Agent Builder, the embeddable ChatKit UI, extended Evals and a Connector Registry — plus apps that run inside ChatGPT via an MCP-based Apps SDK. We summarize what shipped, what is beta, where lock-in sits, and how agent teams should frame the build-vs-buy decision.
Our evaluation set asked one question and checked one answer, and most real failures happened on the third turn. Kranti et al. benchmark task-oriented dialogue systems as whole conversations. What a multi-turn case has to specify, why a simulated user is a test fixture rather than a judge, and the two failure modes that only appear once a conversation has a history.
Extraction accuracy fell four points over a weekend with no deployment and the same model identifier in every request. A version string is a name the provider controls, not a guarantee about behaviour. What we run daily to catch this, the contractual question we now ask before signing, and what we do when the answer is unsatisfying, which it usually is.
Ahead of Microsoft Ignite 2025 in November, we assess what Azure AI Foundry's Agent Service as a hosted runtime delivers since its May GA, what the agent factory narrative signals, why the Azure prefix is eroding, and which capabilities engineering teams should adopt now versus merely watch.
After eighteen months of running our adversarial prompt set in English, we translated it. Roughly twice as many got through, on a system where most users write German. Kumar et al. build multilingual moderation across seventeen languages precisely because coverage is uneven. What we changed, why translation is not enough to build the test set, and the general lesson about which language a safety property was measured in.