Everything Compounds Now

The Dashboard

I really thought that when I got the opportunity to blog consistently, it would be very focused. That I'd be drilling down on one particular thing per post. And I probably should force myself to do that. But so much is changing so fast that I've always got many different things to say. At least they're all somewhat related.

So let me just start with what I did today. Right as I walked out of the office, I launched a dashboard that gives me a graphical representation of what all my agents are doing. It's a security cam style overview. I really need to do a screencast of this one and include a video with the post because it's awesome. It's got that company aesthetic, the DOS green, Matrix-y, cyberpunk look. There's this neat little grid background with blinking status lights, and it gives me the overview of all my agents at a glance. I can click into any of them and drill down into what they're working on. It surfaces the standard error logs, any kind of output as they work, everything they're outputting to their respective terminals. I can read all of that through a GUI now.

Next up I'm going to set up a VPN so I can monitor all this stuff remotely. I usually use Pritunnel because it's easy, has a ton of functionality out of the box, and it's free as long as you're not scaling to enterprise. I've had developers in the past get all elitist about it, like "why don't you just use OpenVPN directly?" And I'd say, man, I don't have time for that. Pritunnel is easy and I can throw it up in no time. But now I can just do things. I'll probably roll my own VPN client powered by OpenVPN because that's all Pritunnel is, a wrapper for OpenVPN. And I can probably roll my own faster than I can install Pritunnel, which is wild.

Email to Pull Request

I've got a client that I've been building a SaaS product for, and we're in a revision phase now. The core functionality is there, and all we need to do is change button colors, eliminate excessive white space, formatting and layout stuff. There's some deeper feature work too, but we're clearly in revision mode for this milestone.

So I added a Slack channel. And when I say channel, I mean in the context of how OpenClaw has channels for Telegram and WhatsApp. I already had a Telegram channel, so this is a new adapter for Slack. What it does is watch a Slack channel where my client shares bug reports and revision requests. He prepares really neat documents with screen captures, marks them up with pen, exports them as PDFs, and shares them. I really appreciate that because the images highlight exactly what he wants changed.

The pipeline downloads all those attachments and sends them through an OpenAI interface because the GPT-4o model has the best bang for your buck on image processing. It really understands the images. It can look at a PDF printout with handwritten pen marks and parse the notes accurately. Anthropic has a model that performs about the same but costs a little more. Both are cheap, really. But I hooked up the OpenAI interface specifically to get the image understanding into the pipeline.

Now, I could have installed the Slack app directly on my client's workspace, but he owns the channel and I'd have to walk him through creating a token, adding all the right scopes. I just didn't feel like it. So I pivoted. Instead of a Slack message, the client creates an email, attaches all the files, sends it to me, and I forward it to Gus. As soon as Gus receives an email with a specific subject line, the pipeline fires. I reused the existing email skill since I already had a Google API client authenticated through the GOG CLI for Gmail, Docs, Calendar, all those Google products. The Slack adapter still exists and I'll come back to it because there's more functionality we can add there. But to get a viable product I could use today, I pivoted to email. Now it supports both.

The thing is, a lot of these bugs and revisions touch many different parts of the system. So it didn't make sense to do it all in one run without me monitoring it. I added an extra planning step that analyzes all the client's requests and bug reports, organizes them by which parts of the system they affect, and creates separate GitHub issues for each strategic chunk. Then it prioritizes them, executes them in order under a parent task, makes the changes, runs tests, creates new tests if needed, fixes anything it broke, and pushes everything to the same branch so it all lands in one pull request.

I just ran one end to end and it went great. So now, multiple times throughout the day when the client has revision requests or bug reports, I forward the email to my agent, he goes to work, and I get a pull request to review at the end of it. And I was doing all of this while building with K'NEX and Legos, making bread, folding clothes, doing laundry, spending time with my son.

Everything Stays Alive

What's crazy is how the stuff I'm about to list off feels lightweight now, like an "oh, by the way" kind of thing. These are things that would have taken me all week before.

We added generative engine optimization features to the SEO audit pipeline. I need to review it to understand it better, but if it works the way it looks, that's outstanding. I added a feature to my Upwork scan that analyzes all the different technologies across the job postings I'm targeting, so I can trend the tech they're working with, build demos around it, share content related to it. I said I was going to do that yesterday. Today it's done.

And everything I've been doing is still in place. That's the part that's different. We're actually compounding. In the past, I would sprint on a thing, it would be cool, I'd be proud of it, and then I'd move on to another project and that first one just dies. It goes dormant and sits there. Unless it was actual client work where the people I built it for went on to use it, personal projects would just fade. I might not even blog about it. I might not even share that I did it.

Now, while I was doing all this stuff today, I got Telegram messages that my SEO audit report is in and needs to be addressed, with a whole list of action items. I got a list of Upwork jobs with recommendations, strengths and weaknesses of each one, explanations of why they were recommended. I got the weekly report showing issues with my sitemap, the kind of thing you only want to run weekly because it's tied to Search Console. These things are all still running. They're all alive and my agents are making sure they stay that way.

Talking vs. Doing

I used to feel like a stoner talking about all the things I was gonna do, and then never executing on 90% of them. Now I'm saying a thing on Sunday and executing it on Monday. I know there are things I'm forgetting about that I did today. I can already foresee that I'm going to think of things, execute on them, and forget I even had the idea. And then I'll find out tomorrow in some report or some log monitoring alert that the thing I thought of has not only been documented, but it's been shipped.

It won't be long before I'm getting proactive recommendations for things to do. We planned a content analysis engine. And there's more stuff that I've already forgotten about. I'll find out tomorrow when I get the recommendation that I need to execute on it. And we're going to execute on it. It just gets better and better every day.