Eban Bisong

Founding engineer turned founder. Building talk it out, get clear, make your move. For people who think out loud 👇

[Eban Bisong @ebanbisong]

Hey I'm Eban.

- A career building software on small teams: first engineer hired, co-founder, tech lead, then VP of Engineering
- Recently left the VP seat to build on my own, zero outside investment
- ~20 AI agents running on my own server around the clock: they capture what I say, draft my follow-ups, run my companies' digests, and fix themselves when they crash
- Building Senvi in public, one piece of infrastructure at a time

I've spent my whole career building software, and somewhere along the way I learned I'm just as good at building the teams around it. But what really hooked me wasn't a language or a framework. It was the moment I realized the ceiling was gone.

With Claude Code on one side and an always-on agent like Hermes or OpenClaw on the other, you get both halves. The reactive one builds whatever you ask. The proactive one runs on its own while you sleep. Put them together and you can build almost anything.

That combination, reactive plus proactive, is the lens I see everything through now: build the system once, let it run, and compound.

Here's what I'll be posting about:

- Running my life and my companies on AI agents I host myself
- How to build personal AI infrastructure that does real work, not demos
- What actually breaks when agents run 24/7, and how I fix it
- Turning a technical edge into a product and an audience at the same time
- The traps that quietly kill solo builders before they ship

What's got me hooked is how much one person can build now. So I'm building in the open and sharing how it actually works.

My goal for 2026 is to show you the unedited version of building a one-person company that runs on agents.

Follow if you want the full build.

Let's go.

[Eban Bisong @ebanbisong]

Nobody replies to a message that was obviously blasted to a thousand people.

Two things actually get someone to respond.

One, it's about them. Something they recently posted, or something on their profile. It doesn't even have to tie back to what you're selling. It just has to show you're actually interested in them.

Two, it doesn't sound like AI slop. People can tell in a second. The real skill is getting your AI to sound like a person, not a bot.

So that's the system. My VA screenshots their profile or a recent post, and a skill has Hermes write the message off it, in a way that actually sounds human.

And honestly, that's the whole thing. People respond when it feels like you actually looked.

[Eban Bisong @ebanbisong]

July 4th is my favorite holiday.

We get to build our own paths here, chase whatever we're into, and we're not doing it alone, there's a whole community building right alongside us.

Even the smallest niche can turn into a real living, and it doesn't have to be your own company either, plenty of people contribute their own vision to a bigger company.

Man, I'm proud of this place.

Happy 250th Birthday, America.

[Eban Bisong @ebanbisong]

I tried to hand my cold outreach straight to a system and step away.

I hadn't done it enough myself to know what I even wanted from it. What good looked like. What I was actually going for.

So how was the system supposed to know?

That's the thing nobody tells you about scaling. You can't hand off something you haven't run yourself yet.

So I sent a few messages by hand first. Enough to get the reps, tweak what wasn't working, and get clear on what I actually wanted.

Then I built a system for it. Handed it to my VA, with Hermes handling the writing.

And honestly, that's the only reason it runs now. I knew exactly what I was giving it.

[Eban Bisong @ebanbisong]

Is anyone else's Fable 5 still disabled for Claude Code?

[Eban Bisong @ebanbisong]

A small rule I’m using more:

If I ask an agent to do the same thing twice, I try to stop asking.

Turn it into a skill.
Schedule it in Hermes/OpenClaw.
Create a hook.
Write the cold-readable doc.

I want the second ask to be the last time I ask.

[Eban Bisong @ebanbisong]

@greg_rog this is the part I’m trying to learn.

Would love to hear how this looked at the beginning.

Did you focus hard on one company, delegate, then rotate? Or did you start with multiple things already moving?

I’m early in this journey with a VA and trying to learn how the leverage actually gets built.

[Eban Bisong @ebanbisong]

A step I added at the end of my agent coding workflow:

Let it write the code.

Then let it quiz me on what it wrote.

Shipping = output.
Understanding = ownership.

Fast is great. But I still need the mental model clear enough to know what decision comes next.

[Eban Bisong @ebanbisong]

@gregisenberg 1. OpenClaw (testing Hermes stability rn)
2. Anthropic (goated)
3. Claude Code (just feels better)
4. Cloud models (still hunting for a use case I'm sure about)
5. Bootstrap
6. Best time
7. Layoffs first, then a wave of new companies

[Eban Bisong @ebanbisong]

I maxed out my Cursor BugBot limit for this month.

genuinely curious what everyone else is running for AI code review right now.

tried CodeRabbit, found it too noisy. and BugBot's $1-1.50 per review adds up fast at volume.

what's actually working for you?

[Greg @greg_rog]

The single most valuable AI skill in 2026:

Knowing what to automate.

Not how. The what.

Most founders automate the easy stuff and leave the hard stuff manual. The hard stuff is where the time is.

The easy stuff was already fast.

[Eban Bisong @ebanbisong]

Claude Fable 5? 👀

[Eban Bisong @ebanbisong]

@aarondfrancis a hook that updates the readme on every push. low stakes if it's wrong, runs on something you already do, and kills a chore you always forget. that's the one that got me going

[Eban Bisong @ebanbisong]

Had Claude audit my Claude Code history this week.

/forge, my full build command that takes a feature from spec to shipped. I ran it 46 times in two months, every single one by hand.

The routine stuff, the scheduled summaries, the check-ins that run on a cron, the cleanup that fires when something changes, all of that already loops without me. Wired it up, mostly forgot it exists.

A full build pipeline doesn't work that way, though. You can't just automate the trigger and trust the rest, because it's a different beast entirely.

The busywork loops the moment you add one line. A feature build hits a judgment call at every turn, and right now, those turns still need a person in the seat.

Looping the busywork is the easy part and honestly everyone's figuring that out right now.

Looping the thing that builds the whole feature is the problem I actually want to solve.

That's what I'm working on this week.

[Eban Bisong @ebanbisong]

Now that AI writes most of the code, I still make myself learn the mental models

When a big explanation comes back I break it down and say it back one piece at a time til we're aligned

The more I understand, the better the ideas I bring when we're designing

[Peter Steinberger 🦞 @steipete]

Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore.

You should be designing loops that prompt your agents.

[Eban Bisong @ebanbisong]

i've been doing NLP outcome exercises with my AI agent. kept failing the same quiz. every time, jumping to solutions before defining what i wanted. same pattern that produces bad prompts. if yours aren't landing, it might not be the prompt.

[Eban Bisong @ebanbisong]

@WizLikeWizard If you haven't already, I'd recommend gaining a solid understanding of OpenClaw's architecture and docs. It's helped me be more specific in telling Opus how to configure my instance

[Eban Bisong @ebanbisong]

checking out lossless-claw for openclaw

instead of throwing away old context when the window fills up, it compresses into layered summaries and keeps the originals

when the agent needs detail, it drills back down to the source

[Eban Bisong @ebanbisong]

builders avoid selling because it feels like a different skill. it's not. it's the same thing...figure out the problem, scope a solution, deliver it. the only difference is you're doing it in a conversation instead of in code