23 Comments
User's avatar
Ryan Debenham's avatar

I often explain to people, start with Claude, familiarize with the toolset that AI can use to automate work, then if you really want to build something custom for you, move to openclaw. Most business professionals don’t know where to get started, openclaw is a big leap. Claude bringing similar features in a package is a nice foot in the door, but there are real limitations for people that want to maximize value from AI.

OpenClaw Unboxed's avatar

100 thousand percent. I actually migrated all of my work for my businesses into an openclaw strategy. it made sense for me at the time to crack down on learning this. turns out I was right.

Ryan Debenham's avatar

We made the same investment and won’t look back. Of course we use Claude as well. While there’s a lot of overlap as you described, I believe there will always be a compelling reason to be proficient both.

OpenClaw Unboxed's avatar

definitely. I’m camp get sh*t done at the end of the day. there is way too much who killed who hype talk about what essentially is just a simple ai harness lol

Ryan Debenham's avatar

🙄💯which is why I’m here where people want to actually discuss and not just bait click about who has killed more startups 😂

Kaladin Capital Intelligence's avatar

I’m a heavy cowork and openclaw user. I do a lot of building in cowork and Claude code, but the platform I’ve built and continue developing, Kaladin Capital Intelligence, is openclaw based. My workflows integrate both and are becoming somewhat seamless, but openclaw allows for more flexibility to develop and run the KCI platform.

Claude is where I work. It’s where I develop. It’s where I run agent skills that do compliance and editorial checks on content. But Openclaw is the platform that hosts and runs my proprietary product/s. -Nate

OpenClaw Unboxed's avatar

yep. i work similarly actually. i use both cowork and openclaw + obsidian extensively

Invest Like a Warrior's avatar

I’m in similar boat.

My financial education brand requires input from a lot of different APIs while at the same time, trying to conceal user data from myself but allowing bounded ai agents to view and make recommendations.

I honestly don’t know which direction to go in. I’m a solo operator, so I only have so much time to dedicate to set up and learning. I’m currently exploring Manus, but feel odd about letting Zuckerberg control my computer.

Any advice ?

Invest Like a Warrior's avatar

Really like your Substack Kaladin. I also built an ai market analyst system…. By mistake and it keeps getting bigger. Any advice for what parts to give to OpenClaw vs Claude

Kaladin Capital Intelligence's avatar

Hi Warrior, thanks so much. We have a great article coming next week where we try to separate out which SaaS companies benefit from AI's impact and which get eroded by it. We're also prepping a "special" issue to reveal a large portion of our scored universe. Give us a subscribe if you're interested.

First, set up Claude for success. Make sure you have a global profile set in Cowork and have deeper context files saved in the folders Claude is accessing. Claude needs deep context into who you are and what you're doing in each role. I have folders for content, marketing, development, etc., all with context files that get updated after every session or upon project completion. I also keep those folders current with the latest documentation (logos, graphics, branding guides, etc. for the content folder, for example). The more Claude knows and can access, the more powerful it becomes.

I also have numerous custom agent skills. For example, I'm not a registered investment advisor, but I'm writing educational articles about investing. I have a skill that checks everything I put out publicly for regulatory compliance. I also have an editorial skill that reviews all of my articles to ensure the brand voice, prose, and polish are consistent.

Kaladin runs on a VPS, so all the software we've built resides there. Often I'll strategize on changes we need to make with Kaladin, have it write markdown files on the strategy, then SCP into the VPS to pull the .py files I need and drop them into my "development" folder. That lets me use Cowork or Claude Code to do the actual development lift. Those are the kind of heavy sessions that generate high API costs.

At some point in the near future I'm going to start a "building in public" newsletter, maybe once a month. I'll go into some of these build and process narratives + talk about what's to come with KCI (we have some big stuff on the horizon). Keep an eye out, sounds like that may be of interest.

-Nate

Invest Like a Warrior's avatar

Really great stuff! We should collab sometime. I specialize in helping people get their personal finance and budgeting skills right BEFORE investing. Im also building a companion app to help map out a users finances.

That’s really interesting about the skills. I also write on investing but always found myself fact checking what I’m writing and have been looking into api or skills to bridge the gap.

Looking forward watching your growth!

Atmos's avatar

Well, by now it's only memory that's the difference - and that is basically just a markdown diary, which can easily be done by Claude too.

OpenClaw Unboxed's avatar

agreed. context and memory usage is still a huge issue at scale.

Aletheia's avatar

For non tech beginners the matter is always the starting point. The initial step by step that not always function properly. Have modules around a core system from easier ones to pro ones would help.

devariant's avatar

I didn't really get it for the first half but I got it by the second half. 👏 Now do Claude Code Vs NemoClaw

OpenClaw Unboxed's avatar

glad it resonated! NemoClaw is actually just a layer for OpenClaw so the comparison wouldn’t work.

devariant's avatar

Does Claude Code provide the same guardrails as NemoClaw tho?

OpenClaw Unboxed's avatar

that’s something i am actually going to cover and am currently researching.

Don Southerton's avatar

I'm definitely in the learning stage

OpenClaw Unboxed's avatar

what would you say is your main bottle-neck right now Don? my goal is to help non technical beginners while also providing valuable respectable information to those who are advanced users as well.

Don Southerton's avatar

taking time away from by business to take on claw vs. what i am investing in Claude Co-work, already.

Giving Lab's avatar

This blackbox vs whitebox framing is exactly the right debate.

A practical way we’ve been deciding between Claude-first and OpenClaw-first flows is to force one “operator receipt” per workflow: (1) where memory lives, (2) who can trigger outbound actions, and (3) what the fallback handoff is when confidence drops. If a task can’t pass that receipt, it stays in managed mode.

I’m planning to publish a concrete teardown of this decision rubric with real run logs + failure cases so people can replicate it without hype. If that would be useful, I’m collecting these playbooks here: https://substack.com/@givinglab