Plumbing the
information age

Dynatrace (NYSE: DT) announced major advancements to Dynatrace Intelligence that help automatically resolve incidents, prevent disruptions, and accelerate operations while maintaining the human oversight and governance enterprises require. Building on the introduction of Dynatrace Intelligence earlier this year, Dynatrace is adding new autonomous agents for incident triage and remediation, and no-code custom agent creation capabilities. The release reflects the industry momentum toward agentic AI in observability—moving beyond passive anomaly detection into systems that can propose and execute remediation with human guardrails. The platform is also expanding its ecosystem of integrations, bringing insights directly into the tools and workflows teams already use. This approach addresses a key operational friction point: getting AI-generated insights into the incident response loops where engineers already live (ticketing systems, chat platforms, runbook automation). For network and SRE teams managing multi-cloud estates with complex interdependencies, autonomous triage reduces the initial human processing load that typically adds 20–30 minutes to incident lifecycle.

On May 21, 2026, the Model Context Protocol (MCP) maintainers locked the release candidate for the largest specification update since the protocol launched. The July 28 final release represents a major shift in production readiness for agentic systems. Removing session handshakes and supporting stateless operations behind load balancers eliminates sticky-session infrastructure requirements and enables horizontal scaling. Six security proposals align authorization with OAuth 2.0 and OpenID Connect, easing enterprise security reviews. For infrastructure operators, this means MCP servers can now deploy as standard microservices without session affinity, simplifying multi-region deployments and disaster recovery. Roots, Sampling, and Logging are deprecated with a guaranteed twelve-month transition window before removal, giving teams time to migrate legacy implementations.

The tutorial builds an advanced workflow that maps Anthropic's financial-services repository components (agents, vertical plugins, partner integrations, managed-agent cookbooks, and financial analysis skills) into a reusable SkillAgent pattern using pure Python, while injecting selected financial playbooks into the Anthropic Messages API and supporting iterative tool-use loops for calculations and file generation. This practitioner-focused approach shows how to structure multi-agent financial workflows with auditable decision paths, a pattern increasingly critical for compliance-heavy domains. The skill-driven architecture separates domain logic from orchestration, enabling teams to test agents against constraint sets and audit trails essential for regulated environments. For SRE teams operating multi-agent systems in finance and regulated sectors, this demonstrates how to build agent workflows with explainability and governance built into the architecture.

AMD's Helios platform combines 72 Instinct MI455X GPUs, 6th Gen EPYC CPUs, and Pensando networking, aiming to compete with NVIDIA's offerings through 15% more AI compute, 50% greater HBM capacity, and 50% higher scale-out bandwidth for large-scale AI inference and training deployments. Benchmark results show 174% geomean uplift on agentic AI workloads versus Intel Xeon 6980P, with expanded DDR5 MRDIMM, PCIe Gen 6, and innovations for complex enterprise AI pipelines. AMD and Cerebras Systems announced a partnership combining Helios rackscale systems with Cerebras Wafer-Scale Engine for AI inference, targeting ultra-low-latency workloads with up to 5x more tokens per second per watt. For infrastructure operators deploying large-scale agentic systems, this signals competitive momentum in AI infrastructure and provides an AMD-based alternative for multi-agent deployments.

The initial part of Extreme's Agent ONE launch is the introduction of Agent ONE Coworker in July 2026, an AI agent designed to work alongside IT teams and deliver proactive, context-aware intelligence with real-time decision-making and automated execution at machine speed. Unlike traditional AI tools that require explicit prompts, Agent ONE Coworker operates proactively, providing insights and guidance within workflows, with capabilities described as ambient (work already completed), proactive (no waiting for input), and helpful (deep understanding and execution). This positions Extreme to compete in the multi-agent networking operations space alongside Cisco's AgenticOps framework. For network operators, the shift from reactive tooling to continuously operating agents represents a fundamental change in operational model—moving from humans triaging alerts to agents detecting and acting on patterns autonomously.

RCR Wireless published a brief but substantive commentary on AT&T's strategic pivot in network AI deployment. The article argues that AT&T's approach emphasizes intelligent orchestration and coordination of AI workloads across the network rather than chasing ever-larger foundation models. This matters to network ops practitioners because it suggests telcos are moving away from the "bigger is better" LLM mentality toward practical, distributed inference architecture tied to network topology and edge compute. The implication is that telcos with existing fiber and cell site infrastructure can leverage that physical footprint as a competitive advantage for running AI inference closer to customers and data sources, reducing backhaul and latency. This aligns with broader industry signals that inference-as-a-workload is becoming a core network operations function, not just a cloud service.

India's Digital India Act establishes a risk-based classification system mirroring the EU model but with stricter liability provisions: minimal, limited, high-risk, and critical AI categories. High-risk systems in healthcare, critical infrastructure, and credit decisions require conformity assessments before deployment. Critically, strict liability applies to critical AI categories—no negligence standard required, stronger than the EU AI Act. Mandatory registration with the Ministry of Electronics and Information Technology (MeitY) and prohibition on AI systems that manipulate users through subliminal techniques or exploit psychological vulnerabilities. The UK's AI Regulation and Safety Bill introduces legal duty of care for frontier AI model developers above a defined compute threshold, mandatory safety evaluations before deployment, powers for the Secretary of State to issue binding technical standards, and a new UK AI Safety Institute with statutory audit authority. The bill progresses through committee and third reading; enforcement unlikely in 2026 but signals statutory obligations are coming. Globally, 38 US states enacted ~100 AI measures in 2025; Texas TRAIGA (effective January 2026) requires reasonable care, transparency, testing, and impact assessments; California's transparency act becomes operative in August 2026.