Explore fresh perspectives and expert takes on design, development, and digital innovation.
Agents rarely fail because your prompt is slightly off. They fail because “resume after tool failure” is underspecified: retries, partial progress, and state handoffs aren’t durable, so you can’t safely continue — and you can’t observe what really happened.
Read more →
Hybrid (BM25 + vectors) plus a cross‑encoder reranker is now the default RAG advice, but it can make real systems worse. Here’s a practical, eval-driven way to decide when to rerank using similarity margins and failure-mode buckets.
Read more →
Tool-using agents fail differently to chatbots: they can cross system boundaries. The Permission Boundary Pattern gives you an implementable blueprint for agent identities, per-tool scopes, short-lived credentials, and end-to-end auditability so overreach is detectable and revocable.
Read more →
Vector-only RAG fails in predictable places: IDs, SKU-like tokens, exact clauses and compliance language. A production hybrid stack (BM25 + dense + reranking + ACL-aware filtering) fixes this, and you can prove it with a simple evaluation harness.
Read more →
Teams are adopting MCP to standardise tool access for agents, but the hard part is governance: auth, failure behaviour, observability, and versioning. This article covers the integration patterns that keep MCP tool estates secure, debuggable, and maintainable in production.
Read more →
Your agent is great until it hits a messy inbox: the hard part is not the model, it’s routing, permissions, handoffs, and getting to a real ‘done’ without inventing a new queue. This article lays out a concrete pattern for in-thread, tool-using resolution that keeps work where it started and makes outcomes auditable.
Read more →
Agentic automation fails in production for boring reasons: you can’t diagnose incidents without drowning in traces, and you can’t ship changes without regression gates. This is a practical playbook for OTel-first tracing, span/trace volume controls, and evals as release gates so on-call can actually operate the system.
Read more →
Most agent demos fail in production for the same boring reasons: retries, partial failure and missing state. A ‘durable agent’ stack treats agents as long-running workflows with checkpoints, idempotent tools and an audit trail you can trust.
Read more →