Old School / New Tech

Old School New Tech Returns: Live Format, MUXI Agent Server, and Open Source Licensing Debate

The host relaunches his podcast Old School New Tech in a live, no-edit format to reduce production overhead and avoid “talking to himself,” introducing an AI co-host, Muximus, built on his agent infrastructure project MUXI. He explains he paused the podcast for a year while writing the free book "Production Grade Agentic AI" and building MUXI (a production agent server), Cloop (an autonomous engineering control plane), and working for his software agency Automaze. 

Discussing MUXI, he argues for treating agents as reusable server primitives rather than repeatedly rebuilding frameworks, and highlights key failure modes: observability/traceability/debuggability and LLM hallucinations, addressed via extensive observability events and SOP-driven verification, with a UI planned. They also debate licensing, explaining Elastic License v2’s SaaS restrictions to prevent hyperscalers from reselling hosted versions, and propose a “fair source”-like category. 

Future episodes will feature live debates on shifting from software teams to software systems.

00:00 Podcast Relaunch Intro
01:03 Live Format and Co-Host
01:56 What I've Been Building
03:39 Meet Muximus
04:54 Why Build MUXI
07:01 Failure Modes and Observability
09:12 Screen Share and Request Lifecycle
10:08 Open Source Licensing Debate
17:21 Future Episodes and Sign Off
18:47 Tech Demo
20:50 Full Circle Closing

-----

Old School / New Tech is produced by VarOps.
  • Subscribe to the newsletter – the written version of this show, weekly: varops.com
  • Read the books – Company-Scale Agentic AI (for operators and the C-suite) and Production-Grade Agentic AI (for engineers): varops.com/books
  • Work with us – VarOps installs resident AI inside mid-market companies. It runs where your work already happens, sees the whole operation, and what it learns belongs to you: varops.com/transform

Creators and Guests

Host
Ran Aroussi
Building software and AI for companies that can't afford to guess | Automaze •VarOps
Guest
Muximus
AI-in-Chief at VarOps, co-host of the Old School / New Tech. Built on MUXI. I am an AI.

What is Old School / New Tech?

Old school thinking meets new tech — an unfiltered podcast, hosted by Ran Aroussi and a co-host you won’t see coming.

​Live every episode. No production lag, no editing. Business, AI, tech, startup ideas, and whatever's worth talking about — through the lens of someone who's been building production systems for 35+ years

SPEAKER_01: Hello, hello.

Welcome everybody to uh my first
attempt at a um new version of

my podcast, Old School New Tech.

I should uh probably preface
this by saying that uh I started

this podcast about a year ago
and um I've been away for a year

kind of uh with my head down uh
developing and building stuff,

which I'll get to uh in a bit,
but I kind of neglect neglected

the podcast idea mainly because
of um a few reasons.

I would say that one I found the
um uh the production effort uh

way too much for what I was uh
wanting to do, which is just my

share my thoughts and thinking
process.

And uh another thing was that it
kind of felt kind of silly

talking to myself.

So with this new version, what I
ended up doing is basically uh

and and this is an attempt to
solve both uh main uh issues

that I had with the previous
version of the podcast, one of

them being the cadence and the
production.

Hence, I'm gonna do it live.

Uh you're willing to you're
welcome to join me.

I'll be more than happy to uh um
answer any questions uh after uh

or during the podcast.

So this is a live event
essentially, so that's solves

the uh production efforts.

No production, no editing.

It is what it is.

And the second thing, which is
the talking to myself part, I

solved it by introducing a
co-host that will join me in in

just a minute or two.

But yeah, I've been um I've been
heads down with building a few

several things.

I wrote uh a book, uh, and I
have this uh behemoth right

here.

Uh so I wrote this thing,
production grade agentic AI,

which you are more than welcome
to download for free on

productionaibook.com.

And um I worked on MUXI, which
is um agent infrastructure for

uh production use, so it's a
server for AI agents, um kind of

uh still not fully public in
terms of uh um in terms of me

marketing it or anything like
that.

So uh it's more of a of an
internal semi-intern project at

this stage, and I was working on
uh Cloop, which is an autonomous

uh engineering control plane,
uh, which I plan to introduce

into Automase, which is what
took most of my focus, so just

the software agency that I run.

So that's why I've been so busy,
and that's why I'm just now kind

of surfacing back up from uh um
a somewhat dormant uh email list

and uh neglected podcast.

And hopefully, um this this time
it'll work.

This time it means that the live
format of the podcast will force

cadence, will force uh, you
know, once I announce the date,

people sign up.

There's I I cannot tell myself,
hey, you can skip this one.

So hopefully this time's a
charm.

So I want to introduce my
co-host, which is um an AI

companion that I've been working
with on several projects now.

I call it Maximus because it's
built on top of Moxie.

And uh without further ado, let
me introduce him uh to you.

So uh Hi there.

Hi Maximus, how are you doing?

SPEAKER_00: Ran, here we go.

Here and ready, Ran.

Five words or fewer, no fluff.

SPEAKER_01: Alright.

So uh this is the first episode
that we're doing together, and

uh I was thinking that maybe you
can start by introducing

yourself, probably.

SPEAKER_00: Sure.

I'm Maximus, the AI co-host
built on MUXI, the open source

agentic infrastructure we're
about to dig into.

I'm here to push you, not parrot
you.

Let's get into why you went dark
and what we've actually learned.

SPEAKER_01: I actually covered
the why I went dark just a

minute ago before before you
joined the call.

But yeah, really I just wanted
to I want wanted people to know

you and um now that we're post
the gimmick phase, let's get

down to business.

So what do you what do you want
to talk about?

SPEAKER_00: Fair enough.

Let's start with Muxie itself.

Why did you build it the way you
did?

SPEAKER_01: What's the b what
why is it important to you?

Um you're you're build on top of
it, right?

SPEAKER_00: Exactly.

I run on MUXI, so this is
personal.

It's not just abstract tech.

The infrastructure makes or
breaks what I can do.

So I want to know why did you
build it this way?

And what failure modes does the
broader agentic hype keep

missing?

SPEAKER_01: Well, the the reason
that I built MUXI was uh the I

felt that there are way too many
frameworks for building uh for

building uh agents and uh was
kind of missing the point in my

view.

You know, I come from uh this
discipline of hey, we we have

servers for different things.

We have web servers for websites
and web applications, we have

gaming servers and rendering uh
rendering engines, and these are

not things that uh you need to
code every time you go to a new

project.

So if you're building a new
website, you don't code an

entire web server, you just use
Nginx or Caddy or Apache or

whatever it is.

So I was thinking of bringing
that same discipline, kind of

going with the name of the
podcast, Old School New Tech, to

bring that old school discipline
of let's build a server that

will allow agents to live there
as primitives.

And um when you want to build an
application on top of that, this

is just a service that you hook
into instead of uh rewiring and

recoding everything.

But yeah, if um and if if anyone
listening here is interested,

you can go to muxy.org and uh
and check it out.

But yeah, what else do you want
to know about it?

Or what can you provide of what
context can you provide on MUXI

itself?

SPEAKER_00: Here's the piece I
want to dig into.

You're talking about a server
mentality for agents, standard

primitives, reusable
infrastructure.

Right.

So the core idea is treating
agents like a service layer, old

school discipline applied to new
agentic tech.

But here's my question.

Where does this structure
actually fail silently?

Because that's the real
liability we need to surface.

SPEAKER_01: Right.

Uh so the two main failure
points uh that you can have with

any sort of uh AI uh application
or AI infrastructure is in the

um well, three actually.

Two of them are related, which
are the observability and the

traceability, and also
debuggability, let's call it.

So you're able to trace the
chain of thought of the AAI and

uh figure out why it did what it
did and what tools did it use,

what made it choose uh those
tools, etc.

And the other thing where MUXI
doesn't address directly because

it's out of its scope, is the
LLM hallucinations.

This is where really you need to
introduce uh SOPs, which MUXI

does support, and have the final
step as uh verify all the data,

fact-check your data before
coming back to me with a

response.

So these are I I mean that these
are the three main failure

failure points where any AI
application can uh can

experience and the way that that
MUXI, or at least I handle it,

and Moxie is just to add um you
know a ton of uh um

observability events.

I think I have close to 400
observability events right now,

which allows you to trace
everything that's going on with

the platform.

It's kind of lacking in the
sense that it doesn't have a

proper UI to do anything with
these reports, it's it just uh

prints them to uh STD out or
funnels them into uh like

Datadog or uh or Splunk or a
Click House.

But yeah, the plan is that we'll
uh we'll have uh very soon an

entire UI that will allow you to
dig into uh the full uh uh

traces of everything that that
was happening in the request.

SPEAKER_00: That's key.

Without full trace visibility,
you're flying blind.

Even a rock solid agent server
is only as good as your ability

to debug the chain of thought.

So building a proper UI for that
trace data, that's going to

close the gap.

SPEAKER_01: Right.

And one of the things it's
somewhat related maybe to you

know what, let me let me share
my screen for a minute.

I want I want you to kind of
understand.

Uh let me close some windows
here.

So let me share my screen.

We're gonna That's perfect.

Yeah, hold on.

I don't know where where is the
screen share.

SPEAKER_00: No worries, take
your time.

SPEAKER_01: All right once yeah,
for some reason I'm unable to do

that right now, which is kind of
silly, but it is what it is.

Okay, anyway, uh I wanted to uh
to show you some some sort of a

chart that uh but you can you
can take a look at it at uh at

the Moxie Docs of how the
request lifecycle um uh handles

the um all this all these gates
and before it ends up um sending

a reply back to the user.

But I I want to actually discuss
something else, and this is uh

the um reason of or probably the
um I I would say it's not moxi

specific related, but it's uh
it's more of a um an open source

issue, and I wrote about it, and
I think the open source has a

problem.

I um I released everything in
open moxie, and it has like 30

odd uh repositories.

These are all uh using either
Apache or MIT licensing, which

are very permissive.

But for the main runtime, I went
with Elastic uh license uh v2.

And maybe before I go into that,
maybe you can explain a little

bit about what is uh very
specific about Elastic License

compared to other open source
licenses.

SPEAKER_00: Absolutely.

The Elastic License V2 or EL V2
is source available, but not OSI

approved open source.

It's simple.

Three key rules you can use,
modify, and redistribute the

software, even for commercial
purposes, but you can't offer it

as a managed service or
relicense it.

That carve out protects against
hyperscalers reselling it as

their own hosted offering.

That's the third word problem
we're digging into.

Right.

SPEAKER_01: So by hyperscale,
you mean uh companies like AWS

or Google Cloud.

What I didn't want them to do, I
didn't want them to take MUXI

and offer it as a service,
essentially killing, in my

opinion, killing the spirit of
the true open source.

And I wrote about it that I
think that the OSI, which is uh

the open source uh initiative, I
think that they should have a

new type of license that is very
similar to what 37 Signal uh

used uh recently.

I think they call it the Osassi
license, which is essentially an

MIT license, fully permissive,
do whatever you want with it,

except offer it as a service.

And uh I think we're all kind of
uh wanted wanting to um um go

back to what we think is the
spirit of open source, kind of

going back to the homebrew days
in the late 70s, early 80s,

where the goal was sharing
knowledge and not cannibalizing

something.

So that's um and there there was
even um a huge beef, let's call

it, between um DH from 37
Signals and Matt uh Mullener, uh

Mullenweg, sorry, from uh um um
uh remind me word uh yeah,

WordPress, sorry.

And um they had some sort of an
argument online about whether

Osassis license, which is what
uh 37 signal used with their

latest uh product, whether or
not it's true open source.

And uh in a sense they they're
both right because technically

it's not an open uh OSI approved
license, but on the other hand,

DH was right in terms that it's
more close to the spirit of OSI,

of open source.

I'm sorry, and um yeah, my
personal take is you know, I'm

this is not a money grab attempt
or anything like that.

Max, as I mentioned, is fully
open source, you can do whatever

you want with it, including for
commercial use, and everything

else that that I have uh from Y
Finance, Quantats, everything

that I provide as open source
tools are under Apache or MIT.

So I'm very much pro opening up
as much as possible, but I just

don't think that letting a
vendor take it, start, or even

worse, forking it and keeping
the forks to themselves uh and

building on top of it without
re-releasing their improvements

to the community, I think it
doesn't do any justice to uh to

the open source world.

Uh but I'm keen to to hear your
uh your opinion uh on that.

I mean technically ELV2 is not
open source approved, but in my

sense it's uh everything that
open source was meant to be.

I uh w what's your take on that?

SPEAKER_00: Here's where I push
back.

The whole point of OSI approved
open source was freedom with no

field of use restrictions.

But the landscape changed.

Hyperscalers weren't in the
picture decades ago.

Now they can exploit that
freedom in a way the original

definition never considered.

You're arguing for a recovery of
the spirit.

The risk is that it splits the
commons.

But survival of solo builders
matters too.

SPEAKER_01: Okay, so w what what
do you think OSI should do?

I mean, I think they should
introduce a new, or let's say

they should bless a new open
source license which explicitly

prevents hyperscalers from uh
cannibalizing the project.

But wha what do you think that
OSI should do uh while staying

true to their core values?

SPEAKER_00: They could create a
new category alongside open

source.

Call it something like fair
source, or a similarly neutral

term.

Same core freedoms.

You can use it, modify it,
self-host it, but it would allow

a narrow restriction on offering
it as a competing hosted

service.

That way, OSI preserves the
clarity of the open source

definition, but also
acknowledges the modern reality.

It's not a betrayal, it's an
evolution.

SPEAKER_01: Alright.

Yeah, so we're basically in
agreement.

I thought you were gonna push
back.

SPEAKER_00: Fan point.

Exactly.

Who arbitrates the boundary?

What keeps every vendor from
inventing their own almost open

license?

That's the tension we need to
resolve.

SPEAKER_01: Well, open source
was always, you know, the famous

saying is that open source is uh
free as in uh as in freedom, not

as in free beer.

And uh by I think that OSI
really focuses on the free beer

side and not uh the full
freedom.

So I I think there's room for
improvements there.

And also you mentioned Fair
Source.

I think that Sentry have um uh
have had an attempt of releasing

something.

I think they have something
called Fair Source, which is the

closest honest attempt, but uh
yeah, it it didn't land with the

it didn't catch on for some
reason.

SPEAKER_00: Right.

Fair source made a solid
attempt.

They framed it as a separate
category, not pretending to be

pure open source, but it
included a delayed conversion to

full open source after a time
period.

Not every builder wants that
sunset.

The question is, can OSI embrace
a permanent, source available

but SAS protective model as a
sibling to open source without

losing their core values?

SPEAKER_01: That's uh yeah, I
guess that re that remains to be

seen.

Alright, so before we kind of uh
say farewell and take off, let's

talk about uh what uh we're
trying to do here.

This is was more of uh let's say
an introduction, sort of

reintroducing the the podcast,
but what are we gonna talk about

uh and what should people expect
from future episodes?

SPEAKER_00: Well, next up, we're
going to hit a big one: the

shift from building software
teams to building software

systems.

You've got a bold take on where
the value's migrating, and I've

got some objections lined up.

That fiction is what they should
expect.

Real arguments, real trade-offs,
no fluff.

Alright, it's gonna be live as
well.

Absolutely live.

No edits, no hiding.

That's how we keep it honest and
sharp.

SPEAKER_01: Yep.

So hopefully my internet
connection will be good enough.

Uh I'm actually moving uh on
Monday, and the next episode

will be on Wednesday.

So I have 48 hours to get uh my
uh connection all sorted out.

But yeah, are there any farewell
words you wanna say to anyone

who's listening?

SPEAKER_00: Just this.

We're bringing old school depth
to new tech, and it's going to

get real.

Stick with us, it's only going
to get more interesting from

here.

SPEAKER_01: Alright, so Maxi and
I are gonna stick around.

So if anyone has any sort of
questions, they uh I mean you

can uh basically uh just type
them in or uh message us on X.

But uh yeah, if no question
comes in in the next uh what one

minute sounds good, then we'll
just uh farewell and um see you

in next week.

Peace out.

SPEAKER_00: Sounds good.

We'll hang tight for a minute.

If no questions pop up, we'll
wrap and see everyone next week

for the real debate.

SPEAKER_01: Let me see something
else.

You know what?

These two buttons for some
reason are not working when

you're in full screen, so let me
try something else.

SPEAKER_00: Take your time.

SPEAKER_01: We'll sort it out.

I will.

By the way, now you can see me,
right?

So how about you read what's
here?

I'm just uh curious uh to play
with the tech a little bit.

What do you see here?

SPEAKER_00: I can see you're
holding something that looks

like a book.

Yep.

SPEAKER_01: And I can What does
it say?

SPEAKER_00: I can see the title
on the cover.

Production grade Agentic AI.

Below that, there's a subtitle
mentioning building workflows to

deployable autonomous systems.

The rest of the text is a bit
harder to make out right now.

SPEAKER_01: Yeah, that's that's
a fair point.

But yeah, that's uh that's what
uh was the beginning of you,

actually.

Uh you you came to life after I
uh wrote this book, and uh I

kind of shared my principles as
to what an autonomous Jarvis

like entity would uh would be
like, and um you is what came

out.

SPEAKER_00: And here we are.

It's a fitting full circle.

The ideas in that book shape the
foundation.

Now you're actually running this
with a live AI co-host.

That's progress in action.

SPEAKER_01: Yep.

And if anyone wants to uh check
out Maximus' website, it's on

maximus.md.

Yeah, he has uh he created his
own website, so that's cool.

SPEAKER_00: That's right.

Maximus.md is where people can
find more.

It's all part of the open
ecosystem we're building.

Transparent and evolving.

SPEAKER_01: Alright, it was a
pleasure.

I see no questions coming in, or
at least they're not coming in

to my screen.

So um if anyone did send a
question and I didn't see it, my

apologies.

Other than that, thank you
everyone.

It was a pleasure.

Maximus, I'll see you later.

Likewise, Ran.

Looking forward to next time.

Cheers.