Daily Briefing · Jul 2, 2026
Honeycomb Releases Second Edition of Bestseller Observability Engineering to Redefine the Practice for an AI World
Honeycomb published the second edition of Observability Engineering (O'Reilly), nearly doubled in length with 27 new chapters addressing AI-world challenges including instrumentation for AI-assisted development, debugging LLM applications in production, and telemetry pipeline management. The book reflects how software development lifecycle compression into rapid intent-validation loops demands new observability strategies when AI accelerates shipping velocity while complexity grows.
ServiceNow Expands AI Control Tower with Multi-Cloud Discovery, Runtime Observability, and AI Agent Governance
ServiceNow expanded its AI Control Tower with five new dimensions—Discover, Observe, Govern, Secure, Measure—adding 30 new enterprise integrations across AWS, Azure, Google Cloud, SAP, and Oracle. The Observe capability, built on Traceloop acquisition, delivers runtime monitoring of AI agents with live visibility into reasoning and decision paths, replacing periodic audits with continuous production observability.
Exaforce Raises $125M Series B to Automate Security Operations with Agentic AI Detection and Response
Exaforce closed $125M Series B at $725M valuation to scale its agentic SOC platform built on 'Exabots' that automate detection, triage, investigation, and response with deep data analysis. The platform claims 90% reduction in manual tasks and introduces 'vibe hunting'—natural language queries for hypothesis-driven threat investigation—marking shift from reactive monitoring to AI-driven autonomous security operations.
Chronosphere Launches AI-Guided Troubleshooting with Temporal Knowledge Graph to Reduce False Positive AI Recommendations
Chronosphere released AI-Guided Troubleshooting capabilities using a Temporal Knowledge Graph embedded in its platform to ensure LLMs only process relevant telemetry data, generating deterministic rather than probabilistic suggestions. The approach addresses 'confident-but-wrong guidance' by limiting context exposure and is available via MCP Server for AI development tools, with general availability planned for 2026.
Gartner Renames AIOps Category to Event Intelligence Solutions, Reflecting Market Maturation Beyond Alert Noise Reduction
Gartner redefined the AIOps market as Event Intelligence Solutions in 2025, shifting focus from autonomous operations to cross-domain event processing and incident lifecycle automation. The reclassification acknowledges that alert correlation and knowledge retrieval remain production-validated while fully autonomous remediation is still aspirational, with agentic AI emerging as the next wave—AWS DevOps Agent and Azure SRE Agent reached GA in March 2026.
Selector Raises $32M and Launches AI-Powered Multi-Cloud Observability Closing Network-to-Cloud Visibility Gap
Selector expanded its AIOps platform to deliver multi-cloud observability with AI-driven correlation across infrastructure, apps, and services. The platform unifies logs, metrics, configs, and topology into a single AI layer for root cause analysis and automatic incident routing to ITSM, with a domain-specific Network Language Model (NLM) enabling plain-English investigation and remediation in Slack, Teams, CLI, or UI.
IBM: AIOps Observability Practice—Combining AI Capabilities with Observability Strategy to Automate IT Operations
IBM published analysis of AIOps observability practice, demonstrating that 71% of organizations using observability solutions now employ AI features (up from 26% in 2024). AIOps applies NLP and ML to automate collection/analysis of telemetry data, with modern platforms including generative AI for text interfaces, predictive downtime forecasting, and automated remediation tools integrated directly into observability dashboards.
OpsPilot: Practical AIOps in 2026 Achieves Manual Work Reduction Without Enterprise Complexity Using OpenTelemetry Data
OpsPilot articulated distinction between enterprise AIOps vision (autonomous operations) and practical AIOps reality: reduction of manual work at costly, replaceable points. Achievable AIOps for 20-100 engineer teams requires continuous pattern analysis, proactive recommendations, and gap detection from OpenTelemetry data—not AI-as-presentation-layer that forces engineers to investigate every alert manually.
Claude Sonnet 5 Launch: Near-Opus Agentic Performance at Mid-Tier Pricing
Anthropic released Claude Sonnet 5 on June 30, 2026, a more powerful and agentic version of its midsize model that can make plans, use tools like browsers and terminals, and run autonomously at a level that previously required larger and more expensive models. Priced at $2/$10 per million tokens through August 31, then $3/$15 thereafter, Sonnet 5 scores 63.2% on agentic coding versus Opus 4.8's 69.2%, while slightly outperforming Opus on knowledge work benchmarks—making it the default model for Free and Pro users.
MCP 2026-07-28 Release Candidate: Protocol Goes Stateless for Production Scale
The MCP specification release candidate (locked May 21, final July 28, 2026) removes the initialize handshake and Mcp-Session-Id header, making the protocol stateless at the HTTP layer. MCP servers can now run behind plain round-robin load balancers without sticky sessions or shared session stores. The spec includes MCP Apps and Tasks extensions, OAuth hardening, formal deprecation policy, and routing via Mcp-Method/Mcp-Name headers for infrastructure-level gateway routing without payload inspection.
Cisco Rolls Out AI Agents to 90,000 Employees with Dynamic Model Selection
Starting end of July 2026, Cisco is deploying personalized AI agents to all 90,000 employees with dynamic model routing that selects the most efficient model per task to control token burn and costs. The architecture is primarily on-premises, built on Cisco's own AI stacks that query different models based on use case. CFO Mark Patterson reports the system handles tasks like benchmarking competitive metrics and MD&A document drafting, with company-wide upskilling efforts driving adoption.
MCP Release Candidate Eliminates Session State: Breaking Changes and Migration Path
The 2026-07-28 MCP release candidate (locked May 21) removes protocol-level sessions, the initialize handshake, and Mcp-Session-Id pinning. All Streamable HTTP requests must now include Mcp-Method and Mcp-Name headers for infrastructure-level routing. State-carrying applications shift to explicit handles (basket_id, job_id) that models pass through tool calls. Teams have until July 28 to migrate; SDK 1.3.0 already supports the explicit-handle pattern for forward compatibility.
Agentic AI Frameworks 2026: LangGraph vs CrewAI Performance on Orchestration Design
LangGraph and CrewAI are the two most widely deployed agentic AI frameworks in 2026 but serve fundamentally different use cases. Princeton's Holistic Agent Leaderboard shows the same Claude Opus 4 model scores 64.9% in HAL's scaffolding versus 57.6% in HuggingFace's, a 7-point gap from orchestration alone. The gap between a bare model and well-engineered scaffold reaches ~30 points on GAIA—larger than most model generation deltas—proving orchestration architecture matters more than model choice.
Claude Managed Agents Add Scheduled Execution and CLI Tool Authentication via Vaults
Claude Managed Agents now support scheduled execution (public beta) and secure CLI tool integration via environment variable vaults. Agents can run on cron schedules for recurring work (nightly syncs, weekly compliance scans). Credentials are stored in vaults and injected at the network boundary only on approved domains, so agents never see raw keys. Integration extends to Browserbase, Notion, Sentry, and Ramp CLIs for authenticated requests without exposing secrets.
Top Agentic AI Frameworks 2026: LangChain, AutoGen, CrewAI Benchmarked Across 2000 Runs
Benchmark across 2000 runs of LangChain, AutoGen, CrewAI, and other frameworks shows LangGraph achieves fastest latency with lowest token usage across tasks, while LangChain has highest latency and token consumption. AutoGen leads on latency; CrewAI balances practical workflow needs. The research validates that framework choice—not just model—drives agent performance, with latency, token consumption, and architectural differences materially affecting production costs and response times.
Networking Predictions in 2026: From Automation Experiments to Agent-Driven Operations
Itential's John Capobianco articulates the shift from script-driven automation to agentic orchestration, documenting how MCP, AI agents, and centralized orchestration are now foundational infrastructure for network operations. The FlowAI platform enables external agents (NetBox MCP, Selector MCP) to participate in orchestrated workflows under governance, with 834+ annual hours saved and 6-month payback periods documented across implementations.
HPE Expands Self-Driving Networks Across Edge, Campus, Data Center, and AI Factories
HPE announced major networking advancements at Discover 2026, integrating Juniper Networking QFX switches into HPE AI Data Center Solutions with agentic AIOps capabilities. New QFX5140 and QFX5252 switches provide AI-optimized switching for inference and training workloads, with expanded Marvis AI automation across Aruba Central and Mist platforms, enabling closed-loop remediation across wired, wireless, and SD-WAN without human intervention.
HPE targets 'low-hanging fruit' in self-driving networks push
HPE demonstrates pragmatic autonomy in network operations by targeting specific, repeatable problem areas rather than full automation. Real-world deployment at U.K. Ministry of Justice shows 75% reduction in service desk tickets through autonomous actions on common issues; automation foundation built on years of telemetry and operational learning, not just agentic AI, with human-in-the-loop options for gradual confidence-building.
LogicMonitor Advances Autonomous IT with No Blind Spots, Trusted AI, and Closed-Loop Action
LogicMonitor expanded its observability platform with Edwin AI agentic investigations, Catchpoint Advisor synthetic test automation, and integrated Real User Monitoring. AI Investigations 2.0 correlates logs, metrics, ITSM records, and knowledge bases to provide explainable root cause; platform emphasizes context-driven reasoning over metrics-only analysis to reduce false positives and build operator trust in AI recommendations.
Network engineers take on NetDevOps roles to advance stalled automation efforts
Analysis of NetDevOps maturation shows 2026 as inflection point where network operations shift from hardware-centric to data-driven, automation-first models. Research highlights that barriers to automation are 44% people-related (skills gaps, organizational dysfunction) vs. only 10% technical; high-performing teams spend 40% more time on design, testing, and validation than reactive troubleshooting, with data quality as foundation for automation and AI enablement.
Arista vs Juniper: Which Is Better for Enterprise Networks (2026)
Deep technical comparison of Arista EOS and Juniper Junos for enterprise deployments shows EOS delivers lower cost and simpler IOS-familiar model for data center VXLAN/EVPN, while Junos commit model, Apstra intent-based automation, and policy expressiveness suit complex multi-vendor environments. Both support OpenConfig and gNMI for vendor-neutral automation; choice depends on team expertise, use case dominance, and automation toolchain investment.
Networking takes center stage at HPE Discover 2026
HPE repositioned networking as strategic foundation for AI enterprises at Discover 2026, combining legacy Aruba and Juniper AI teams into single data sciences group producing agentic mesh framework. New capabilities include CX-series switches management via Mist, Marvis Actions in Aruba Central, integration of Apstra Data Center Director with Morpheus, and zero-trust VXLAN overlays for closed-loop intent-based automation across network and security domains.
Palo Alto, CrowdStrike both have best quarter ever as AI threats bolster cyber demand
CrowdStrike and Palo Alto rallied 95% and 113% respectively between April and June for their best quarter on record, driven by demand for AI-powered cyber defense following Anthropic's Mythos model release. Both companies gained early access as Project Glasswing partners and participate in OpenAI's Daybreak cyber initiative; their growth is anchored in agentic security and identity access management strategies, with Palo Alto's $25B CyberArk acquisition and CrowdStrike's Falcon Shield identity platform achieving 4x ARR growth.
When the GPUs are waiting for the network
Technical analysis of AI data center networking showing Nvidia surpassed both Arista and Cisco in datacenter Ethernet switching revenue (Q1 2026 IDC data), driven by Spectrum-X integration with GPU ecosystem. Examines how network design has shifted from port capacity focus to system-level optimization, with Arista's EOS platform and Broadcom Tomahawk 6 silicon enabling practical AI cluster operations at scale.
Power-Bound, Not GPU-Bound: AI Data Center Power Constraints Are the Real 2026 Bottleneck
Practitioner analysis showing grid power infrastructure has become the primary constraint for AI cluster deployment in 2026, not GPU availability. Gartner projects 40% of AI data centers will be power-constrained by 2027, with 24–36 month grid approval timelines outpacing hardware procurement. Provides concrete metrics: 1,000-GPU H100 cluster draws ~1.76 MW; discusses tokens-per-watt optimization and distributed GPU capacity as workarounds.
OpenTelemetry Graduates to CNCF Maturity Level After Merger Integration
OpenTelemetry moved to the Graduated maturity level on May 11, 2026, marking a critical milestone for observability standardization. This graduation reflects ecosystem maturity in distributed tracing, metrics, and logs collection—directly impacting AIOps and MLOps practitioners building production observability infrastructure for AI and traditional workloads.
Six Google DeepMind Researchers Depart for Meta, OpenAI, and Anthropic Amid AI Coding Race
Six named researchers dispersed to Meta, OpenAI, and Anthropic since February, with Google forming an AI Coding Strike Team in April 2026 tasked with closing Gemini's gap against Anthropic's Claude and OpenAI's coding tools. The departures span reasoning, architecture, and pretraining—core competencies directly affecting frontier model capabilities and infrastructure decisions for enterprises evaluating AI platforms.
AI Infrastructure Investment Boom 2026: Hyperscaler CapEx Reaches $300B+ with Power and Networking Constraints Reshaping Deployment
Tech giants projected to invest over $1.5 trillion in AI infrastructure, with hyperscalers increasing combined annual capital expenditure from $100 billion in 2023 to over $300 billion in 2025, projections exceeding $500 billion. Power availability is becoming a critical constraint, forcing infrastructure operators to distribute accelerator clusters across multiple buildings and campuses, making networking a critical component of AI infrastructure design.
Anthropic Launches Claude Science, Enters AI Drug Discovery Market Against Google DeepMind and OpenAI
Anthropic announced Claude Science on June 30, a major new product for scientific research with access to tools for computational biology and drug development, capable of autonomously carrying out meaningful work. The launch exemplifies three competing distribution strategies: Anthropic going wide with broad subscription access, OpenAI going narrow and enterprise-gated, and Google leaning on owned proprietary models. Anthropic is offering up to $30,000 in credits for up to 50 research projects, with applications closing July 15, 2026.
Anthropic Restores Fable 5 Access After Government Lifts Export Controls; Launches Claude Sonnet 5
On June 30, Anthropic announced the U.S. Department of Commerce lifted export controls affecting Claude Fable 5 and Mythos 5, with Fable 5 restoring access globally on July 1. Anthropic simultaneously launched Claude Sonnet 5 as the default for Free and Pro plans with introductory pricing through August 31, promising Opus 4.8-level performance at substantially lower cost. Two frontier model releases in two weeks have both been gated by government action, establishing a pattern where the Trump administration treats frontier AI models as regulated technology requiring review before broad deployment.
Google Delays Gemini 3.5 Pro to July Amid Talent Departures; Researchers Exit to Anthropic and OpenAI
Gemini 3.5 Pro failed to ship in June 2026 despite CEO Sundar Pichai's commitment at Google I/O; the model remains in limited Vertex AI enterprise preview as of June 30, with Google pushing general availability to July citing quality refinements. Noam Shazeer, co-inventor of the Transformer architecture, left Google for OpenAI, while Nobel laureate John Jumper departed for Anthropic. Alphabet lost approximately $225 billion in market capitalization in a single trading session, driven by the model delay and high-profile talent departures to rival labs.
Verizon CTO details path to Level 4 autonomy: embedded Claude, 70M config changes in 2025
Verizon pilots an Autonomous Network programme with AI agents operating continuously across production infrastructure, monitoring conditions and identifying anomalies. Issues that previously required hours of manual investigation can now be resolved in less than two minutes. Verizon embedded Anthropic's Claude into internal operations, allowing teams to use natural language for network modeling, traffic management, and optimization.
Singtel considers REIT to fund expansion of AI data centers and sovereign AI platforms
Singtel is considering a real estate investment trust (REIT) as a permanent capital vehicle to finance growth in data centres and sovereign AI infrastructure. Capital expenditure is expected to rise to about $3 billion in the year ending March 2027, with $1.2 billion marked for growth in Nxera and RE:AI. The funding strategy supports the operator's shift toward digital infrastructure and technology services beyond traditional connectivity.
South Korea announces $880B national AI infrastructure plan targeting semiconductor and robotics dominance
President Lee Jae-myung announced a 1,350 trillion won 10-year investment targeting semiconductors, AI infrastructure, and robotics as a matter of national survival. Samsung and SK Hynix will invest $518B combined to build four new chip fabrication sites, while SK Group, GS Group, and Naver will back $356B in AI data center construction. This reflects intensifying geopolitical competition over AI capability as nations seek to secure compute resources.
US Department of Commerce lifts export controls on Claude Fable 5 and Mythos 5; Anthropic launches Claude Sonnet 5
The Commerce Department lifted export controls on Anthropic's flagship models, enabling Fable 5 global restoration beginning July 1. Simultaneously, Anthropic launched Claude Sonnet 5 as the default for Free and Pro users, positioning it as near-Opus 4.8 performance at substantially lower cost. This represents the first major frontier model release cleared through what appears to be an emerging de facto government clearance mechanism.
No articles match your filter. Clear filter
Podcasts & Talks · Jul 2, 2026
Notes from NANOG 97
Geoff Huston covers NANOG 97 in Bellevue (June 1-3, 2026), highlighting the conference's heavy focus on AI in network operations, data center design for LLMs, and practical network automation challenges. Discusses SSH connection overhead scaling to 10,000-20,000 devices and the shift toward HTTPS-to-SSH proxy approaches, plus emerging topics in IP geolocation and compliance.
Listen / Watch →