AVM.io

Observability and Security Are Converging in the AI Era

Datadog has announced more than 100 new capabilities aimed at helping organizations manage the growing complexity of AI, security, and modern cloud operations. The company introduced these updates at its DASH event, focusing on autonomous operations, better visibility, and faster issue resolution across the entire software and AI lifecycle.

One of the biggest highlights is the expansion of Bits AI, Datadog’s AI-powered operational assistant. Bits AI can now automatically detect, investigate, and even resolve infrastructure and development issues with minimal human intervention. It also supports AI agents, helping teams debug and monitor agent behavior while integrating directly into tools like Slack and Claude.

Datadog also launched AI Guard, a new security capability designed to detect and block malicious AI agent behavior, including prompt injection and agent manipulation attacks. Another key announcement was Bring Your Own Cloud, which allows customers to process and store observability data inside their own cloud environments to improve visibility and control costs.

The broader message is clear: as AI systems become more autonomous, organizations will need stronger observability, governance, and security to manage operational complexity at scale.

Do you think autonomous operations will become the new standard for enterprise IT and security? Share your thoughts below.

The Future of AI Won’t Just Depend on Models—It Will Depend on Infrastructure

AWS has officially launched Graviton5, its newest custom-built processor designed to support the growing demands of agentic AI and modern cloud workloads. According to AWS, Graviton5 delivers up to 25% better performance than the previous generation, with improvements across web applications, machine learning inference, databases, analytics, and high-performance computing workloads.

What makes this especially important is the focus on infrastructure for the next generation of AI systems. Agentic AI applications require real-time reasoning, orchestration, and the ability to manage many simultaneous tasks efficiently. Graviton5 addresses this with 192 CPU cores, faster memory, lower inter-core latency, larger cache sizes, and improved network and storage bandwidth.

Major companies such as Meta, Uber, Snowflake, Airbnb, SAP, and Atlassian are already using or testing Graviton5 for large-scale AI and enterprise workloads. AWS is also emphasizing energy efficiency and stronger cloud security through its Nitro Isolation Engine, designed to provide mathematically verified workload isolation.

The broader message is clear: the future of AI will not depend only on better models. Infrastructure, performance, scalability, and efficiency are quickly becoming equally important competitive advantages.

Do you think AI infrastructure will become as strategically important as AI models themselves? Share your thoughts below.

We've kicked off an internal AI Engineering Book Club.

Over the next ~12 weeks, a cross-functional group from our teams (engineers, QA, DevOps, etc) is reading AI Engineering: Building Applications with Foundation Models by Chip Huyen (O'Reilly, 2024). One chapter a week, one hour together.

From foundation models and RAG to agents, fine-tuning, evaluation, and inference optimization, we're working through what it really takes to ship production AI systems and connecting every chapter back to the real problems we're solving day to day.

No ML background needed. Just curiosity and a willingness to dig in.One hour a week, leveling up together.

AVM is at the Databricks Data + AI Summit 2026 this week. If you’re here and would like to connect, let’s meet. Looking forward to seeing customers, partners, and friends across the data and AI community.

The Next Big AI Opportunity May Be Hidden in Industrial Operations...

Shell is expanding its partnership with C3 AI to move beyond basic predictive maintenance and toward fully autonomous AI-driven maintenance operations. Today, Shell already uses the C3 AI Reliability Suite to monitor more than 30,000 critical assets across its operations. The next step introduces AI agents that can not only detect equipment problems but also investigate root causes, create work orders, check inventory, and even initiate procurement requests with minimal human involvement.

Previously, predictive maintenance systems mainly focused on anomaly detection—alerting engineers when equipment behavior looked unusual. Now, agentic AI is helping automate the entire workflow around those alerts. By combining real-time sensor data with operational and business systems such as SAP, the AI agents can understand context, recommend actions, and eventually automate certain maintenance decisions.

The goal is to reduce downtime, improve operational efficiency, extend equipment life, and increase safety while lowering maintenance costs. This marks a major shift from AI systems that simply provide insights to systems that can independently take action.

Do you think agentic AI will become the standard for industrial operations and predictive maintenance in the coming years? Share your thoughts below.

The Future of Cybersecurity May Depend on Browser Visibility

The State of Browser Security Report 2026 from Keep Aware highlights a major shift in enterprise cybersecurity: the browser has become the primary workplace and, increasingly, the primary attack surface. Employees now spend most of their time working through SaaS applications, AI copilots, GenAI tools, and browser-based workflows, making the browser far more than just a web access tool.

The report explains that traditional security approaches such as network controls, endpoint protection, and static policies are struggling to keep up with how work actually happens today. AI-powered browsers and embedded AI assistants are introducing new risks because they can read, write, summarize, upload, and share sensitive data autonomously. At the same time, attackers are increasingly targeting browsers through phishing, malicious extensions, session hijacking, OAuth abuse, and AI-driven social engineering attacks.

Keep Aware’s findings show that many employees use personal AI accounts for work-related activities, often exposing sensitive company data outside corporate visibility and governance. The report also warns that browser extensions and AI agents are creating new security blind spots that traditional tools cannot fully detect.

The key takeaway is clear: organizations must treat the browser as a critical security control point with its own visibility, monitoring, and governance model.

Do you think enterprises are ready for a future where the browser becomes the new security perimeter? Share your thoughts below.

‍

Amazon Is Solving the Hardest Problem in AI: Trust...

Amazon is betting that the future of AI isn't better chatbots, it's reliable AI agents that can plan, reason, and take action with minimal human involvement.

Unlike traditional generative AI, agentic AI is designed to execute multi-step workflows, adapt to changing conditions, and deliver real business outcomes. To make this possible, Amazon combines foundation models, purpose-built infrastructure, custom AI chips, and its Nova Act platform, which trains reasoning, orchestration, and tool usage as a single integrated system.

A key focus is reliability. Amazon uses large-scale simulated training environments where AI agents learn through trial and error before operating in real-world scenarios. The goal is to achieve the trust and consistency businesses need for mission-critical processes.

The results are already emerging. Organizations are reporting significant productivity gains, while Amazon is using AI agents internally to automate billions of compliance checks and accelerate software development.

The next phase of AI may not be about generating content, it may be about getting work done. How do you see agentic AI transforming your industry? Share your thoughts below.

Project Glasswing Reveals a New Cybersecurity Reality...

Project Glasswing is demonstrating how quickly AI is changing cybersecurity. Anthropic’s restricted AI model, Claude Mythos, scanned more than 1,000 open-source projects and identified 23,019 vulnerabilities, including over 6,200 classified as high or critical severity. Independent validation confirmed that more than 90% of sampled findings were genuine issues. The results helped organizations uncover major security flaws, improve browser security, and even prevent a fraudulent $1.5 million bank transfer.

However, the most important takeaway is not the number of vulnerabilities found it’s the fact that fewer than 1% have been patched so far. AI is now capable of discovering security weaknesses far faster than organizations can analyze, prioritize, and remediate them. In other words, the bottleneck in cybersecurity is shifting from finding vulnerabilities to fixing them.

This highlights a new reality for security teams: vulnerability management, patching processes, and supply chain security may soon become more critical than vulnerability discovery itself. As AI-powered security capabilities continue to advance, organizations must prepare for a future where both defenders and attackers have access to increasingly powerful tools.

Do you think enterprises are ready for a world where AI finds vulnerabilities faster than humans can fix them? Share your thoughts below.