Engineering notes
Since 2023 we have documented what actually works in AI systems — with numbers, sources and honest limits.
Agents act, so they need governed identities. This article defines agent identity, examines Microsoft Entra Agent ID and the MCP authorization model, and lays out least privilege at the tool boundary, delegation-aware audit trails, and shadow-agent discovery — including what each control does not solve and which control planes we expect next.
GPT-5 replaces the model picker with a real-time router that chooses between a fast path and a reasoning path per request. We examine how the router decides, why the API keeps explicit controls, what launch week revealed about predictability, and how evaluation practice must change when the product is a portfolio behind one name.
Declarative agents specialize Microsoft 365 Copilot through a JSON manifest — instructions, knowledge, and OpenAPI actions — running entirely on Microsoft's orchestrator with no hosting of your own. We examine the schema as of August 2025, governance in the admin center, the metering economics, and the concrete cases where pro-code custom engine agents remain the better choice.
A field guide to the retrieval stack that became the 2025 default: BM25 and dense vectors merged with reciprocal rank fusion, then reordered by a cross-encoder reranker. We cover the measured lifts from Anthropic and Cohere, a realistic latency budget, three production rerankers, and the query classes this baseline still cannot answer.
Google has donated the Agent2Agent (A2A) protocol to the Linux Foundation. AWS, Cisco, Google, Microsoft, Salesforce, SAP, and ServiceNow are founding participants. We examine what the protocol specifies, what it deliberately leaves out, and why neutral governance is the precondition for agent interoperability across organizational boundaries.
Agent journeys end at systems you cannot test against at volume: payment providers, ERP backends, partner APIs. We apply service virtualization to agents — simulated counterparts implementing the real contract, holding state, injecting faults — and argue that every simulated result must carry an explicit label. With verified data from τ-bench, τ²-bench, and Microcks 1.12.0.
On June 12 and 13, 2025, Cognition and Anthropic published apparently opposite advice on multi-agent LLM systems. We compare 'Don't Build Multi-Agents' with Anthropic's research-system post — a 90.2 percent lift at roughly 15x token cost — and derive a decision rule: parallelize decomposable breadth, keep shared-context work single-threaded.
Prompt wording no longer decides system quality; the contents of the context window do. 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.
Claude Opus 4 and Claude Sonnet 4, released May 22, 2025, 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.
The MCP specification revision 2025-03-26 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.
On April 9 2025 Google announced the Agent2Agent protocol (A2A) 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.
Agent evaluation in 2025 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.