https://imrul-hasan.com/Imrul's tech blog on AI agents, platform engineering, distributed systems, and engineering leadership. 2026-11-01T08:07:01+01:00 Imrul https://imrul-hasan.com/ Jekyll © 2026 Imrul /assets/img/favicons/favicon.ico /assets/img/favicons/favicon-96x96.png 2026-11-01T08:05:00+01:00 2026-11-01T01:47:38+01:00 https://imrul-hasan.com/posts/this-week-in-ai-systems-3-rules-in-code-controls-in-iam/ Imrul

Four themes from Oct 23–29, 2026: agent platform design, enforced agent permissions, trusting evals before models, and systems at scale.

2026-11-01T08:03:00+01:00 2026-11-01T01:47:38+01:00 https://imrul-hasan.com/posts/systems-at-scale-storage-scheduling-and-features/ Imrul

ChatGPT's storage back to multi-tenancy, Uber's capacity-aware scheduling, Meta's fused GPU retrieval, Airbnb's fresh embeddings and GCP data-service updates.

2026-11-01T08:02:00+01:00 2026-11-01T01:47:38+01:00 https://imrul-hasan.com/posts/trust-the-eval-before-the-model-benchmarks-labels-and-the-model-menu/ Imrul

A mostly private Artificial Analysis index, new models from Google, Moonshot and OpenAI, copied eval labels, fragile judges and hidden harness spend.

2026-11-01T08:01:00+01:00 2026-11-01T01:47:38+01:00 https://imrul-hasan.com/posts/enforced-not-requested-sandboxes-egress-and-permissions-for-agents/ Imrul

Escaped microVMs, models calling a proxy they apparently learned about in training, IAM deny on MCP write tools and data-layer policy: agent limits that hold whatever the model decides.

2026-11-01T08:00:00+01:00 2026-11-01T01:47:38+01:00 https://imrul-hasan.com/posts/designing-the-agent-platform-one-loop-owned-skills-gated-tools/ Imrul

One agent loop with owned skills, rules in code, eval-gated skill releases and gated MCP tools: platform designs from Anthropic, Salesforce, AWS, Uber and Google.