Operations | Monitoring | ITSM | DevOps | Cloud

If you are able to build it better than your vendor, then change your vendor. Not build it. Rishi builds large-scale observability systems at Last9, focusing on reliable and cost-efficient telemetry infrastructure, and writes about the practical lessons learned while operating ClickHouse, VictoriaMetrics, and OpenTelemetry in production.

A team ships at a steady pace for most of the year. Then an audit approaches, and delivery slows. Engineers get pulled off feature work to support the audit, producing the configuration exports, logs, and environment checks that the evidence depends on. The slowdown lasts as long as the audit does. It is tempting to read this as a team that needs to move faster or be bigger. It is usually neither.

LLM token cost is the price a provider charges per token a model reads or writes, quoted in dollars per million tokens. Input and output bill at separate rates, with output priced at roughly 5x input. As of September 2026, published rates range from under $0.10 to more than $180 per million tokens on top-end reasoning tiers. In late 2025, Hardik Sonetta of Thomson Reuters Labs published a warning about the most common prompt caching mistake in production.

Changing a Kubernetes resource request takes two numbers: what the workload requests, and what it uses. The CloudZero MCP server now returns both, by cluster, namespace, or workload. This gives you a number you can defend. Usage comes back as P95 over the date range you query, 30 days by default. When an engineering lead asks whether a service runs on a smaller request, that is the figure that settles it. Over-provisioning and under-provisioning show up on the same query.

Pop quiz: If you spend $40,000 per month on Anthropic, and you’ve got two customers, what’s your cost per customer? If you bypassed the easy answer of $20,000 and said, “Scott, you old trickster, that’s not enough information to answer that question,” you’ve won today’s prize: a lesson in the perils of average costs. Let’s flesh out the situation: You put an AI feature in your product, a document assistant powered by Claude.

A log line arrives as plain text and leaves as a record you can query. Six steps sit between those two states. Each one adds something useful, and each one costs you time, CPU, or storage. Most teams never look at that chain until a search comes back empty. Here is what log processing does to an event, step by step: By the end you can look at your own chain. You will know what each step buys you. Six steps turn a raw log line into a searchable record.

Sensitive values reach your logs without anyone deciding they should. A debug line prints a whole request object. An error message carries the query string. A customer email address is suddenly stored in three systems. PII redaction in logs then gets treated as one setting to switch on. In practice it covers four separate treatments. The value is already inside the message before log ingestion finishes. In this blog, you will: By the end you can write a rule for each field and defend it.

In a single morning, a coding agent can pull an open-source model, connect to an unvetted MCP server, and execute a code-optimizing skill from the web. In the rush toward agentic automation, these AI assets quietly bypass traditional security reviews, creating new attack vectors across the software supply chain. Closing this blind spot has been the driving force behind the JFrog AI Catalog since its launch at swampUP 2025.

Learn how NaaS helps MSPs add connectivity services and respond faster to customers without building a global network. For many Managed Service Providers (MSPs), connectivity is still the part of a customer solution they control the least. An MSP may manage the cloud environment and secure access to it, but adding a circuit can still mean carrier lead times and manual coordination. Options may also narrow when the customer enters a new market. This reality doesn’t match with customer expectations.

Datadog’s 2026 State of AI Engineering report showed organizations’ LLM inputs swelling rapidly as context engineering expands. In March 2026, 69% of all input tokens in Datadog customer traces were for system prompts: internal instructions, policy definitions, and tool guidance providing context and guardrails around the user input. This suggests that most context engineering spend among Datadog customers is going toward optimizing repeating system prompts in heavily scaffolded agent systems.