Claude Code Ultraplan Moves Terminal Work to the Cloud

Dive into Ultraplan, Anthropic's cloud-backed workflow that offloads heavy compute from your workstation to a dedicated web session. We explore how you trigger it from the CLI, the GitHub-only requirement, and why it runs on Anthropic's cloud. See the rich web review surface with architecture outlines, inline comments, and emoji reactions, plus how teleport returns a finalized plan to your local terminal. We'll discuss the implications for productivity, security, and the future of asynchronous, developer-friendly collaboration — and spotlight.

Note:  This podcast was AI-generated, and sometimes AI can make mistakes.  Please double-check any critical information.

Sponsored by Embersilk LLC

Transcript

SPEAKER_01 0:00

So uh the other day I'm just staring at this completely locked up terminal screen, right? Because my local AI is just it's chewing through this massive code base refactor.

SPEAKER_00 0:10

Oh, I know that feeling. It's like watching digital paint dry.

SPEAKER_01 0:13

Exactly.

SPEAKER_00 0:13

Yeah.

SPEAKER_01 0:14

And I'm sitting there unable to get any actual work done, thinking, like, why are we still letting local compute hold our thinking hostage?

SPEAKER_00 0:22

Right. It's a huge bottleneck.

SPEAKER_01 0:23

Yeah. So today we're diving into Enthropic's latest release notes to explore this really cool new Claude Co. preview feature, you know, for version 2.1.1.91 and up. It's called UltraPlan.

SPEAKER_00 0:34

Such a great feature.

SPEAKER_01 0:36

It really is. And our mission for this deep dive is to see how this completely frees up your machine so you can focus on building the future.

SPEAKER_00 0:42

Aaron Powell So that local hostage situation you mentioned, um, UltraPlan basically solves that by shifting all that heavy lifting from your machine over to a dedicated web session.

SPEAKER_01 0:52

Aaron Powell It's sort of like, well, think of it like hiring a remote sous chef, right?

SPEAKER_00 0:57

Oh, I like that. That's a great analogy.

SPEAKER_01 0:59

Aaron Powell Yeah, you give them this incredibly complex recipe, but instead of them taking up all the counter space in your own kitchen, they prep all the ingredients in a completely separate cloud kitchen. And you just keep cooking locally.

SPEAKER_00 1:10

Exactly. Mechanically, you just trigger it right from your command line. You can use like the slash ultra plan command or just type the keyword ultra plan in your prompt.

SPEAKER_01 1:19

Aaron Powell Wait, you can also just escalate a plan you already started locally.

SPEAKER_00 1:22

Yeah. Yeah. You can totally hand it off midstream. It securely streams your terminal context right to Anthropics Cloud.

SPEAKER_01 1:28

Okay, but offloading like an entire code-based context, that sounds like an infrastructure nightmare. They can't possibly support every enterprise cloud yet.

SPEAKER_00 1:37

Aaron Powell Well, no, they don't. That's actually a pretty crucial guardrail in the documentation. Right now, it strictly requires a GitHub repository.

SPEAKER_01 1:44

Aaron Powell Okay. Just GitHub.

SPEAKER_00 1:46

Right. And it absolutely must run on Anthropic's own cloud. So, you know, no Amazon Bedrock, no Google Cloud Vertex, or Microsoft Foundry.

SPEAKER_01 1:56

Aaron Powell makes sense. They probably need that super tight native back-end integration to stream the repo securely. And you know, if you're building custom AI tools and dealing with exactly those kinds of integration headaches, well, that's actually a perfect time to mention our sponsor, Embersilk.

SPEAKER_00 2:11

A very smooth transition.

SPEAKER_01 2:13

Thank you. If you need help with AI training, automation, integration, or software development, like really uncovering where agents can make the most impact for your business or personal life, you've got to check out Embersilk.com for all your AI needs.

SPEAKER_00 2:27

Finding the right integration is everything. And uh back in your terminal, while the cloud is doing all that work, you actually get these clever little CLI status indicators.

SPEAKER_01 2:37

Like a progress bar.

SPEAKER_00 2:38

Kind of. It's this diamond symbol that tells you if Claude is actively drafting, or um maybe it's paused because it needs a decision from you, or if it's finally ready for review.

SPEAKER_01 2:49

Aaron Powell See, that's the part where I was a bit skeptical at first. Offloading is great, but what if the AI's plan is just, you know, slightly off?

SPEAKER_00 2:55

Aaron Ross Powell Right. A giant wall of text is useless if you can't easily tweak it.

SPEAKER_01 2:59

Exactly. Like how do I change the recipe before it executes and ruins the dish.

SPEAKER_00 3:03

Well, that's the brilliant part. They don't just dump the output back into your terminal. You actually open this really rich web review surface.

SPEAKER_01 3:10

Oh wow. So it's a full UI.

SPEAKER_00 3:11

Yeah. It gives you this whole outline sidebar so you can navigate the architecture. You can leave inline comments on specific passages to force revisions.

SPEAKER_01 3:20

No way. Like a Google Doc.

SPEAKER_00 3:22

Pretty much. You can even drop emoji reactions to just like quickly signal approval or concern to the model. Aaron Powell Okay.

SPEAKER_01 3:29

I love the visual iteration. But eventually this code has to actually run on my machine. How does it get back?

SPEAKER_00 3:35

So you have options here. If the plan is completely self-contained, you can just approve it to execute entirely on the web. It can even open a GitHub pull request for you right there.

SPEAKER_01 3:45

Aaron Powell That is incredibly cool. But what if I want it local? There was something in the notes about teleporting.

SPEAKER_00 3:50

Yeah, teleport. Which, you know, sounds like marketing jargon, but sci-fi. You're right. But it's actually super straightforward. When you hit teleport, the web UI packages that finalize plan into a structured payload and pushes it back down the WebSocket connection.

SPEAKER_01 4:06

Oh, so it just injects it right back into your waiting local terminal session.

SPEAKER_00 4:10

Exactly. And from there you implement it locally or maybe start a fresh context window or just save it to a file.

SPEAKER_01 4:16

I mean, this is such a massive shift. We're finally moving away from like waiting on our tools to this state of totally frictionless, asynchronous collaboration.

SPEAKER_00 4:24

Aaron Powell It really is. It frees human developers to focus entirely on creative problem solving instead of, you know, just watching processing bars fill up.

SPEAKER_01 4:34

It's an incredibly optimistic vision for the future, like human creativity completely unchained.

SPEAKER_00 4:40

Truly.

SPEAKER_01 4:42

Well, if you enjoy this deep dive, please subscribe to the show. Hey, leave us a five star review if you can. It really does help get the word out. Thanks for tuning in.