leaky valve cover gasket

Hey ya’ll -

I spend about an hour each morning training agents. I'm not sure if I love it or hate it.

I definitely wouldn't say that they're great employees just yet, but they're inexpensive (aside from my time), they're improving, and they generally do a good job helping me do more things more quickly.

But I'd rather talk to people.

Thanks for reading!

~Eric

Eric Boggs
CEO @ RevBoss

PS - as a friendly reminder, you’re receiving this email because you filled out a form on revboss.com at some point or because we interacted on LinkedIn at some point.

I’m the proud creator (owner? caretaker? future employee?) of a growing horde of robot minions that help me with:

Prospecting — separate routines for cold leads, warm leads, and triggered leads.
Content — ideas, drafts, comments, etc. all in one worker.
Company updates / ideas — this one is the least developed, very much WIP.

(The RevBoss team has already built some deeeep stuff that automates and prioritizes client work…and much of the stuff I’m building will turn into client-facing products — so holler if you’re curious about this kind of thing, RevBoss can help.)

Everything at RevBoss is connected to our Claude instance, so I can generally find whatever I’m looking for by asking an agent to dig into a call transcript or some other data source. It’s really fast and really convenient, but it means I’m doing quite a bit less of the “what are you working on?” type meetings with the team because can more easily “watch” RevBoss happen as a spectator.

I think this is generally positive — less time in meetings means more times doing the work, everyone can work more quickly, everyone can know what’s going on more easily, etc.

But I definitely miss the people collaboration.

That’s a leaky valve cover gasket on my beloved 2003 Toyota Tacoma. I don’t drive it very much and decided to get it running over the holiday weekend in large part because my oldest turns 15 this summer and will inherit my truck next summer. :)

Got it running no problem (new battery and cable) and drove it around the neighborhood feeling good about myself…until I smelled the burning smell and noticed the smoke under the hood.

The truck sat my driveway over the fall / winter and I think that the seals rotted or cracked. Not a catastrophic problem, but an annoyance that I could have avoided with just a little more care and attention.

I’ve done in-person coffee meet-ups with non-RevBoss-related work friends 2 weeks in a row and I’m looking for more. Reply to this email if you want to chat — can be a zoom…or a meet-up if you’re in Durham.

Anyway, back to the robots…

I had a bit of a last mile breakthrough over the past couple weeks when I stopped trying to run all of my agents through a web app and instead started building local agents in Claude Code.

I have Claude Code set up to:

-- dig through my sent mail (Gmail connector)
-- browse a few online accounts (Claude in Chrome plug-in)
-- receive email in a bcc (Sendgrid inbound parse + webhook)

...and use artifacts from these processes as a part of it's daily routines.

I set all of this up because I got tired of trying to tweak the instructions via chat and instead wanted to train the agent by having it watch what I actually do.

Every agent does a quick re-training step on every start, using the relevant bits I've told it to look for to shape it's work and instructions. It's exactly how I like to work -- lazy, smart, and effective! 🤠

It's made a noticeable impact and I expect (hope!) it’ll keep getting better:

-- the draft content ideas (like this one!) are sharper, more on-brand.
-- the warm-ish prospecting email drafts require increasingly less tweaking.
-- the permission-less value assets I'm building are basically on autopilot.
-- the "thinking about the business" feedback is actually useful

I'm doing all of this locally but many of the outputs are getting published on public URLs (via Netlify) or getting appended to company accessible artifacts in Google Drive.

So far these processes are just setting up the work for me to do -- I still have to send the email, post the thing, etc. This keeps me in the loop and serves as the quality assurance / training step in the workflow.

It feels a ways out before I let it actually do the work for me, though I'm honestly not sure if it actually needs to get all the way to that point.

Or maybe I’m just resisting the final nail in the “sip a cocktail and let the agents talk to the agents!” utopia coffin that we’ve been promised because I think it sounds like a crummy existence.