Less Miserable Tech
  • Home
  • Leave a Review
  • Main Site
  • Field Notes
  • Start Here
  • Episodes
    Season 1 1 All Episodes 2
  • Follow
    iHeartRadio podcast player icon iHeartRadio
    All Players
  • Search
Contact
Contact
Search
Oct. 30, 2026

Make Your AI Agent Less Miserable

Make Your AI Agent Less Miserable
Less Miserable Tech
Make Your AI Agent Less Miserable
Your browser does not support the audio tag.
iHeartRadio podcast player icon iHeartRadio iHeartRadio
More
YouTube RSS Feed Spotify PocketCasts Podchaser Apple Podcasts Castbox Podcast Addict Amazon Music TuneIn
iHeartRadio podcast player icon iHeartRadio iHeartRadio
More
YouTube RSS Feed Spotify PocketCasts Podchaser Apple Podcasts Castbox Podcast Addict Amazon Music TuneIn
Season 1
  • Show Notes
  • Transcript

My AI agent could do useful work. The little things that made it frustrating to use kept getting in the way.

While setting up Milo, my Hermes agent, I kept running into friction: long silences, progress updates that just said “Python,” and no easy way to stop a long answer or get the agent into a conversation.

Each annoyance became one small fix: more useful progress updates, a Stop button, a Join button. The interaction got smooth enough that I wanted to use it regularly.

In this episode:

  • Finding the little things that make you avoid using a tool
  • Turning vague status messages into useful, sometimes funny updates
  • Solving one friction point at a time
  • Why your simplest useful fix may look different from mine

Try this with your own agent: “Here's the problem I'm trying to solve. What's the simplest fix?”

Less Miserable Tech is a podcast and field journal about real workflows: where they get stuck, what small fixes actually help, and how to make systems fit the people doing the work.

Website: https://lessmiserabletech.com/

Watch, listen and read transcripts: https://podcast.lessmiserabletech.com/

Speaker:00:00:00

I was setting up my own agent.

 

Speaker:00:00:01

So there's a framework called Hermes that you can put, uh, AI

 

Speaker:00:00:05

agents inside of, and it's a whole framework for managing stuff and doing

 

Speaker:00:00:10

things, and it's like, okay, great.

 

Speaker:00:00:11

I mean, I've been using AI for three years.

 

Speaker:00:00:14

We were on the b- we've been on the bleeding edge all the way up until now.

 

Speaker:00:00:18

And so I started setting up my agent, but I kept having all these little

 

Speaker:00:00:22

friction points, things that bothered me.

 

Speaker:00:00:24

It's like I got in Telegram.

 

Speaker:00:00:26

Okay, great, but you gotta press and hold to talk to it, and I'm driving.

 

Speaker:00:00:30

It's like, well, I don't wanna drive.

 

Speaker:00:00:30

Can… Is there more of an interaction we can just talk, right?

 

Speaker:00:00:34

And it's like, "Well, sure. You can do that in Discord." And I'm

 

Speaker:00:00:36

like, "Okay, great. Let's set up Discord." So that little pain in

 

Speaker:00:00:40

the butt, but I got it set up.

 

Speaker:00:00:42

And then it's like, okay, I can talk, but then there's these long silences.

 

Speaker:00:00:48

I'm like, "Ooh, are we still talking? What's going on?" 'Cause

 

Speaker:00:00:51

it's not interactive, right?

 

Speaker:00:00:52

It's not in the immediate.

 

Speaker:00:00:54

What, what happens is you, you tell Hermes, or in my case, I named him

 

Speaker:00:00:57

Milo, uh, you give him an instruction, and he goes through a series of

 

Speaker:00:01:02

steps to answer your question.

 

Speaker:00:01:04

He does an amazing job, and it has, it builds, uh, information

 

Speaker:00:01:07

about you and your problems and what you're working on over time.

 

Speaker:00:01:10

I have the archive of dubious certainty is set up for me

 

Speaker:00:01:14

that has all these, this data.

 

Speaker:00:01:15

I use Honcho for back-end memory, as well as this local stuff that it does.

 

Speaker:00:01:19

So it takes a while.

 

Speaker:00:01:21

So I said, "Well, can you at least tell me what you're doing so I

 

Speaker:00:01:23

know you're not dead?" Right?

 

Speaker:00:01:24

That the thing's actually working, 'cause sometimes tech fails.

 

Speaker:00:01:28

It goes, "Sure.

 

Speaker:00:01:29

Here's what I'm doing." And it literally said, 'cause it's running

 

Speaker:00:01:32

these things called Python scripts with code, and it goes, "Python Python.

 

Speaker:00:01:39

Python.

 

Speaker:00:01:39

I'm like, "Oh my God, can you please…" Oh, at first it was,

 

Speaker:00:01:43

"Terminal, terminal, terminal." Like, "Can you please tell me what you're

 

Speaker:00:01:45

doing?" He goes, "Python, Python."

 

Speaker:00:01:47

I said, "Oh, Milo, you're killing me here. Can we please, can you make it more

 

Speaker:00:01:53

interesting?" So I'm using Groq with a Q, which is a high-speed inference system.

 

Speaker:00:01:58

I'm using it to do quick analysis, to just give it the whole thread

 

Speaker:00:02:01

and have it give you back something to tell me about what's happening.

 

Speaker:00:02:05

And suddenly now I'm getting intelligent stuff, like when I do a long, a

 

Speaker:00:02:10

long text, it comes back and says, "Oh, I'm glad you took a breath."

 

Speaker:00:02:13

Uh, and it analyzes what I said, and then it tells me what it's doing.

 

Speaker:00:02:18

And when it's like, "Uh, running that search again. Hopefully we're gonna find

 

Speaker:00:02:22

something." You know, I'm, I'm chasing that down like a caffeinated truffle pig.

 

Speaker:00:02:26

You know, 'cause I have dry wit, I have a little bit of humor in it.

 

Speaker:00:02:29

It makes it fun.

 

Speaker:00:02:30

So while I'm driving along, I can do that, then all I have to do

 

Speaker:00:02:32

is unmute, talk to him, and mute.

 

Speaker:00:02:36

Well, then of course he goes on these long, long explanations, 'cause sometimes

 

Speaker:00:02:40

I'm using Codex subscription, which is actually GPT, and sometimes it can go

 

Speaker:00:02:45

very long, just like I'm doing right now.

 

Speaker:00:02:48

And it was like, "Okay, you're killing me here.

 

Speaker:00:02:50

How do I stop you?" So then I had to create a stop button.

 

Speaker:00:02:54

So now I have a stop button.

 

Speaker:00:02:55

It's like, well, then getting him to join the chat was difficult,

 

Speaker:00:02:57

so I have a join button now.

 

Speaker:00:02:59

And what I've done is each time a little piece of friction, things that made it

 

Speaker:00:03:03

difficult to use the technology to get what I wanted done, you identify it, and

 

Speaker:00:03:09

then you solve the smallest usable fix.

 

Speaker:00:03:12

Uh, uh, 'cause once you identify that problem, now you can solve that.

 

Speaker:00:03:16

And now of course I've done that- 20 different ways to now where I find

 

Speaker:00:03:21

a very smooth experience talking and dealing with, um, with, with Milo

 

Speaker:00:03:26

and my Hermes agent, and I'm able to do a lot of things now that make it

 

Speaker:00:03:30

super easy because I enjoy doing it.

 

Speaker:00:03:32

It-- the friction of the actual interaction is low enough that I

 

Speaker:00:03:36

use it regularly, where before it was a big old pain in the butt.

 

Speaker:00:03:40

By the way, I'm gonna a- create my own web app interface to make that even easier.

 

Speaker:00:03:45

Now, the lesson for most people is, like, your problem isn't my problem.

 

Speaker:00:03:49

Your goofiness… Like, I chose a very specific voice, and it was a

 

Speaker:00:03:52

voice that cost a little bit of money.

 

Speaker:00:03:54

It's from Groq with a Q. They have the best voice, in my opinion, for whatever.

 

Speaker:00:03:59

I needed that intonations.

 

Speaker:00:04:00

I needed its way that it said things and phrases.

 

Speaker:00:04:03

It made it more natural for me.

 

Speaker:00:04:05

And so by doing that, um, uh, again, I use it.

 

Speaker:00:04:10

Your issue's gonna be something completely different.

 

Speaker:00:04:13

Like, I've set up now these, these agents for two other people.

 

Speaker:00:04:16

They're using them completely differently than me.

 

Speaker:00:04:19

Their friction points are different, and they'll tell me, and I can

 

Speaker:00:04:21

help them out, or they tell their own agent, and he'll go fix it.

 

Speaker:00:04:24

So that's what you need to identify.

 

Speaker:00:04:26

What is it that slows you down and makes you go, "Ugh, I don't

 

Speaker:00:04:30

wanna do that." It's like, "Okay, yeah, I gotta go do that again.

 

Speaker:00:04:32

I gotta copy and paste 30 times, or I gotta go deal with this thing that

 

Speaker:00:04:36

I hate dealing with." And when you're able to, um, identify that, then we can

 

Speaker:00:04:43

help-- Not we, but you can figure out then, okay, how do we solve that best?

 

Speaker:00:04:46

What's the tool?

 

Speaker:00:04:47

What's the process?

 

Speaker:00:04:48

What's the thing that's gonna fix that thing?

 

Speaker:00:04:51

It's not about, "I created a whole new system. I created salesforce.com for

 

Speaker:00:04:56

myself. Isn't it amazing?" It's like, well, no, I solved the problem, and

 

Speaker:00:04:59

now I can move on to the next thing.

 

Speaker:00:05:01

And it's about making tech less miserable while using it and actually being

 

Speaker:00:05:06

productive and useful in solving things.

 

Speaker:00:05:09

And the same thing can be true for you.

 

Speaker:00:05:11

So, um, is this when I say leave a comment and subscribe?

 

Speaker:00:05:16

Like, I don't know.

 

Speaker:00:05:17

I'm just saying think about that and identify your friction,

 

Speaker:00:05:21

and the tool may reveal itself.

 

Speaker:00:05:22

Or just ask, "Here's the problem I'm trying to solve." Ask your own a- agent,

 

Speaker:00:05:27

"Here's a problem I'm trying to solve.

 

Speaker:00:05:29

What's the simplest fix I might have for that?" And then ask for a more

 

Speaker:00:05:32

complicated fix because sometimes they give you some really janky stuff.

 

Speaker:00:05:36

But it's like, that's okay, and now you're solving something that's real.

 

Speaker:00:05:41

Cheers.

 

 

Less Miserable Tech

Technology should make work less miserable. Not become a whole extra job.

Less Miserable Tech is a podcast and field journal about real workflows: where they get stuck, what small fixes actually help, and how to make systems fit the people doing the work.

The lessons we’ve learned the hard way become the principles we use when the next messy problem shows up.

  • Leave a Review
  • Contact
  • Main Site
  • Field Notes
  • Start Here
  • Episodes
  • © 2026 Less Miserable Tech
  • Podcast Website by Podpage