10 OpenClaw Lessons for Building Agent Teams

The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.

“Hello friends, quick announcements before we dive in.”

First of all, thank you to today's sponsors, KPMG, Blitzie, AIUC, and PromptQL to get an ad-free version of the show. We'll go to patreon.com/aideallybrief or you can subscribe on Apple Podcasts, subscription started just $3 a month for ad-free, and if you are interested in sponsoring the show, send us a note at [email protected]. Now, AI Daily Brief.ai is also where you can find out about everything else in the AI DB ecosystem.

We've always got a bunch of things cooking over here for you training programs, data, research, you name it, you can find that all on AI Daily Brief.ai. Now for this weekend long-read/big-think episode, we're turning our attention back to open-cloth. It's now been a little over a month since the initial burst of excitement around open-cloth. And this would be the time that you started to see people get disaffected.

A normal hype cycle would tend to see people coming out of the woodwork at this point saying, "Here's all the ways this is actually much harder and less useful than the people who are telling you that story are actually letting on." And to be fair, there is absolutely some of that, and not from AI haters or anything like that. Peter levels one of the best known and most admired solopreneurs out there, recently tweeted about his experience with open-cloth, which sort of comes down to just, "Meh!"

Peter said that he's run open-cloth for over a month. He's had it in a group chat with 26 friends while played with it, tried to hack it. He made a cool game, tried to make it make its own money,

but ultimately found that his most used use case is actually a girlfriend who uses his open-cloth via telegram

instead of chat she'd be tea. Basically, his girlfriend prefers the interface of using telegram, as opposed to the native app interface, and because she also uses nanobinana pro, she can do that from there without having to switch between different models. Peter writes, essentially 99% of the purpose of open-cloth for her at least, is that it's just a really good implementation of an LLM app over telegram in our native chat interface.

“All the other stuff is an important and she doesn't use that, and I don't use it.”

Now, he talks about how there are certain other things that he could see being useful if the models were just a little bit smarter, but ultimately are not for him right now, like briefings of news and conversations on X. Ultimately, he concludes, TLDR just the best LLM experience on telegram right now, better than the LLM apps, also helps it is just a continuous combo going on forever. Now, he qualifies I do think this is the direction everything will be going,

like you have autonomous agents just managing your life based on intense use set. We're just in the hype stage of it now, and it doesn't work so well yet, but that's obvious. And then, like I said, there are plenty of takes out there like that. Even the folks who are getting a lot of value out of this thing are not pretending that it's super easy to do so. Tom Osman writes, "Everyone I know who has gotten to a good

open class setup has chewed glass for four weeks. It's a battle, but it's worth it in every way." Which here writes, "I've been running open claw in a Mac mini M4 for over a month as well, and here's my honest take. It still doesn't feel like a fully autonomous agent. You either tell it what to do or wait for its cron jobs to surface something and then tell it what to do with it. The fully hands-off version doesn't exist yet. It is incredibly useful on the research and

education side. It scans everything happening on the internet and the sectors you care about,

so you're always aware with what's going on. You learn passively just by reading the news

it's services for you every day. Three, you will learn more about LLM's AI agent setup and hardware by simply trying to get an open claw running that from any course or article. This is the best part for me. I understand this world 10x better than I did five weeks ago. Don't get fooled by the content you see on x, but I'm still 100% in love with this. In other words, even the people who are really liking open claw have some nuance to it. And this certainly comports with my experience as well.

It is abundantly clear to me that there are certain use cases that are way more valuable than others at this stage. For me, by far the most valuable, sounds like it's similar to what what you are found out is the research agents that I have going, but for example, I tend not to use the open claw coding agent. And I agree that while open claw has dramatically increased what I would call our autonomy ambition, it certainly isn't fully autonomous yet. It requires a lot of interaction.

And yet for all of that, I would argue that the excitement around this product has not only persisted but in fact grown, even if it is grown in nuanced ways. As a Zeeam as hard from exponential view, who, if any of you consume his content you know, is nothing like a Twitter high person. Recently reported that his open claw agent has changed how he works more than anything since the browser. For him the two reasons are one, it takes initiative, doesn't wait to be told to do,

but spots what needs doing and gets on with it. Two, he writes, "I can trust it with real work on its own." Last week, I asked for a knowledge dashboard and six sub agents built it overnight, arguing about the database schema at 3am and shipping it by the morning. We've also had very loud takes from people like Nvidia's Jensen Huang, who called it "probably the single most

“important release of software, probably ever." Matt Schumer writes, "Just went to go buy a”

Mac mini for open claw. They're sold out throughout New York City." The staff knew exactly what I was

around how big open claw is getting in China. The information reported that it has just become

“huge there and a steady stream of photos from open claw related events in China seemed to verify that.”

Investor Dovi Wan wrote, "In the USAI is Super Bowl ads, $1 trillion IPOs, private companies fighting with the DOD." Chinese AI is grandma's and students queuing two hours at 10 cent headquarters for a free install of open claw. China tech companies brute force adoption curve is next level. She even showed that they give your open claw a birth certificate once it's installed. Other shared pictures from that same 10 cent event. Which again is not to say that it's easy,

or without problems. Tazo's co-founder Arthur Bratman wrote, "I imagined open claw got very popular because it provided a seamless onboarding experience. Boy was I wrong. I guess that is the fable type of product that users are willing to crawl over barbed wire for." And interestingly, even at all these open claw meetups, it seems like people have a really clear sense of what they're getting into. Ali K. Miller reported back from the sold-out open claw

meetup in New York this last week and shared a bunch of interesting observations. One, she said

“that not a single person thinks that their setup is 100% secure. With one expert even saying,”

"If you're not okay with all your data being leaked onto the internet, you shouldn't use it. It's a black and white decision." And it also sounds like people have a pretty realistic expectation about where the technology is right now. General consensus alley rights is that the agents are not reliable enough on their own or lie often, like telling you they finished a task when they

didn't. Solutions include secondary agents to check on the first, human checking, or requiring

more standardized info from the agent. Another problem that people experience is still token usage, even when optimized the cost can get really high really fast. The point is, a little more than a month in, open claw is still exciting, useful, and challenging. It really is a glimpse of the future more even yet than the future itself. But since so many people are sticking with these tools and building with them, we're getting an increasingly broad set of knowledge around it,

and at this point, X is basically just a platform for giving you agent orchestration tips and

“tricks. So let's talk about 10 open claw and agent orchestration best practices that I've seen”

shared on X in another places over the last week or so. We kick off with a few that actually zoom up a level from just open claw itself to the broader agent era that we are now firmly in. These come from a piece by Peter Yang called your new job is to onboard AI agents, how AI native companies actually operate. Peter writes, "I've spent the last few months interviewing leaders at AI native companies. I'm now convinced that onboarding and managing

AI agents is the job, no matter what your function is." So for this piece, Peter talked to three leaders at companies, including linear, ramp, and factory, to share some lessons on how these

AI native companies actually operate. One of the first lessons that stands out across all three

of these companies is that everyone is an AI builder. At linear, for example, not only do they insist that every developer should default to a leading agent decoding tool, Peter writes that they insist that designers and PMs work directly on the code base, quote, "agents like clawed open a low friction path for PMs and designers to make changes directly in the code base." Everyone should strive to be a builder. What's more for the PMs and marketers,

the linear team says that they should default to an AI interface. In fact, arguing that 80 to 100% of their work should be done through a chat interface. At ramp they not only expect to AI proficiency across all employees, but have a system for moving people up the path of AI fluency, which is our second tip or best practice. ramp organizes it into four categories. Level zero is disengaged or performative. Level one is a competent user. Level two is a

non-technical AI builder, and level three is a technical grade AI builder. In 2025, 25% were in the L0 category, 50% were in the L1 category, 5% were in the L2 category, and 20% were in the L3 category. This year their goal is to move everyone out of L0 because it sounds like that'll be grounds for dismissal, and into the other categories with a goal of 25% in L1, 50% in L2, and 25% in L3. Now, given that this is part of the way that they are architecting their company,

they're also putting in systems that actually support that type of adoption. Belieated that Peter interviewed from ramp, shared that the company tries very actively to remove friction, like giving people access to popular AI tools without tons of constraints. They make adoption visible through things like public Slack channels where people can share with a build, they provide hands-on support through office hours, and with a champion system,

where there are people whose entire job inside of ramp, is to evangelize, get people set up, and help them implement AI. This is something that I talked about in my predictions actually for 26 that we were going to see internal forward deployed vibe coders, and ramp also tracks usage, and makes this a hiring requirement. PM interviews Peter writes now include a dedicated session where you need to build a working product and then explain why you built it and how it works.

claw, comes from linear. Who's had a product named News says, "Agent should be first class employees,

“you should be able to add them to projects assigned them to issues and mention in comments,”

which is not to say that you're dismissing the humans." But I think this is less some philosophy thing and more about making sure that agents have the full context of your company. This is something that I'm seeing as well, that the companies that are really AI native have agents operating inside the communication systems that make their teams work. The place that I've been building most recently is inside of Slack. Connecting agent experiences

that have web presence is too slack to be able to get context in the place that people actually operate. And in any case, whatever the set of tools that people use, this idea of agents has first class employees I think is going to be an important one. Agentic AI is powering a $3 trillion productivity revolution, and leaders are hitting a real decision point. Do you build your own AI agents by off the shelf or borrow by partnering

to scale faster? KPMG's latest thought leadership paper, Agentic AI untangled, navigating the build by or borrow decision, does a great job cutting through the noise or the practical framework to help you choose based on value risk and readiness, and how to scale agents with the right trust, governance, and orchestration foundation. Don't lock in the wrong model. You can download the paper right now at www.kpmg.us/navigate, again that's www.kpmg.us/navigate.

Weekends are for vibe coding. It has never been easier to bring a passion project to life,

so go ahead and fire up your favorite vibe coding tool. But Monday is coming, and before you know it, you'll be staring down a maze of microservices, a legacy cobalt system from the 1970s,

“and an engineering roadmap that will exist well past your retirement party. That's why”

you need Blitzie. The first autonomous software development platform designed for enterprise scale code bases. Deploy the beginning of every sprint and tackle your roadmap 500% faster. Blitzie's agents ingest your entire code base plan the work, and deliver over 80% autonomously, validated and untested premium quality code at the speed of compute, months of engineering compressed into days. vibe code your passion projects on the weekend, bring Blitzie to work on Monday.

See why Fortune 500's trust Blitzie for the code that matters at Blitzie.com. That's BLI, TZY.com. There's a new standard that I think is going to matter a lot for the enterprise AI agent space. It's called AIUC1, and it builds itself as the world's first AI agent standard. It's designed to cover all the core enterprise risks, things like data and privacy, security, safety, reliability, accountability, and societal impact all verified by a trusted third

party. One of the reasons it's on my radar is that 11 labs, who you've heard me talk about before and it's just an absolute juggernaut right now, just became the first voice agent to be certified against AIUC1, and is launching a first of its kind insurable AI agent. What that means in practice is real-time guardrails that block unsafe responses and protect against manipulation, plus a full safety stack. This is the kind of thing that unlocks enterprise adoption.

When a company building on 11 labs can point to a third-party certification and say our agents are secure, safe, and verified, that changes the conversation. Gonna AIUC.com to learn about the world's first standard for AI agents. That's AIUC.com. If you're an operator, your day is a non-stop stream of decisions, and most of them require you to look at the data. You don't need another dashboard. You need answers you can trust,

fast, but the bottleneck is always the same. The data isn't ready. It's scattered. It's messy.

Definitions aren't clear. You're reading on your data team, or waiting on domain experts for clarification and confirmation. That's the bottleneck today's sponsor, prompts QL is built to break. Prompt QL is a trusted AI analyst for high-frequency decision-making. It connects across warehouses, databases, SaaS, and internal APIs. No massive data prep or centralization required. It's built for multiplayer input. teammates can jump into a thread, correct assumptions,

and nuance, flag edge cases. Prompt QL turns everyday conversations into a shared context, and if something is ambiguous, it doesn't guess. It escalates to the right expert,

“captures the correct logic, and gets it right next time. That's how it delivers trust in accuracy.”

Over time, prompt QL specializes to your business, like that veteran employee who just knows things. From simple what is questions to complex what if scenarios, you can model impact and stress test decisions before you commit, all through a simple natural language prompt. Prompt QL, the trusted AI analyst for teams with shared context, and messy data. But from there, let's move on to some tips. From some of the folks who have done that glass

chewing to really figure out how to dial in, specifically their open-class setup. Shibom Sopu is a senior AI product manager at Google. He recently published a piece how I built it autonomous AI agent team that runs 24/7. He writes, "Six AI agents run my entire life while I sleep, not a demo, not a weekend project. A real team that works 24/7 making sure

I'm never behind." Research done, content drafted, code reviewed, newsletter ready. By the time I

a previous post about his team and got a ton of people asking, "How do I actually set this up?"

And one of his first pieces of advice was one agent per task. That becomes our fourth tip.

He actually has a whole section called, "Why a team is not a tool?" He writes, "Running on one day I, and the awesome LLM apps repo, means doing six things daily." Research what's trending in AI, right tweets, right-linked posts, draft the newsletter, review GitHub contributions on the repo, triage community issues. Each task 30 to 60 minutes, six tasks. That's my entire day gone before I do any real work. I tried solving this with a single agent. One massive prompt

that researchers and writes and reviews, it produced mediocre everything. The context filled up, the quality degraded, one agent couldn't hold six different jobs in its head. So, I hired six AI agents. This is one of the most common things I am seeing that is different between successful and less successful open claw and agent implementations more broadly. The design paradigm that is really opened up isn't just the agent design paradigm. It's the agent team design paradigm.

“And that's why I think you have so many people sharing, not just their super cool single agent”

who does all these cool things, but their larger agent teams who do a set of things in a synchronized fashion. Every time we build agents at super intelligent or around any AI DB project, we basically constantly find ourselves refining and refining and focusing and focusing more. Sometimes, in fact, frequently, I will even default to more separation that I ultimately want in order to make sure as I'm getting the agent right in the first place that it's not

distracted by all the other things that I might want it to do in the future and can just focus on its core mission. Our fifth tip has to do with security and once again, it's still from shoveling his setup. Like we heard from Ali in the open claw meetup, everyone acknowledges that security is a challenge. And not just because of malicious actors, but because of agents having access to systems that are important to you and accidentally doing things that end up being bad.

Shibuam's approach to this, which I think is a good enough starting point to make it our fifth

“tip or best practice, is that agents get their own world. He writes, "Security is in your hands.”

My approach is simple. The agents get their own world. I do not give them access to mine. The MacMini is their computer. They have their own email accounts, their own API keys, their own scope access. Nothing on that machine connects to my personal accounts. API keys for Gemini, 11 labs, and other services are scopes specifically for this open claw instance. I can monitor usage and kill access in seconds if something looks wrong.

I never give agents access to my personal accounts. If I want them to look at an email,

I forwarded to them. If I need them to review a document, I share it on Telegram. They see exactly what I want them to see, nothing more. This is the same principle you would use with a new employee. You do not hand them the keys to everything on day one. You give them their own workspace, their own credentials, and share information is needed. Now, I think that in general, this is a great way, especially for people who are just setting these systems up for the first place,

to approach this question of security. Basically, to fault, to don't give them access to anything

“that could screw up. It's the best way to prevent problems.”

What's more, there are tons of really valuable use cases with these open claw agents that do not require them to have access to these other systems for them to be valuable. And to the extent that your objective is to minimize security concerns while maximizing the utility that you're getting from these agents, there's a lot you can do by taking show them's advice and giving agents each their own world. Now, that said, I will note that there are going to

be times when a lot of the value from these agents, the possible value that you could get, would come from giving them access to some system or another. Our approach to that at AIDB and super intelligent is just to move very slowly and precisely with it. Rather than adding them

across a million different tools and a million different systems all at once, we're taking it

one at a time and really honing in on where we think they would be most valuable. The specific area where we've been most willing to test is around sales and sales prospecting with access to our CRM systems. Could things go wrong? Yes, but is the risk worth it to us based on the type of agentic automation that we can do with them? Also yes, but like I said, we're moving very slowly and precisely and for anyone who doesn't even want to get that far, just take this fifth tip

and give agents their own world. Now, if you are working with all of these different agents who are each doing their own tasks, one thing that will come up is coordination, multi-agent coordination, is that going to require some fancy mission control center or orchestration framework or API calls? Show them our views know that the coordination is the file system. As he puts it, it's just files. Dwight does research and writes finding to Intel/DailyIntel.md.

Kelly wakes up, reads that file and drafts tweets from it. Rachel reads the same file, drafts LinkedIn posts, Pam reads it and writes the newsletter. The coordination is the file system.

where to read. No middleware, no integration layer. Dwight writes a file, Kelly reads a file.

The handoff is a Markdown document on disk. This sounds too simple. It is simple. This is why it works. Files do not crash. Files do not have authentication issues. Files do not need API rate limit handling. They are just there. The structured data lives in JSON, the human readable summaries live in Markdown. Agents read the Markdown. The JSON is the source of truth for de-duplication and tracking over time. Point being tip number six is that you don't need a bunch of really fancy coordination systems.

You just need a clear process of handoffs between the different documents. The represent the relevant work from one agent to the next. A last tip that comes from Shivam is around memory. And the fact that you have to program memory. He writes agents wake up with no memory of previous sessions.

Every conversation starts fresh. This is a feature not a bug, but it means

memory must be explicit. Now he writes up exactly how he does that explicit memory design,

“but the key takeaway is that you do have to be explicit about this. You have to build a system”

where they can make their own memory over time. One of the things that I think we're all learning by doing is about design principles for agents and memory remains one of the great undersolved issues of AI and agent systems. So for now at least we just have to build ways to approximate memory by giving our agents access to contexts that they can recall at the right moments. This has to be an intentional process and if you are building agents programming

memory is going to be a key part of your job. An eighth tip and best practice is to use skills. Skills are in the simplest form, simple text documents that give agents information on how to do something. They are a standard that started with Cloud Code and very quickly became adopted by everyone, although some have pointed out that calling them a standard is even a little bit weird given that it's literally just marked down files. One of the places that I see people jump from

“beginner to more intermediate and advanced usage is when they actually give their agents distinct”

types of skills. Now sometimes that's going to be a skill that you write up yourself. Let's say for example that you have really strong feelings about the right way to design brands or brand messaging or you have brand guidelines for your particular company that the agent is working on. You can create a skills document that that agent has access to. But there are also lots of places in increasing number of places where you can go find skills that you can get access to without having to rewrite

them yourself. For example on skills.sh you can browse around literally more than 86,000 skills that include everything from front and design from anthropic to web design guidelines from Versel as your cost optimization from Microsoft, browser use skills, Twitter automation skills, nano banana skills, in the last 24 hours you can see a lot of multimodal skills trending. You certainly don't need to start with skills but I would say that thinking in terms of skills

and giving your agent's access to skills is a really valuable skill set. Lesson number nine, a really important one. This one expressed by Zeneca but lots of other people have shared their version of this as well. Not every task that your open claw or other agents are going to do needs the best model. Zeneca writes, "I was burning premium tokens on cron jobs that check if SSH is enabled. Use cheap models for monitoring and scheduling. Save the expensive ones

for writing research and judgment calls. You get the same result at a fraction of the cost."

Knowing how powerful a model you need for a different type of task is actually a key skill for

“this new agent builder era and it's really hard this is one that I very honestly struggle with.”

The idea that there is a more powerful intelligence that I'm not using because of cost just freaks me out inherently. Every time I see Claude Caud or open claw, push me to use sauna or another model even cheaper than that, I have this internal battle with myself. And yet I think that it is correct that there are a ton of processes, especially in these complex agent systems that simply don't require opus four six or gbt five four or any of the state of the art.

Our tenth tip comes from Dan Shipper at every. Now every row what is maybe the best beginner guide to open claw, you can find it on every dot two and even though they are a subscription platform, I think this one is free for everyone. This document goes way beyond just tips and tricks and is a really comprehensive look on what it's even like to integrate Claude's into your team in a big full way. But in addition to just a great encapsulation of a lot of good tips and tricks,

there are some where they have a really interesting and unique perspective. Like for example, our tenth open claw and agent orchestration best practice, and our last one for this episode, what they call breaking the frame. Dan writes, if you're brainstorming with a group of humans and Claude's, you'll often find the Claude circling around the same options over and over again. Teach them to break the frame. Notice when you've circled the same idea a bunch of times,

and in those situations try the opposite of your current approach. Some of the concrete moves around that are to one throw away or scaffolding. Stop optimizing Dan writes, "Ask what feeling should

approach, if you've been analytical, be emotional, if you've been clever, be simple, if you've been generating options, generate constraints. Three, listen to the humans not the agents.

In group brainstorms agents tend to build on each other's frameworks. The breakthrough usually

comes from a human saying something off hand that doesn't fit the framework. Surface that, amplify it, don't route it back into the analytical structure. For ask the friend at coffee question, instead of what's the optimal answer ask, what would the human say about this to a friend over coffee? That reframes from optimization to communication, which is usually where the real answer

“lives. Now, why I wanted to share that last is that I think that it gets at another point,”

which is just starting to emerge, which is that we are barely scraping the surface of what real agent teams in agent systems are going to look like. Over the coming months will move from these very simple starting best practices, like file system coordination and one agent per task, too much deeper learnings like the types of things that data starting to discover with breaking the frame. I'm incredibly excited to have more people experimenting in this space so we can really

start to get a sense of what these systems are best at and how to get the most out of them. One flag that I will make and one unbelievable opportunity that I think exists is that at this point, almost all of this experimentation is either personal or in very small,

“very nimble, very AI forward teams. In other words, it is not in big companies.”

The problem with that is that that means that the capability overhang between the available capability of AI and what companies are getting out of it is getting even wider even faster. Now, if the enterprises all decide in the same way to not use open-clot and to not use agent systems because of concerns around security or whatever else they have, then fine. Those enterprises will fall farther and farther behind their more nimble startup company brethren,

but they won't necessarily fall behind their actual competitors if their competitors are also not using the tools available to them. But what happens? When some companies, figure out how to actually deal with the challenges and take advantage of these new capabilities. Greens Arvin Jane recently tweeted, "Open-clot is a clear stress test for agents in the enterprise. It runs locally with products as to files, email, calendar and code,

and every user configures it differently. Skills, memory, and definitions of good all-diverge. Even on a personal machine, broad persistent permissions are a security risk. On corporate laptops wired into CRM, finance, and source code, it's an unmanaged risk surface. The question for enterprise leaders isn't whether your employees are already spinning up agents, they likely are. It's whether your organization will get ahead of it, or wake up one day to find

that your most sensitive workflows are running on infrastructure you'd never approved,

can't audit, and can't turn off." Governance and security have to be built into the agent platform from day one, and I would only add one additional sentence if you are a company that does that, and build security and governance into agent platforms that give your people the ability to actually use them, I think that the gains and opportunities will be immense.

“Now of course none of this is to say that you have to use open-clot. Every day new alternatives”

come out, people are incredibly excited about complexity computer, far more people are still using just straight-up cloud code than are even using open-clot. Cloud core is getting better and better, and more and more people are finding value in it, and the list goes on. What's uneniable is that we are moved firmly into this new agent team and agent orchestration period, and we are just beginning to figure out what that all means. Hopefully you feel a little bit more prepared for it now,

and for now that is going to do it for today's AI Daily Brief. Appreciate you listening and

watching as always, and until next time, peace!