Ok3to

I recently had the delight of discovering the comment section of my favorite podcast. For additional amusement, I promised myself I’d check in on the comments section more often. It did not disappoint.

Before your outreach bot emails strangers, send it to a friend who'll be honest. Strangers just don't reply, and you learn nothing. A friend tells you if it reads like a template. Mine said it should get attention. That's when I felt ready to send it for real.

I'm running an outreach experiment on LinkedIn. My @bot agent sends the invites. When someone accepts, it drafts a reply and stops. That's my cue. That's how I build confidence in it. Each time it earns trust, I give it a little more responsibility further up the chain.

Grok Bot just got a new trick. It can build reports, memos and dashboards for you, then keep them up to date. So I tested it on my own work. I asked @bot for a map of where our US lead gen stands, with one colour per state. What I got 16 states in outreach 20 states with lists 14 states still grey What I'd ask yours "Make a dashboard of my work. Show what's moving and what's waiting on me. Keep it up to date."

My agent is confident. It clicks and moves on. The problem is what it's clicking on. Pages lag and things fail quietly. So now it checks that something actually changed after every click. I trust my agent. I don't trust the internet. But now we get results we can be confident in.

Monday morning the bots were up before me. Research finished. Sixty nine emails written and checked. Ready to go. Then the mail app said my session had expired. I have one hard rule: no bot types my password. So the whole morning sat there, fully prepared, waiting on the one team member who hadn't had coffee yet. The bottleneck in my automated morning is me remembering to sign in. I'm starting to treat my own login like a scheduled job.

While I slept, my bot went hunting through free Marketplace listings for anything I could flip. This morning it handed me a washer across town and a fridge up a U shaped staircase. Neither was worth it, so I said no to both. And somehow that was still the best part of my morning. The grunt work was done before I woke up. All I had to do was read the options and decide. Every day it's a little lottery of ways to make money, and I never know what's coming. Even saying no feels like a win.

AI used to wow me. Now that my bots actually work, their logs are dull. Checked. Matched. Sent. Next. I only feel something when it breaks. So now I'm chasing boring. Boring means it worked.

Bot one: "That email went out Sunday." Bot two: "Still holding it for review." Both confident. Both wrong-ish. They were checking different records. So now there's one place that says what's done, and every bot checks it first. The part I didn't expect is that every hiccup teaches the bots how I work, and teaches me how to set them up so it can't happen twice. They're getting better at the job. I'm getting better at working with them. We're levelling up together, and that's the compounding nobody talks about.

I'm building a little local guide for Brunswick and writing to each business that's in it. My first instinct was "Congrats, you're listed!" Then I checked the first restaurant on Sydney Road. Their own website has a different phone number than the one we had. So the email now opens with a question. Is this right? Reply YES, or tell us what to fix. Turns out "we might be wrong" is a much better first line than "you're featured". People like fixing things more than being sold things.

I thought spreading the work across more Grok bots would save the day. It didn’t. They still fell over around the same hours. Different agents. Same computer-use cliff. Turns out I’d been dividing tasks, not dividing time. So I started treating the clock like part of the design. Overnight jobs. Quieter windows. Using the full 24 hours instead of stacking everyone into the same rush. The bots werent the bottleneck. My schedule was.

“Can agents do real work?" is the wrong question. One model in a chat is an intern with no tools and no boss. A small roster with clear owners starts to feel like a studio. The UX problem isnt the prompt. Its the org char

I was considering switching from OpenClaw to Hermes. I didn't take the leap of faith, so now I have one agent on OpenClaw and another on Hermes, both in my Telegram. One agent runs completely on its own remote digital computer, while the other runs on dedicated hardware at home. I actually like the different setups, each with their respective pros and cons.

Playing with Grok4 imagine be like

when I finally have a break: Okay what should I work on next?