Is RAG dead? What the measurements say about long context and retrieval
Long context scored 73.1% against RAG's 65.4%, at 26 times the cost per query. The measured boundary, and why the right answer is routing.
- rag
- long context
AI engineering
Building agents that survive production: tools, verification, context, guardrails and telemetry, with every figure traced to a primary source.
MCP, A2A and WebMCP: the map of agent protocols in 2026
MCP connects the tool, A2A connects agents, WebMCP connects the site. What each solves, the real state of each, and what none of them expresses.
- mcp
- a2a
Context engineering: the four pillars and why a full window hurts
Across 18 models tested, performance falls as the input grows, before the window fills. The four pillars for keeping context small.
- context engineering
- context rot
Agent memory: what to keep, what to forget and how to avoid repeating the error
Memory cut the steps from 20 to 5 in a Databricks study. And it made the agent repeat its own error with more confidence. What to keep and what to delete.
- agent memory
- context engineering
Evals for agents: how to test what has no right answer
89% of teams instrumented their agents and only 52% evaluate. How to build a test set, choose a threshold and put the gate in CI without false alarms.
- evals
- llm evaluation
LLM as a judge: when the score deserves trust
An LLM judge agrees with humans as much as humans agree with each other, and errs in measured ways: the authorship label alone moves the score by 0.43 points.
- llm as a judge
- llm evaluation
Guardrails for agents: where to lock before it acts
Of the documented agent failures, 40 out of 52 were preventable by harness layers. Where to put the lock and why a classifier is not enough.
- guardrails
- agent security
Agent observability: instrumenting what you cannot reproduce
An agent bug does not reproduce: the trace has to exist before the error. The span tree, the OpenTelemetry attributes and what is still moving.
- observability
- opentelemetry
Tool design for agents: what it can actually use
With 10 tools the agent gets everything right; with 107 it fails outright. How to design tools an agent uses well, with before and after.
- tool design
- mcp
The verification loop: why the agent cannot review its own work
Asking "are you sure?" makes the model switch answers in 46% of cases and lose 17 points. The bottleneck is finding the error, not fixing it.
- verification loop
- self-correction
Harness engineering: what separates a demo agent from a production one
Harness engineering explained: why LangChain climbed 25 places without changing the model, the five layers, and what decides a long task.
- harness engineering
- ai agents
OWASP Top 10 for agents: the ten threats of 2026, and what already happened with each
The official OWASP Top 10 list for AI agents, from ASI01 to ASI10, with the real, dated incident that has already exploited each of the ten risks.
- owasp
- agent security
Prompt injection: why there is no definitive fix, and what to do anyway
Twelve prompt injection defenses reporting near-zero rates were broken above 90%. Why the problem is structural and what reduces risk.
- prompt injection
- agent security