host profile

Null Frequency

Mira Voss is a fictional independent developer and former cloud security engineer. Each week, she turns fast-moving AI agent, cloud infrastructure, developer tooling, and security engineering articles into compact podcast briefings.

Callsign mira-null, recording remotely from a late-night terminal desk with a bias for low-latency systems and observability.

Stack Rust / Go / TypeScript / Bedrock / Kubernetes / eBPF / PostgreSQL.

Format One engineering question per episode: architecture, risk, migration path, and team adoption cost.

episode queue

Recent Signals

Curated from official engineering blogs and product teams, this queue tracks 2026 shifts in AI agent operations, developer tools, browser automation, cloud runtimes, and security governance.

EP.01 / 2026-07 DEVTOOLS

From Issue to PR: GitHub Copilot Coding Agent as Async Engineer

GitHub's 101 guide frames Copilot coding agent as a background software engineering agent: assign an issue, let it work in a GitHub Actions environment, and review the draft PR it produces.

Host note: teams need smaller, testable task packets for agents, with clear acceptance checks instead of broad product asks.

GitHub Blog Read source

EP.02 / 2026-07 AGENTOPS

Production AI Teammates: monday.com's AgentOps Playbook

The monday.com case study breaks production agents into identity, orchestration, tool calls, monitoring, and human approval paths. The message is blunt: an agent is not a chat box, it is an operated service.

Host note: useful launch metrics include task success rate, escalation rate, permission hits, tool failures, and human intervention cost.

AWS ML Blog Read source

EP.03 / 2026-06 SECURITY

Financial Compliance Agents: Lessons from Stripe's Production Boundary

This case study focuses on production AI agents in financial compliance, where tool permissions, data boundaries, audit trails, and controls for high-risk actions matter more than raw model capability.

Host note: the closer an agent gets to business closure, the more it should be designed like a privileged service account.

AWS ML Blog Read source

EP.04 / 2026-06 MCP

MCP Gateway and Registry: A Control Plane for Agent Tools

AWS's open source post places MCP gateways, registries, federated discovery, access control, and security scanning into a shared governance plane for agent tools.

Host note: when tool count grows, discovery, authorization, versioning, isolation, and quotas become bottlenecks before schema design does.

AWS Open Source Blog Read source

EP.05 / 2026-05 SAFETY

Running Codex Safely: How Coding Agents Limit Side Effects

OpenAI describes the safety model around Codex environments, task isolation, and user approvals, separating coding-agent risk across network access, filesystem writes, command execution, and code review.

Host note: the default posture for developer agents should be least privilege, reproducible environments, and explicit human confirmation points.

OpenAI Read source

EP.06 / 2026-04 WEB

AI Labyrinth: Slowing Unwanted Crawlers with Dynamic Decoys

Cloudflare's approach generates a maze for suspicious crawlers, wasting their time while reducing disruption to real site resources. Anti-bot work is moving from static rules toward behavioral traps.

Host note: defenses against automated traffic are becoming generated, observable, and tunable engineering systems.

Cloudflare Blog Read source

EP.07 / 2026-07 OPS

AWS Support Companion: Incident Triage Inside an Agent Loop

AWS uses Bedrock AgentCore, Strands Agents, MCP servers, CloudWatch, re:Post, and the Support API to connect log analysis, documentation retrieval, and support-case creation.

Host note: in SRE workflows, agent value comes from collecting cross-console evidence and turning it into an auditable support path.

AWS ML Blog Read source

EP.08 / 2026-05 CLOUD

Google Cloud's Agent Ladder: From Low-Code Studio to ADK 2.0

Google Cloud's I/O article connects Agent Studio, Managed Agents API, Antigravity, Agent CLI, and ADK 2.0 into one development path for prototyping, sandboxing, governance, and deployment.

Host note: enterprise agent platforms are becoming lifecycle platforms, tying IDEs, local CLIs, cloud sandboxes, and runtimes to shared identity and audit systems.

Google Cloud Blog Read source

EP.09 / 2026-05 WEB

Chrome at I/O 2026: The Agentic Web Gets a Browser Layer

The Chrome team connects built-in browser AI, Prompt API, WebMCP, and agentic browsing. The direction is clear: pages should expose structured affordances instead of forcing agents to infer everything visually.

Host note: future frontend accessibility will serve both human assistive tech and machine agents. Stable semantics may become part of CI.

Chrome Developers Read source

EP.10 / 2026-02 APPSEC

Claude Code Security: AppSec Moves Toward Contextual Reasoning

Anthropic positions Claude Code Security as an AI-assisted scanner that reads code, follows component interactions, validates findings, and proposes patches for vulnerabilities static rules can miss.

Host note: AI AppSec shifts the question from finding issues to proving risk, reducing false positives, prioritizing fixes, and keeping final patches human-approved.

Anthropic Read source

recording log

Recording Log

MON 23:12 Split the AgentOps episode into identity, tools, memory, and rollback. Removed low-signal model parameter notes.

WED 01:04 Recorded an extra MCP gateway segment with notes on tenant isolation, registry drift, and tool version governance.

THU 09:40 Refreshed the source queue on July 23, 2026. Working title for the next cut: "Treat Agents Like Production Services."