The AI Catchup -- April 15, 2026

Welcome back. We are calling this the orchestration week. Anthropic dominated the headlines with three releases that all point in the same direction -- AI coding work is moving from "ask and wait" to "schedule, monitor, and orchestrate." Warp, Cursor, and Ramp shipped meaningful complements to that story, and a head-to-head comparison of the new wave of scheduled agents is our piece of the week.

Let us get into it.

This Week in AI

The big story is Anthropic's orchestration cluster. Three drops landed in five days, and they fit together in a way that is hard to ignore.

First, Claude Code Routines shipped, letting you schedule agents on cron, trigger them via API, or wire them to GitHub events. Routines are the missing piece for unattended work -- nightly dependency upgrades, scheduled triage, on-demand release notes. The agents run, post results to Slack or GitHub, and you read the diff in the morning.

Second, the Claude Code desktop app got a full visual rebuild. The new interface treats long-running agents as first-class citizens, with a session tree, live output panes, and a sidebar that surfaces every running Routine. If you have been using Claude Code primarily in the terminal, the new desktop is worth a fresh look.

Third, Anthropic's Thariq Shihipar published a practical guide to managing the 1M context window. It is the most useful writeup we have seen on context economy -- when to compact, when to fork, what actually fits in a million tokens, and where the real bottlenecks are. Required reading if you are running Claude Code on large codebases.

Tool Spotlight

Warp shipped what they are calling universal agent support, which positions the terminal as a full agentic development environment. You can now run Claude Code, Codex, OpenCode, and Gemini CLI inside the same Warp session, with vertical tabs for parallel agents and a unified command surface across all of them. It is the cleanest implementation of the "terminal as ADE" idea we have used so far, and it makes switching between agents on the same task feel native rather than clumsy.

Practice of the Week

Ramp's internal AI workspace, Glass, is reportedly hitting 99 percent weekly active usage across the company. We dug into the playbook in Internal AI Workspaces: The Ramp Glass Playbook for Company-Wide AI Adoption. The piece pulls out the patterns that any company can copy -- how Ramp picks problems for AI, how they handle data access, how they keep adoption sticky. Glass itself is internal only, so this is not a tool you can install. It is a blueprint.

Tutorial of the Week

Eric Zakariasson published a small but clever workflow this week -- a prompt that mines your existing Cursor chat history to auto-generate rules and skills. We wrote it up as a step-by-step in Auto-Generate Cursor Rules and Skills From Your Chat History. If you have been using Cursor for months without ever writing a rule, this is the lowest-friction way to bootstrap a useful set from work you already did.

What We Are Watching

Cursor 3.1 introduced interactive canvases -- the agent can now render a small interactive UI inside the chat instead of returning a wall of text. You ask for a comparison, you get a sortable table. You ask for a flow diagram, you get a draggable one. We are still figuring out when this is genuinely better than text and when it is novelty, but the early use cases for spec exploration and config tuning are promising.

Comparison of the Week

The scheduled-agent space went from one player to five in the span of a quarter. We put them in a head-to-head: Scheduled AI Coding Agents in 2026: Claude Code Routines vs Cursor Automations vs Codex vs Warp vs Gemini CLI. The short version -- Claude Code Routines wins on triggers and ecosystem depth, Cursor Automations win on UI polish, Codex is strongest for pure background coding, Warp wins for terminal-native workflows, and Gemini CLI is the budget choice. The full table breaks down pricing, trigger types, and observability for each.

Quick Hits

Chrome Skills shipped on April 14. Google rolled out Skills in Chrome, letting you save reusable Gemini-in-Chrome prompts and trigger them from the address bar. It ships on Mac, Windows, and ChromeOS in English-US, and saved Skills sync across signed-in devices. A lightweight take on the same pattern Cursor and Claude Code have used for a year, but the distribution is going to be massive.

That is it for this week. The AI Catchup publishes every Tuesday. If you found this useful, subscribe to get it in your inbox.

Until next week -- stay caught up.