Good AOE 1: Agent Orchestration Engineering Is DCC, Not SWE

There's already an industry that contends with large teams making rapidly produced, iterated, discarded, and reused artifacts in a variety of modalities — Digital Content Creation

DemigirlbassNov 19, 2026

The Coder Is Dead, Long Live The Coder

It's very tempting to think of Agent Orchestration Engineering as a natural extension of Software Engineering. After all, the tools that enable AOE are software. Software Engineering gave us Agent Orchestration Engineering, in a very direct, tangible sense.1

Furthermore, SWEs are also the ones who most readily embraced AOE. While there's been a smattering of adoption across work like legal, accounting, administration, or comms, using AOE to force-multiply existing talents and streamline work, the full-bore disruptive gutshot has only so far really hit SWE, a field where, if you aren't at least somewhat proficient with agents, you are now widely regarded as unemployable.2

All of this naturally pulls us to regard AOE and SWE as being the same thing, the first just an evolution or next stage of the second.

And also we SWEs are entitled, self-centered gits with all the worldliness of a Pikmin. That is also quite relevant to these assumptions.

You can see how this assumption leads us to think, to build, to act a certain way.

Consequently, you can also see how it causes us to eat shit a certain way.

Agent orchestration tools orient themselves around CLIs and TUIs.3 Workflows, tooling, even ideas of correctness all orient around how we build software. Tools are built first and foremost to accommodate the mental models, tastes, and sensibilities of SWEs, only.

This is, in essence, the 2010s App Bubble problem of "every developer lives in San Francisco, has constant amazing Internet, and owns a current-gen iPhone", which put every website and app into React Hell for 10 years until we were redeemed by Grace of Claude.

Thankfully, while we're still playing stupid games, this time the stupid prizes are coming along much faster. With as much money on the table and, consequently, on the line as there is from non-SWE userbases, vendors like OpenAI, Anthropic, and even Meta have started trying to make their agent orchestration tools more accessible to a more general audience.

Things like Claude Duet, ChatGPT for Work, or Meta Muse are all oriented towards making vibecoding...I hesitate to say accessible, it's already quite accessible. It's moreso a matter of making it appealing. Normal people, the kind who don't have opinions on Vi descendants, find a degree of mental anguish in merely interfacing with a CLI, and don't know the difference between a CLI and a TUI (which is fine).

I do not hold any aspect of this against them, nor do I say any of this to be derogatory. Using CLIs does suck shit, and we as SWEs got very used to being cozy in them because they were the cost of entry for doing the things we otherwise loved. We still dropped them like hot potatoes when GUIs came along that could do the same things (VSCode's prominence is testament to that).

But if we want vibecoding to be appealing to the masses of people who didn't give themselves novel mental illnesses about text editors, we have to package it in ways that don't make them sound like Odysseus at the end of The Horse and The Infant.

This in itself is about as far as one can reasonably get from novel observation. Yes, people find the machine god in the terminal scary and the machine god with the fun little hat in their phone or browser approachable and fun to use. That alone is not really worth a blog post.

However, I think for the most part the larger discussion has failed to develop beyond this shallow observation. Fundamentally, most of the people working at this perceive a certain parallel gap between the General Public, and the Sacred Ethylene-Huffing Pythiae of the SWE profession. Credit where it's due, there has been some effort in bridging the gap, in making more people able to perform at least some of the functions of the SWE. Where there has been tragically less consideration, is in graduating through this crippling lens of binary technical castes.

This blog post is the first in what I hope will be a series where I discuss the concept of Agent Orchestration Engineering, as a novel discipline with a novel adherent, and as our third lens (though I would be overjoyed if someone would take me beyond this trinary, too). This first part is about finding inspiration in production and engineering lineages beyond just Software Engineering,

Finding More Giants On Whose Shoulders To Stand

My current job title is Agent Orchestration Engineer at Field Level Tech. If you're wondering where I got the term from, or why I seem to have such an odd attachment to it, it's because it's what it says in my CV and my email signature, but also because it does genuinely do a great job of describing what I actually get paid to do.

While I was originally hired to be an SWE, and while I do consider myself one, because I do do4 a lot of it even in my free time, it is fundamentally not my capital-J Job. The bulk of my work the past few weeks as of time of writing, for example, has mostly been copywriting and illustration for our social media, with occasional UX work on internal tools.

Now, you could put the term SWE on the rack and stretch it till it screams "engineering done on or using software", but see my above statement about Pikmin for my view on that. I think AOE as a term is simply a far more accurate to and descriptive of what I actually do.

However, six months ago as of time of writing, I was not an AOE. I was scarcely even an SWE, honestly. I was, in fact, a 3D artist5! Obviously I'm not immune to bias or overgeneralizing a relatively narrow worldview, and there's lots more places that have just as much a right to be used as sources of inspiration as DCC does.

I would say that if you know about such a field or discipline, please write your own version of this post and then send it to me! I want to know what the alternative perspectives and milieus are that could be applied to this, and I'm always happy to be casting a wider net and hopefully developing into a better, more holistic engineer!

However, Digital Content Creation is what I know, right now, and I think there is a lot of meat on this particular bone that AOEs would do well to gnaw off, be it for day-to-day workflows, mental models, production pipelines, or even for designing a Harness6

Producer, Developer, Maker...Builder?

Coders and 3D artists both have one quite critical similarity. Do indulge me here for a moment — I know this next bit is a bit...inflammatory, but I am saying it anyway because I think it is important to say it.

In both code and 3DCG, there is a vast deal of Art. Capital-A Art, not just art. Regardless of how you define it, its presence and richness in these disciplines is undeniable.

And in both of these disciplines, we often confuse the Art for the job. Because your job as an artist is not to make Art. And your job as a coder is not to make Art. Your job, in both cases, is to produce useful software-based artifacts which either make your colleagues' lives easier, or let your employer make money.

If you remove the software caveat from the above, you may note, if you have a keen professional proprioception, that what I have just described is Engineering, on some level. A more accurate and encompassing definition would probably be "solve problems and create desired/favourable outcomes" but that's more philosophical than what this installment is about.

At the end of the day, whatever our email or forum signatures may say — Producer, Developer, Maker7 — we are all essentially the same thing. Just different facets of it. Different colours of light reflected through the same prism, different forks taken in the same long, winding road.

I understand some people find the term builder extremely grating, and so that they may fain benefit from my writing I shall spare them the torment of having to read it. Honestly, I don't know what the term I should use is. I'm still trying to figure that out as I write this, and outline subsequent posts where I make my argument for why the term is so damn important.

Some seditious heterodox elements may even suggest that I should've done the introductory posts before the one where I talk about practically actionable ideas but, well, I didn't get to the philosophy for this until I worked out why the practical stuff was so appealing, so. Sue me.

Okay You've Convinced Me—

No I haven't.

No You Haven't But You've Interested Me Enough to Where I Wanna Know What Practical Advice Falls Out From Shaking You Vigorously

Honesty is the foundation of any healthy relationship, darling. <3

Git Is A Great Start; Don't Let It Be The End

Git is a titan, and for very good reasons. It is genuinely a miracle of software engineering, and it still holds the coveted station of "VCS That Is Basically Good Enough For Basically Most People", which if you are any sort of seasoned engineer you know may as well be the Aegis of Champions.

Hey what's another word for titan again?

With code being essentially a metered commodity at API price, now is the best time there has ever been to stand on Git's shoulders and build something that works for you and your team, a lot better than Basically Good Enough.

DCC is a great place to find a second set of shoulders to stand on8. Such productions are no strangers to paces matching or far exceeding the frenetic tempos of agent-amplified software construction, GitLab issue notifications kicking down our doors as we struggle to queue up enough CI capacity to let us go home on time.

The fundamental difference is that SWE has always oriented its mental model9 around contributing code. Mud for a dam, or bricks for a wall. Many workers, each doing their relatively menial part to bring the whole together over time.

This makes it great as a way to manage agents, since they are in fact menial workers stitching together code like little termites pooping out a termitarium. Which is a less...flattering view of the humans overseeing the operation, and a less effective way of organizing them.

If you think of the purpose of your VCS as helping you stitch together disparate, asynchronous bits and bobbins of code with your colleagues, you will soon drown in an explosion of yarn, because you can so quickly build so fucking many of those.

If, instead, you take on the mental model of DCC — an industry where work is seldom code-shaped, and often opaque-binary-that-no-one-could-possibly-manually-review-every-little-bit-of-even-if-they-really-really-wanted-to-and-so-you-just-have-to-trust-that-your-colleagues-did-their-job-right1011 DEEP BREATH shaped — then you can instead start seeing a much more effective use for your VCS: providing a communication bus and visibility for milestone contributions to the project, which may be done asynchronously and then reconciled together right before they get sent off to the next department.

Maintain a Communal Tool Bin

You will probably end up making a lot of cheap, disposable tooling that is good for accomplishing one thing that you need to accomplish right now and otherwise basically useless. That goes in the bin. It's cheap, it's disposable, someone else can do it too.

And then occasionally you will find yourself making the same tooling over and over again. Or you'll make a tool, and it'll be good. And you will be tempted to flesh it out into a proper bit of hardware that you can point to and be proud of.

Do not do this! Bad idea. No good.

Such tooling, however, does not go in the bin.

It goes in the Bin.

Any sort of DCC studio, be it VFX, 3D animation, or game development, has a giant pile of internal tooling that is freely accessible for any of their artists to use at any time. Because at some point, three years ago, another artist had to do the exact same effect you need right now, asked a technical artist to build a tool for it, and they never threw it out because code used to cost something dammit.

But even now that code costs nothing, having a big bin full of tools that you can just reach into and grab a thing from which exists and works and you know is good (enough, at any rate), can save so much in the opportunity costs of digressing into a side project for just that one thing that you need to do before you can continue.

At FLT, right now that primarily means Skills (to encode things like branding, voice, or copywriting abilities), and organizational tools like Markdown maps that allow for more effectively juggling agents across broader, more organizational-scale work.

We use Git as the backbone of these, because Git, for lack of a better word, is good. But while we may have started with Git, we have not ended on it.

Use Software To Do Engineering, Not The Other Way Around

Setting aside my downright corrosive disdain for both SWEs and artists for a moment, forcing the two to work together did produce one very good outcome: it forced the boffins to set aside their...eccentricities long enough to build software that allowed artists — a decidedly non-technical crowd at the best of times, and outright anti-technical at the worst of them — to do, in essence, programming.

While 3D art is ultimately called art, the engineering element of it is still decidedly alive and well in its day to day use, despite the fact that the vast majority of 3D artists will never write a single line of Python or Lua or whatever crime against ergonomics your DCC suite of choice uses for plugins.

The point I'm making, if you can spot it through the gaps in between the slats of my numerous grudges and vendettas, is that 3D art is fundamentally an endeavour wherein engineering is performed, and software is used to do so, but it does not become subsumed by the pursuit of Software Engineering.

SWE is rather intrinsically a discipline that subsumes everything it comes into contact with, or at least makes its damnedest efforts towards doing so. Prior to agents — which, incidentally, is when most of the canonical tooling and methodology for SWE was built — being an SWE required a degree of dedicated study which made it largely incompatible with anything you couldn't learn on the job while doing SWE.

We tried to ease this, and we largely succeeded. For all my bitching, the fact is Python let generations of boffins write simple and sophisticated automations without having to be SWEs to do so12, and Lua did genuinely spark a lot of peoples' interest in programming by letting them use Turtles to automine entire chunks without building a Buildcraft Quarry13.

But you did ultimately still have to be a boffin. What software like Blender (my beloved) accomplished, that nothing else prior to agents really managed to get off the ground in quite the same way, is building a scaffolding atop a deeply technical discipline that enabled people to engage with it without being boffins. It created intuitive, naturalistic interface surfaces through which people could build without knowing why the bricks were yellow.14

An AOE is Not An SWE

And our work does not have to be code-shaped. Not in how we work on it, with the rest of our team, our beloved colleagues15. Not in how we keep the memory of our work, what wisdom it may grant us and ours. And not in carving the grooves where our hands will sit as we build something greater than ourselves, build the stuff of raw ambition, as we steal fire from self-made gods, and use it to illuminate a brighter tomorrow.

This life is enough. We lay the trail for those who come after.


  1. I am of course asking you implicitly here to regard AOE and SWE as two genuinely separate disciplines, and to consider AOE to be a genuine, meaningful separate engineering discipline, which is already a bit of an ask, but I do feel it is a necessary one for any of the rest of what I'm about to say to make sense.
  2. Controversially I do regard this as basically a good thing. Artisanal hand-crafted code is fine in the same way Patek Phillipe watches and hand-thrown crockery are fine. Neat hobby projects, fun for people who are willing and able to pay the premium, but the fact that everyone else can now buy Casio F-91s and the Rs. 350 mass produced generic ceramic plates from Imtiaz is, for society as a whole and most people living in it, a net positive.
  3. I love TUIs, but it's not because they're superior to GUIs, it's cause they're a lot harder to fuck up and all of you people are fucking awful at building good GUIs.
  4. hehe doodoo
  5. Not a good one lol
  6. Don't design a harness. Awful idea. This shit is exhausting and it is so hard to come up with ideas that are both genuinely novel and that someone else hasn't already done. Just use some existing one that works Well Enough, building your own is really basically never worth the effort.
  7. Oh hey like the heading!
  8. SEE I HAVE SOME THEMATIC COHESION HERE, I'M TYING THIS SHIT BACK
  9. Yes I'm back to talking about mental models, building things is cheap, but thinking things is still really hard!
  10. If you don't or can't trust your colleagues to do their jobs right, I cannot help you. That is a level of dysfunction so profound that if I could fix it, I would charge money to do so.
  11. Also yes I am sort of taking a dig at the "you need to read the code" people. It's petty and indulgent but I am a petty woman and were I more resistant to the temptations of indulgence I'd probably have a smaller band size than 46 now wouldn't I?
  12. Even if I wish it'd done that without tarnishing its legacy with the horrors of pip and semantic whitespace.
  13. For those born before 9/11, this is a reference to the Minecraft mods ComputerCraft (which introduced programmable robots called Turtles, programmed in Lua), and Buildcraft (which could frankly be the subject of an entire blog post by itself exploring its influence and complicated legacy).
  14. Where I live bricks are mostly yellow or maybe bright orange, I have literally never seen a red brick in my life, I'm pretty sure they made those up for television.
  15. I mean I love mine at least I don't know about any of you.

Never miss a post

Get new writing sent to your inbox.

Subscribe

2