AI Agent Identified 14% Cloud Cost Savings in One Hour, But Missed Critical Context

·1 views

A developer ran an AI agent with read-only access to AWS and GCP billing data, giving it a single prompt to find ways to cut cloud costs by 20% without disrupting services. Within an hour, the agent flagged idle resources including unattached storage volumes, oversized database instances, redundant NAT gateways, and staging environments running overnight with zero traffic, projecting roughly 14% in monthly savings. However, the agent also produced dangerous recommendations, such as deleting disaster-recovery snapshots it misclassified as orphaned and downsizing an instance it believed was CPU-idle but was actually memory-constrained. It also proposed purchasing reserved instances for workloads that were weeks away from being migrated to a different platform. The experiment highlighted both the strength of AI agents in tedious discovery tasks and their key limitation: they reason accurately from available data but cannot account for intent, tribal knowledge, or context stored outside their data sources.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Developer builds Rust-native WMI exec tool by reverse-engineering impacket byte-by-byte

A developer has documented building wmiexec from scratch in Rust as part of ADhammer, a self-described Active Directory audit and validation toolkit. The implementation required reconstructing DCOM activation, OXID resolution, and MS-WMIO object marshaling without relying on existing libraries. Rather than using impacket as a dependency, the developer treated it as a debugging reference, comparing wire-level bytes between the two implementations until a Windows domain controller accepted the output. The entire project is built on a custom DCE/RPC, NTLM, SMB2, and Kerberos stack written in Rust, with Claude Code used as an AI development assistant throughout the process.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

React Forms Explained: Controlled Components, Validation, and React Hook Form

Day 24 of the React Mastery Series covers how to build and manage forms in React, a skill considered essential for frontend developers. The article contrasts controlled components, where React state drives input values, with uncontrolled components that rely on the DOM directly. It demonstrates how grouping multiple fields into a single state object improves scalability over managing separate state variables for each input. React Hook Form is introduced as a library that reduces boilerplate code and minimises re-renders, making it better suited for large forms with many fields. The piece also touches on schema-based validation using Zod, which allows developers to define and centralise validation rules in enterprise-grade applications.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Deploy an MCP-Based AI Agent to Production Using Kubernetes and CI/CD

Deploying a Model Context Protocol (MCP)-based AI agent beyond a local environment requires a structured production architecture covering containerization, orchestration, and observability. The proposed blueprint routes code through GitHub and GitHub Actions into a container registry, ultimately running on a Kubernetes cluster connected to MCP servers and external services. Docker is used to package the agent with a lightweight, non-root runtime, keeping credentials outside the image entirely. Kubernetes manages deployment, scaling, restarts, and rolling updates, while secrets management tools protect API keys and access tokens. Observability tooling — covering logs, metrics, traces, and alerts — provides the visibility needed to maintain reliability once real users and external dependencies are involved.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Git Merge, Rebase, and Cherry-Pick: How Each Strategy Shapes Your Repo History

Git offers three primary strategies for integrating code changes across branches: merge, rebase, and cherry-pick, each with distinct trade-offs. Merging combines branch histories by creating a dedicated merge commit, preserving the exact timeline but potentially cluttering the graph. Rebasing replants a branch's commits on top of another branch's latest commit, producing a clean linear history at the cost of rewriting commit history, making it risky on shared public branches. Cherry-picking copies a single specific commit onto a target branch, which is useful for applying hotfixes without importing unfinished work, though it results in duplicate commits with different hashes. Choosing the right strategy depends on team workflow, branch visibility, and how much historical accuracy matters to the project.

0 comments Read more at DEV Community