Is Berd Worth Using? I Installed It Live and Here Is My Honest Verdict

Good morning, good afternoon, and good evening.

Berd dropped yesterday. It is from Block, it is open source, and it is supposed to work with Buzz. I had done zero prep. No test run, no notes, no idea what I was going to do with it. Listen, we were just there to experiment. I thought it would be a valuable stream to install it together and find out live.

Ninety minutes later I had a verdict on Berd that I did not expect, and a tool I did not plan to build at all.

Here is both.

The install, start to finish

Onboarding is short. What type of work will you be using Berd for? I picked marketing. It recommended three agents: planner, automator, generalist. I kept them. Then it asked me to choose a provider to power Berd, and I went with Codex.

Then you create a project, name it, and choose a folder. I pointed mine at a throwaway test folder, which I paid for later when every single thing I built that day landed in the wrong directory. Rookie move. Point it at a real directory.

Two things I liked immediately. It pulled in my universal skills cleanly, the ones I use across different apps, so my agents directory just showed up. And the website is genuinely funny. You can play with almost everything on the landing page. It cracked me up more than once.

There is also a little bird animation every time it responds. I said out loud that it was funny now but would probably get annoying with use. By the end of the stream it was annoying.

First impressions, about twenty minutes in

I poked at project chats, skills, agents, the canvas. Then I said the thing I actually thought:

It just seems like another chat app.

Cool visually. You put stuff on a canvas, you tag agents, you switch models. But I was having a hard time seeing the power in it. To me it felt like a chat version of Buzz with more visuals tied into it, and honestly I would rather just use Codex.

I went to their own marketing to see if it would change my mind. Your canvas, your rules. Build a crew, not a chatbot. Stop switching tools. Not married to one model. Animate a logo, build a silly 404 page, create an interactive presentation. That helped, actually. Visual first, creative first is how I would position Berd. That is who it is for.

It half convinced me. Not all the way.

The accident that became the real story

While I was poking around, I asked Berd to build me a prop studio. What I meant was a visual aid for videos I record. Everybody and their mother is using an Excalidraw whiteboard canvas right now, and I wanted a fresh approach to it.

The first pass was not what I meant. So I redirected it and gave it a real topic instead of a dummy one: is Grokbot really worth $200 to $300 a month, with five agents that replicate replacing a human employee.

Then it did the thing that turned the whole stream around.

It pulled from my front end design skill, rendered a unified infographic in the styling I have been dialing in for X and LinkedIn, and when I clicked a section, it zoomed in.

I am not mad at it. I am not mad at that at all.

That click to zoom is the feature. It saves me from doing camera zooms in post, moving from one point to the next manually in the edit. That alone is worth the build.

What Clear Mud Canvas turned into

From there it was iteration out loud for the rest of the stream. Every complaint became a prompt:

• The highlighter highlighted anything I clicked. It should only highlight text.
• Add a pen for underlining and annotation. The first pen strokes looked like overlapping circles instead of a fluid line, so that got fixed too.
• Kill the gray margin outside the canvas. The grid should be the canvas, with a little padding at the edges.
• Light mode with orange instead of green, because that is our color scheme.
• Rounded corners on screenshots. No rectangle border around them.
• Hotkeys, because I am talking while I present: M for pan, H for highlight, P for pen, C for clear, and hold space plus left click drag for temporary panning.
• A fit button and zoom controls, since full canvas mode would pan but not zoom out.
• An infinite drawing surface, because I found an invisible border where annotation got cut off.

Then the structural idea. Why tab out to a chat at all to make these changes? So we added a slide in chat panel on the right that spins up a local Codex session against the app itself. It is a REST style workaround, and it means I never alt tab to a terminal while I am working on a canvas.

Autoplay with a nice interval, a popup master prompt per section that I can drag around, a naming and archive system for past props. I pinned it, named it Clear Mud Canvas, and moved it out of test into ~/dev/clearmud-company/canvas where all our internal tools live.

I was not expecting to build what we built that day.

The pipeline, if you want to copy it

The flow that produced it is simple and reusable:

• Grokbot generates an Excalidraw canvas on the topic.
• That canvas gets fed to the internal web app.
• The app renders it in the Clear Mud creative system style.

Then, and this is the part most people skip, once an output finally looked right, I had it write a skill file from the approved version. Train on the final output, not on your instructions. I also offered to keep feeding it the hand tweaked files I actually ship, since I always make minor edits before recording.

Shout out Berd, shout out Codex, shout out Excalidraw. It started in a tool I am not adopting, and I am still giving it credit.

The question that hijacked the stream: is the $300 bot worth it

Chat derailed the demo repeatedly, and honestly the derails were the best part.

The one that kept coming back was whether Grokbot is worth $200 or $300 a month. My answer has not changed. It depends on the responsibilities you give it.

A social media manager who stays on top of trending news and preps posts in your brand voice is minimum $2,000 a month. Most are more. You can hand a lot of those responsibilities to a team of bots on a $200 or $300 plan and still have usage left over. From that angle, yes, it is worth it.

But here is the caveat I want people to actually hear: I would not justify spending money on it to fill a job role that you do not already pay for. If you do not currently pay a social media manager, you are not saving $2,000. You are adding $300 to your monthly expenses.

And the uncomfortable version, which I said on stream and will say here: if humans are doing the amount of work a $300 subscription can do for social media, chances are somebody is exploiting labor laws. I would rather not do that.

One more thing that came up. If you have ever considered Cursor Ultra at $200 a month, look at the $300 Grok plan first, because it lets you claim Cursor Ultra as well. You actually get $500 worth of plans. Do not quote me on the exact plan name or the nuances, I am still learning it, and I only found this out after I signed up for standalone Cursor Ultra by mistake and cancelled it immediately. Verify it yourself before you spend.

I am not here telling you to go spend money. I am spending the money so you do not have to.

Other things I learned live

Grokbot VMs. I said on stream that the VM is shared across all your bots. Then I tested it and reversed myself, twice. Each bot gets its own VM and its own open tabs. Credentials and cookies are shared across them. They get their own profiles, on the same VM.

Nothing is dead. I use Buzz, OpenClaw, Hermes, and Grokbot daily. Grokbot will take market share because it removes the setup friction, not because the others stopped working. It is just sad that what gets attention is saying something is dead.

I think subscription prices go down, not up. Everyone is saying enjoy the cheap prices while they last. I disagree. Grok is a top five frontier lab operating at a fraction of the cost, and the others have to compete on price. That is just my belief. I could be wrong. Nobody knows.

Privacy. If you use any frontier model, privacy is out the window regardless of which data sharing box you unchecked. If you actually care, buy the hardware and host your own model. With Grokbot the real trade is not data ownership, it is that you do not control the hardware or the infrastructure, and there is no self host option.

Fewer surfaces. Twice I lost track of where something was, and I do not even have that much set up. Then near the end I went to check on a prompt I had been waiting on all hour and found it saved as a draft. I never sent it. I squirreled hard. That was a rookie mistake, and it is the whole argument for narrowing what you have on screen.

The verdict on Berd

I gave it a fair shake for ninety minutes, and my minute twenty read survived.

I think Berd is cool. It has the wow factor with the 3D agents and the fun little features, the light mode default is great, the chat feels fluid, and tagging different agents in is nice. It also has persistent memory, which I checked live and confirmed.

But I do not know if I would use it for what it was built for. I wish there was a split option to preview things inside it instead of opening a browser, which is exactly the complaint that made me fork Buzz in the first place. And I am not going to take the time to fork this one.

I think I am just going to continue using Buzz personally.

That is a boring result for a stream I titled after the product. It is still the honest one. And the actual lesson is not the verdict at all: I accidentally built something I am now going to use for every single one of our videos moving forward. The value of testing a tool is not always the tool.

Pretty nifty. Pretty nifty.

Go kick some ass. We will catch you Friday, same time.

Watch the full live stream, squirreling and rookie mistakes included: https://www.youtube.com/watch?v=1eSPTkSy0ug