Anthropic buys Mendral, solo AI agents collapse

Regulated firms can boost AI‑driven development without inflating compliance risk by turning verification into a continuous pipeline rather than a final gate. The approach lets domain experts contribute code faster while keeping security and auditability baked into each commit.

Anthropic has acquihired the founders of Mendral, a startup that built AI‑driven security, reliability and performance agents for CI/CD pipelines. The move lets Anthropic embed Claude‑powered automation directly into software engineering flows as rapid model advances render traditional tool roadmaps obsolete.

Prompt caches shave latency and cost, but their TTL varies wildly: Anthropic drops after ~5 min, OpenAI after ~10 min, and Google after ~15 min. A 30‑second keepalive ping wastes up to eight‑times more money, while a 4‑minute interval saves on Anthropic and avoids needless traffic on others.

Expedia Group launched STAR, an AI‑powered observability platform that parses service telemetry to surface root‑cause clues in seconds. Engineers can now cut mean time‑to‑recovery on production outages by up to 50%, turning raw logs into actionable insights without manual digging.

Single AI agents stumble at enterprise scale because their plumbing, authentication, idempotency, and connectivity, breaks under dozens of tool integrations. Red Hat’s Model Context Protocol gateway gives each agent a cryptographic identity and a governed endpoint, turning custom connectors into reusable, policy‑enforced tool catalogs.

Traditional diagrams show where packets travel, not who can act on them. Modern environments weave identities, OIDC policies, service accounts and CI/CD secrets into a trust graph that attackers exploit. Ignoring this graph means troubleshooting the wrong picture and missing critical breach vectors.

A new multi‑agent AI architecture, leveraging A2A messaging and MCP orchestration, lets 5G core operators auto‑update detection rules in real time, eliminating the biggest SOC bottleneck. The system keeps the security detection engine fresh without constant human triage, boosting response speed and reliability for telecom networks.