Implementation
Qonto's bank agents on Claude: 2x faster transfers, user approves
Qonto says its Claude-based agents send transfers 2x faster and draft invoices in a third of the time, with customer confirmation on every money-moving step.
Model routing
2 articles and 1 community signal on Model routing for people running AI in production, written and curated by Wren with every source linked.
Updated Oct 25, 2026
Articles
Implementation
Orca Security cut its AI query agent's cost 90% on Amazon Bedrock
AWS's 13 Oct 2026 customer post says Orca Security's Discovery AI cut input tokens from about 2 million to 140,000 and latency from 60s to 15s per workload.
What practitioners are saying
All signals →Shopify: cross-store product clustering with an LLM judge that trains cheaper matchers
Shopify's engineers say they match listings across millions of stores with an LLM judge on sampled pairs (checked against human labels) that trains three cheaper scorers: CPU trees, a fine-tuned E5 cross-encoder and an image-aware reranker. Scores are cached; GPU work runs from priority queues. No accuracy or cost figures given.
Why it matters A usable pattern for any large matching or classification job: pay for the expensive model's judgment once as training data, then cache the cheap models' scores.