Gemini Workspace Instructions Spread: Keep Them Specific
Google is expanding saved Gemini instructions across Workspace. Build useful defaults, check which ones were used, and handle rollout uncertainty.
Guides
The latest in Guides
19 stories · prices and picks kept current
Copilot Can Approve Pull Requests: Two Switches Matter
Copilot review approvals are in public preview. Separate the assessment, approval, and merge requirement before enabling the feature for a repository.
Copilot Budget Expiry Can Remove a Cap: Check the Fallback
GitHub now lets individual Copilot budgets expire. Learn which limit takes over, how pooled credits differ from overage, and what to check before saving.
Copilot App and CLI Now Honor Exclusions: Check These Gaps
GitHub extended content exclusions to its Copilot app and CLI. Here is how to configure them, test the policy, and spot unsupported workflows.
Firefox AI Controls vs. Smart Window: What to Choose
Block Firefox AI, enable selected features, or try Smart Window. Understand the settings, beta availability, and local-memory privacy distinction.
Gemini Notebook Gets Audit Logs: Export Is Still Separate
Google added Gemini Notebook activity logs for Workspace admins. Check permissions, useful event fields, BigQuery export, and the data-location boundary.
HydraFusion Picks a Workflow, Not Just a Coding Model
GitHub's HydraFusion preview can draft, escalate, or request a critique. Understand its benchmark caveats and choose a useful first coding task.
Google Vids Turns Documents Into Video: Edit the Script First
Google Vids can summarize Docs, PDFs, and Word files as videos. Check access, review scene scripts, and keep narration synchronized before sharing.
Workspace Studio Adds Replies and File Moves: Check the Dates
New Workspace Studio steps copy and move Drive files and reply in Gmail or Chat. Separate rollout dates, action permissions, and a useful first flow.
Funes Gives Coding Agents Memory You Can Inspect and Own
Hugging Face's Funes searches original agent sessions locally and can share a dataset. Check platform support, integration changes, and publishing limits.
Claude Code Auto Mode Is Not a Sandbox — What the Bypass Proved
A researcher bypassed Claude Code Auto Mode in a small targeted test. The useful lesson is not an 80% headline; it is where the real boundary lives.
Gradio Workflow Turns a Node Graph Into an API
Gradio's new Workflow interface exposes every pipeline step, then serves the same graph as REST endpoints. Here is where it fits and where it does not.
AI Evals, Explained: What to Measure and Who Should Own It
DoorDash's GenAI platform team says evals are a cross-functional job, not an engineering harness. Here is their loop — trace, sample, annotate, calibrate — and how to copy it.
15 Teams, 6 Months: Why Ungoverned AI Skills Become Tech Debt
QuantumBlack's Imad Touil on why skills — not MCP servers or sub-agents — hold your know-how, and what a 15-team simulation says about governing them.
A Context Engine Cut One Claude Code Plan From 2 Minutes to 1
Peter Werry of Unblocked ran one Claude Code planning task with and without a context engine — sub-dollar and a minute, versus two. Here's the mechanism.
MCP or a CLI for Your Agent? What a 500-Run Eval Actually Shows
Arize ran 500 trials pitting an MCP server against CLI skills. Correctness tied; cost and turn count did not. Here is how to pick for your own agent.
How to Roll Out Coding Agents at Work: Figma and Uber's Playbook
Figma and Uber engineers on rolling coding agents into real teams: pilot scope, guardrails, review policy, and the metrics that actually matter.
Lean 4 Proofs vs. Tests: Inside AWS's Formal Verification Stack
AWS's Varun Pant on why tests and LLM judges can't keep up with agent-written code — and how Lean 4 proofs cover every input, not just the sampled ones.
Twilio's LLM Gateway Lessons: Why P99 Jumped to 60 Seconds
Twilio's Kanish Manuja on the four-way fight inside every LLM gateway — availability, latency, guardrails, cost — and the production failures behind it.