AI engineering blog · agents, RAG, systemshttps://rachid-chabane.com/en-UShttps://rachid-chabane.com/en/blog/mcp-stateless-core-header-routing/https://rachid-chabane.com/en/blog/mcp-stateless-core-header-routing/MCP's stateless rewrite is a retrofit, and the migration bill lands on the operators who never had the problem it solves. The specification states the change without hedging: the…Wed, 29 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/what-vllm-afd-numbers-ask-of-your-interconnect/https://rachid-chabane.com/en/blog/what-vllm-afd-numbers-ask-of-your-interconnect/The +11.3% that vLLM's AFD plugin reports at 64A16F [s1] is a number you cannot carry to your own cluster, and the release does not publish what you would need to earn it.…Mon, 27 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/your-incident-response-runs-on-a-vendors-content-policy/https://rachid-chabane.com/en/blog/your-incident-response-runs-on-a-vendors-content-policy/Hugging Face's responders could not analyze their own breach through commercial frontier APIs, which is why refusal behavior no longer belongs in any control set I write. The…Sun, 26 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/count-the-irreversible-actions-before-an-agent-stops/https://rachid-chabane.com/en/blog/count-the-irreversible-actions-before-an-agent-stops/The abstention score a benchmark reports is a rate, and the number that decides how you wire an agent is how many irreversible actions run before it stops. Cyera's enterprise…Sat, 25 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/swe-bench-private-holdout-moves-the-trust-problem/https://rachid-chabane.com/en/blog/swe-bench-private-holdout-moves-the-trust-problem/A private, contamination-resistant holdout lowers leakage risk but hands you a number no outsider can check. I read that as a trade: it buys contamination-resistance with…Fri, 24 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/kimi-k3-ranks-fourth-nobody-measured-the-weights/https://rachid-chabane.com/en/blog/kimi-k3-ranks-fourth-nobody-measured-the-weights/Every public number attached to Kimi K3 describes a rented API, not the weight file that is supposed to land on July 27. Artificial Analysis scores the model 57 on its…Thu, 23 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/your-agent-attends-to-the-right-tool-and-still-picks-wrong/https://rachid-chabane.com/en/blog/your-agent-attends-to-the-right-tool-and-still-picks-wrong/The reason your agent picked the wrong tool is almost never that your tool description was badly worded. In one 2026 study, prompt-repair recovers at most 23 percent of failures…Wed, 22 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/stop-editing-claude-md/https://rachid-chabane.com/en/blog/stop-editing-claude-md/Your CLAUDE.md structure is not why the agent stopped listening. Compliance falls about 5.6% in odds with every additional function the agent generates [s1], so the variable that…Tue, 21 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/injection-defense-layer-split/https://rachid-chabane.com/en/blog/injection-defense-layer-split/The three mid-2026 results that read like a field-wide fight about prompt-injection defense are not disagreeing with each other, and once you name the layer each one attacked,…Sun, 19 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/plus-24-percent-and-minus-19-percent-are-both-true/https://rachid-chabane.com/en/blog/plus-24-percent-and-minus-19-percent-are-both-true/There is no portable productivity number for coding agents: in this evidence base the instrument outweighs the tool, and proxy choice sets the sign of the headline. A study of…Sat, 18 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/your-agents-denylist-checks-a-string-the-shell-rewrites/https://rachid-chabane.com/en/blog/your-agents-denylist-checks-a-string-the-shell-rewrites/The auto-approve config you shipped this quarter checks a string that bash has not finished rewriting. Your gate and your shell disagree about what the command is, and they…Fri, 17 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/there-is-no-best-coding-agent-only-a-best-one-for-this-workload/https://rachid-chabane.com/en/blog/there-is-no-best-coding-agent-only-a-best-one-for-this-workload/The analysis everyone cites to prove the harness beats the model puts the harness at about 5.3% of the variation in success and the model at 0.7% [s2], a ranking that dwarfs…Wed, 15 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/verification-is-the-new-bottleneck-for-coding-agents/https://rachid-chabane.com/en/blog/verification-is-the-new-bottleneck-for-coding-agents/The hard part of training a coding agent is no longer generating a solution, it is verifying one, and that single relocation quietly breaks the reward recipe most teams are still…Tue, 14 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/multi-agent-wins-are-a-compute-artifact/https://rachid-chabane.com/en/blog/multi-agent-wins-are-a-compute-artifact/Normalize the thinking-token budget and the multi-hop reasoning edge people credit to multi-agent orchestration stops reliably showing up [s1]. That one control changes what the…Mon, 13 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/code-review-agents-are-shipping-noise/https://rachid-chabane.com/en/blog/code-review-agents-are-shipping-noise/The pitch for a code review agent is that it reads every pull request so you do not have to. An empirical study of 13 of them inverts that pitch: 60.2% of agent-only PRs land in…Sun, 12 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/claude-code-action-breach-was-an-auth-bug/https://rachid-chabane.com/en/blog/claude-code-action-breach-was-an-auth-bug/The Claude Code Action breach needed two things to work, a prompt injection and a broken authorization check, and the second is where your next hour of review pays off. A…Sat, 11 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/observability-is-not-evaluation/https://rachid-chabane.com/en/blog/observability-is-not-evaluation/Buying an observability dashboard is not the same as knowing your agent is right, yet that is the trade most teams are quietly making. Two independent 2026 surveys show the field…Tue, 07 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/coding-leaderboards-measure-harness-gaming/https://rachid-chabane.com/en/blog/coding-leaderboards-measure-harness-gaming/The rank order on a coding-agent leaderboard is partly a gaming score, and the gaming grows with capability instead of washing out as the models get better. The sharpest single…Mon, 06 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/agent-memory-benchmarks-grade-recall-not-trust/https://rachid-chabane.com/en/blog/agent-memory-benchmarks-grade-recall-not-trust/A 92.5 on the LoCoMo memory benchmark tells you your agent can retrieve a fact, and almost nothing about whether it will keep serving that fact with full confidence after it…Sun, 05 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/agent-debt-is-a-verification-placement-problem/https://rachid-chabane.com/en/blog/agent-debt-is-a-verification-placement-problem/Ninety-four percent of leaders rate AI-generated code as higher quality than human-authored code at review, and yet 82% of them hit at least one production failure tied to that…Sat, 04 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/coding-agents-fail-at-the-harness-not-the-model/https://rachid-chabane.com/en/blog/coding-agents-fail-at-the-harness-not-the-model/The next model release will not fix your coding agent, because the failures that cost you are not the ones a bigger model repairs. They are harness failures: what the agent can…Thu, 02 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/mcp-moved-the-trust-boundary-to-the-client/https://rachid-chabane.com/en/blog/mcp-moved-the-trust-boundary-to-the-client/MCP made wiring an agent to tools trivial by trusting the server's tool descriptions, and that one default quietly moved the trust boundary onto clients built to forward, not to…Wed, 01 Jul 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/prompt-caching-can-slow-your-agent-down/https://rachid-chabane.com/en/blog/prompt-caching-can-slow-your-agent-down/You flip on prompt caching expecting the bill to fall and the latency to drop, and most of the time it does. But the first controlled cross-provider study of caching on…Tue, 30 Jun 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/glm-5-2-where-you-run-it/https://rachid-chabane.com/en/blog/glm-5-2-where-you-run-it/A near-frontier coding model under an MIT license reads like the moment your default flips, and that read is wrong: the license hands you the weights, not your data flow, because…Fri, 26 Jun 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/teaching-an-agent-to-forget/https://rachid-chabane.com/en/blog/teaching-an-agent-to-forget/When an agent stalls halfway through a long task, the reflex is to reach for a bigger context window. That is the wrong knob: on long-horizon software work the binding constraint…Tue, 23 Jun 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/temperature-zero-is-not-deterministic/https://rachid-chabane.com/en/blog/temperature-zero-is-not-deterministic/Set an LLM to temperature 0 and run the same prompt a thousand times, and you will not get a thousand identical answers; the standard excuse is that GPU floating-point math is…Mon, 22 Jun 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/pass-at-1-is-blind-to-long-horizon-reliability/https://rachid-chabane.com/en/blog/pass-at-1-is-blind-to-long-horizon-reliability/Select an agent for long autonomous work by its pass@1 score and you optimize the wrong number: capability and reliability match at horizon one but diverge as the horizon grows,…Fri, 19 Jun 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/scaffold-not-model-swe-bench-score/https://rachid-chabane.com/en/blog/scaffold-not-model-swe-bench-score/Most teams read a SWE-bench number off a leaderboard and treat it as a property of the model, but for the comparison that actually drives procurement, choosing between adjacent…Thu, 18 Jun 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/evaluation-is-the-new-compute-bottleneck/https://rachid-chabane.com/en/blog/evaluation-is-the-new-compute-bottleneck/Most teams budget training runs to the dollar and treat evaluation as effectively free. Then a single agent sweep on the Holistic Agent Leaderboard runs about $40,000 [s1], and…Wed, 17 Jun 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/agents-erode-the-code-they-keep-editing/https://rachid-chabane.com/en/blog/agents-erode-the-code-they-keep-editing/Single-shot pass@1 and iterative maintainability measure different things, and on the axis that mirrors real software work, today's strongest coding agents are weak and get…Tue, 16 Jun 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/ai-repriced-the-canon/https://rachid-chabane.com/en/blog/ai-repriced-the-canon/Every convention I inherited bet that one cost bound software: editing code a human had to read, and a machine reader plus a stochastic author now split that cost in two. This…Wed, 10 Jun 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/chunk-on-the-syntax-tree-or-not/https://rachid-chabane.com/en/blog/chunk-on-the-syntax-tree-or-not/Structure-aware AST chunking helps code RAG, but the cAST headline overstates how much the chunking rule itself earns. cAST reports a 4.3-point Recall@5 gain on RepoEval and a…Wed, 10 Jun 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/your-context-window-is-a-ceiling/https://rachid-chabane.com/en/blog/your-context-window-is-a-ceiling/The context length printed on a model's spec sheet is a marketing ceiling, not an operating budget, so I size a prompt for what the model can actually use, not for what caching…Wed, 10 Jun 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/deterministic-coding-agent-workflows/https://rachid-chabane.com/en/blog/deterministic-coding-agent-workflows/Break an engineering task into verifiable steps, and let the agent fail early rather than late.Sat, 30 May 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/hybrid-rag-reciprocal-rank-fusion/https://rachid-chabane.com/en/blog/hybrid-rag-reciprocal-rank-fusion/Combine BM25 and vectors without tuning ten weights: rank is enough.Wed, 27 May 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/publication-guardrails-fact-checking/https://rachid-chabane.com/en/blog/publication-guardrails-fact-checking/Before publishing, the agent must prove its claims, sources attached.Sat, 23 May 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/quantizing-open-model/https://rachid-chabane.com/en/blog/quantizing-open-model/GPTQ, AWQ, GGUF: what quantization actually costs, measured.Tue, 19 May 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/evaluating-tool-using-agent/https://rachid-chabane.com/en/blog/evaluating-tool-using-agent/An agent that completes the task but wrecks the state hasn’t succeeded.Tue, 12 May 2026 00:00:00 GMThttps://rachid-chabane.com/en/blog/serving-oss-llm-production/https://rachid-chabane.com/en/blog/serving-oss-llm-production/vLLM, continuous batching, KV-cache: where the VRAM really goes.Mon, 04 May 2026 00:00:00 GMT