Claude Code Agent View Explained: Run Multiple AI Sessions From One Terminal
Before Agent View, managing multiple Claude Code sessions looked like this: three terminal tabs, each running a separate session, each losing context when I switched away, no way to see which one was waiting for my input without clicking into each tab individually. Six projects across three repos meant constant window-switching and lost mental state.
Anthropic launched Agent View at the Code w/ Claude 2026 conference on May 6, alongside several other features that change how Claude Code works in practice. This post covers what shipped, how it works, and what's actually useful versus what's still rough.
What Agent View Does
Agent View is a single terminal interface that shows every Claude Code session in one list. Running, waiting for input, finished — all visible at a glance. You launch it with one command:
From this view, you can jump into any session that needs attention, peek at the latest output without entering the session, and start new sessions that run in the background. The interface updates in real time — when a background session finishes or hits a point where it needs your input, the status changes and you see it immediately.
For someone running multiple projects simultaneously, this is the difference between "I need to remember which terminal tab has which project" and "everything's in one list, sorted by what needs me right now."
Background Sessions: The Real Unlock
Agent View would be a nice UI improvement on its own. What makes it genuinely useful is background sessions — the ability to start a Claude Code task, send it to the background, and let it work while you do something else.
Before this, every Claude Code task required my active attention. I'd start a refactor, watch it run, approve file changes, wait for the next step. If I wanted to work on a different project, I had to either finish the current task or abandon it.
Now: start the refactor in the background, switch to a different project, come back when Agent View shows the refactor is either done or needs input. The workflow matches how I actually think about my projects — checking in on multiple things throughout the day rather than giving one thing uninterrupted focus for hours.
The /goal Command: Set a Target, Let Claude Work
The /goal command is the more autonomous cousin of background sessions. Instead of giving Claude Code a specific instruction, you describe what "done" looks like, and Claude keeps working until it gets there or gets stuck.
This is the feature that sounds most like science fiction and works most like a junior developer. It handles the iteration loop — try, fail, fix, retry — without needing you to approve each individual step. You define the success condition, and Claude Code keeps working toward it.
Two caveats from using it:
It burns through context fast. Each iteration adds to the conversation history. A goal that takes 15 iterations to achieve can consume a significant chunk of the context window. For complex goals, I've seen it hit compaction (where Claude summarizes earlier context to make room) within a single goal session.
The success condition needs to be testable. "Make the code better" isn't a goal Claude can evaluate. "All tests pass" is. "The API returns 200 for these three endpoints" is. The more concretely you define done, the better /goal performs. Vague goals produce vague attempts.
What Else Shipped at Code w/ Claude 2026
Agent View and /goal were the headline features. Three other announcements from the May 6 conference matter for daily use:
Rate limits doubled. Claude Code's rate limits were doubled across all plan tiers. For Pro and Max subscribers, this means more turns per hour before hitting the throttle. If you've been running into rate limits during intensive coding sessions, this is an immediate quality-of-life improvement.
Managed Agents and Dreaming. These are Claude Platform features (not Claude Code CLI features) aimed at teams deploying agents at scale. Dreaming is particularly interesting — a scheduled process that reviews past agent sessions, identifies patterns, and curates memories so agents improve between runs. It's a research preview, meaning it's available but not stable. For solo builders using Claude Code directly, these are worth watching but not immediately actionable.
Fast mode updated to Opus 4.7. Claude Code's fast mode now defaults to Opus 4.7 instead of 4.6. Faster output generation at the same price tier. The practical effect: Claude Code responds noticeably quicker during back-and-forth debugging sessions where response latency compounds.
How I Use Agent View Across Six Projects
My current project list: x402 Protocol API, SpeedTap Telegram game, x402watch analytics, Weather Bot, a flight compensation checker, and this blog. Before Agent View, each project lived in its own terminal tab, and switching between them meant losing track of where I left off.
Now the pattern is different. I start my day by opening Agent View. Overnight background sessions show their results — maybe a dependency update finished, maybe a test suite ran and flagged two failures. I address the failures first (they're marked "blocked"), then start new tasks for whatever I'm focusing on that day.
The mental shift: projects feel less like separate commitments and more like a dashboard of things moving at different speeds. Some are actively running. Some are waiting for a decision. Some are done and just need a review before I deploy.
For solo builders running multiple projects on free infrastructure, this is the workflow improvement that has the biggest practical impact. Not because any individual feature is revolutionary, but because the aggregate effect — seeing everything in one place, delegating tasks to the background, checking in instead of babysitting — saves an hour or more per day.
The Limits Right Now
Agent View is a research preview, and it shows in a few places.
Session caps still apply. Background sessions count against your rate limit the same as foreground sessions. Running five background tasks simultaneously burns through your allocation five times faster. Planning around this is necessary.
Context loss on long sessions. Background sessions that run for extended periods hit the same context compaction issues as regular sessions. Critical instructions from early in the session can get summarized away. For important tasks, shorter focused sessions still work better than long autonomous ones.
No mobile interface. Agent View runs in a terminal. There's no mobile app or web dashboard to check session status from your phone. When I'm away from my laptop, I can't see whether a background task finished or is waiting for input. This is the feature I most want to see added.
These are early-version limitations, not fundamental flaws. The core functionality — seeing all sessions, running background tasks, setting goals — works reliably enough that I use it daily. The rough edges will smooth out as the feature matures.
Related guides:
- Claude Code for Non-Developers: What It's Actually Like
- Building a Full-Stack App with Claude Code in One Day
- The $0/Month Tech Stack That Runs All My Projects
Disclaimer: This blog documents practical workflows based on personal experience. Nothing here is financial, legal, or professional advice.
Comments
Post a Comment