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.
System quality no longer hinges on prompt wording, but on the context window's contents. This article defines context engineering as of mid-2025: what enters the window, how compaction, retrieval, tool results and memory files behave, why claimed context lengths mislead, and how measurable token budgets turn context into an engineered resource.
A customer asked why their assistant could not just look it up when the manual is silent. Web access answers more questions and makes every answer unattributable to a source they control, which in a regulated industry matters more than the answer. Where we do allow it, the two properties a source needs before it goes in the corpus, and what we do instead when the manual is genuinely silent.
Memory across sessions is the most requested feature we have, and it is the only one that lets a user change what a different user is told. Dong et al. show memory can be poisoned through ordinary queries alone, with no privileged access. What we scope, what we never write to memory, and the review that turned a two-day feature into a two-week one.
Released May 22, 2025, Claude Opus 4 and Claude Sonnet 4 extend usable coding horizons from minutes to hours: 72.5% on SWE-bench Verified, extended thinking with tool use, memory files, and a seven-hour autonomous refactor at Rakuten. We examine what changes for engineering teams — and where human review must remain.
Accuracy improved for a year and escalations from one user group did not move, because the assistant answered in the language of a manual written for engineers. Simplification is not a prompt instruction that says be simple; done carelessly it drops the qualifier that made the sentence true. What we simplify, what we refuse to, and how we measured whether it worked.
A procurement department asked for the energy footprint of an assistant for a sustainability report, and we had an invoice and no way to answer. What we can measure honestly, what we can only estimate with a stated range, and the design decision the exercise changed: a background summarisation step nobody was waiting for that ran on every document and could run on a fraction of them.
A customer asked whether their agent could talk directly to their supplier's agent. Technically yes; the hard part is that neither side can see why the other decided what it did, so a wrong order has no diagnosable cause. What we build instead of a free-form exchange, the three things we insist a cross-boundary agent interaction records, and where we said no outright.
With its 2025-03-26 specification revision, MCP replaced the HTTP+SSE transport with Streamable HTTP and added an OAuth 2.1 authorization framework. We trace the path from local stdio servers to authenticated remote endpoints, review the first vendor offerings from Cloudflare, GitHub and Stripe, and outline integration patterns and open gaps for enterprise deployments.
Google announced the Agent2Agent protocol (A2A) on April 9, 2025, with more than 50 partners. We examine what A2A specifies — Agent Cards, task lifecycles, artifacts — how it complements Anthropic's MCP, what the draft deliberately leaves open, and why governance will decide whether it becomes the interoperability layer for multi-agent systems.
The passage was correct, well written and three years out of date. Our index held four revisions of the same manual with no notion of which applies, because similarity search has no opinion about time. What we changed in the metadata, why the default has to be the current revision rather than the best match, and the question type we now refuse to answer without a date.
In 2025, evaluating agents means judging trajectories and tool calls, not just final answers. We define step-level metrics, per-task success predicates, and reliability measurement with pass^k, drawing on τ-bench, WebArena, BFCL V3, and SWE-bench Verified — and state plainly what trajectory evaluation does not deliver and why it is harder than RAG evaluation.
OpenAI announced support for Anthropic's Model Context Protocol on March 26, 2025, across its products, starting with the Agents SDK. We examine what the protocol standardizes, what it deliberately leaves open, why cross-vendor convergence changes integration strategy, and what the simultaneous specification revision means for remote MCP servers.