Skip to main content
Vibgrate Graph: over 26% fewer tokens in our benchmark

May 2026

Blog posts from May 2026.

AI & ModelsMay 20, 20266 min read

Gemini 3.5 Lands: Agentic Tool‑Use Meets Million‑Token Context for Migration-Scale Refactors

This week’s Gemini 3.5 launch pushes “agentic” from a demo buzzword toward something migration teams can actually operationalize: models designed to plan, call tools, and iterate. Even more consequential for modernization work, Gemini 3.5 Flash shows up with a 1M‑token context window—large enough to reason over multi-module subsystems, migration runbooks, and dependency graphs in a single pass.

Vibgrate
AI & ModelsMay 16, 20269 min read

Codex from your phone—without chaos: real-time approve/steer controls for safer change governance

Coding agents don’t stop when you leave your desk—and now oversight doesn’t have to either. With Codex accessible in the ChatGPT mobile app, engineering leaders can monitor, steer, and approve work running in remote environments in real time. This post translates “mobile agent control” into practical guardrails for change control, incident response, and modernization workflows.

Peter Chapman
Cloud MigrationMay 15, 20268 min read

Stop Breaking Prod with Prompt Drift: Migrate and Evaluate LLM Prompts Like Code with Amazon Bedrock Advanced Prompt Optimization

As LLM features move from prototypes to production, prompts become a high-churn dependency—more like schemas and configs than “just text.” Amazon Bedrock’s Advanced Prompt Optimization and migration tooling helps teams optimize prompts for a current model or migrate them to new models faster, with built-in evaluation feedback loops to reduce friction and risk when prompts or models change.

Peter Chapman
SecurityMay 14, 20268 min read

After the TanStack supply-chain incident: a pragmatic playbook for cert rotation, package trust boundaries, and smaller CI blast radii

The TanStack supply-chain attack was a reminder that modern dependency ecosystems can propagate compromise at ecosystem speed. Using OpenAI’s published response as a concrete reference point, this playbook lays out practical steps for rotating code-signing certificates, tightening package trust boundaries, and reducing CI/CD blast radius—especially during modernization and high-churn upgrade cycles.

Luke Geaves
DevOpsMay 12, 20268 min read

Regression Tests That Don’t Lie: Capture Real API Behavior to De-Risk Modernization and Stop Contract Drift

API regressions rarely come from the code you changed—they come from the behaviors you didn’t know you relied on. By capturing real API behavior from production-like traffic and replaying it against refactors, you can detect contract drift and edge-case breakages before they ship, without inflating a brittle test suite.

Peter Chapman
Data EngineeringMay 9, 20268 min read

dbt Developer Agent (Preview): Safer Analytics Refactors Grounded in Your dbt Project

Analytics codebases break for the same reasons application code does: hidden dependencies, rushed refactors, and brittle interfaces. dbt Developer Agent (now available in Preview) aims to make analytics engineering changes safer by grounding suggestions in your dbt project—helping teams ship faster without breaking downstream models.

Peter Chapman
DevOpsMay 8, 20269 min read

Streaming-First CI “AI Steps” with WebSockets: Lower Latency, Fewer Timeouts, Better Logs, and Predictable Cost

AI steps inside CI/CD often fail for the same reasons as flaky integration tests: slow feedback, brittle timeouts, and poor observability. With OpenAI introducing a WebSocket-based execution mode aimed at reducing latency in agentic workflows, teams can redesign “AI steps” to be streaming-first—improving responsiveness, failure handling, and cost control without sacrificing reproducibility.

Peter Chapman
Cloud MigrationMay 7, 20268 min read

Managed MCP for AWS: Standardizing Agent Access with Least Privilege, Auditing, and Fewer Bespoke Integrations

As AI coding agents move from side projects into real operational workflows, the biggest risk isn’t capability—it’s uncontrolled cloud access. AWS’s newly GA AWS MCP Server offers a managed, authenticated path for agents to interact with AWS services, helping modernization teams reduce integration sprawl and bring “shadow agents” back under governance.

Luke Geaves
AI & ModelsMay 6, 20267 min read

400K Context Arrives: Using This Week’s New Chat Models to Modernize Large Codebases Without Losing the Plot

This week’s releases are all about scale and reliability: longer context windows for repo-wide reasoning, and a new “Instant” default model tuned to be more accurate with fewer hallucinations. For migration teams, that combination is practical—not flashy—because it targets the two things that derail modernization work most often: missing cross-file dependencies and untrustworthy refactor suggestions.

Vibgrate
Page 1 of 2