Say What You Do
Say What You Do
$150,000. That’s on the low end. Maybe more like $250,000.
Expect to spend that much or more on an outside consultant to stand up a Quality Management System (QMS) compliant with AS9100 Rev D, the aerospace industry’s standard. It’s the standard that says you’re a grown-up big-boy company, that your processes are documented, auditable, and not held together with bailing twine and night sweats. If you want to sell to primes, if you want to bid defense contracts, if you want to play in aerospace at all, ya gotta have it.
Standard playbook: a QMS consultant comes in, spends a few weeks to several months learning your business, then drafts your quality manual, builds out your procedures one at a time, maps your process flows, creates your forms and registers, trains your people, runs internal audits, and shepherds you through the whole certification gauntlet. Six to twelve months. Sometimes longer. If you’re a startup, you might hire a full-time quality manager instead. $120K salary, plus benefits, plus the six months of ramp time before they’ve actually produced anything.
Either way, it’s a serious cash outlay for a company still wearing short pants. This is one of those structural barriers to entry that keeps SmallCorp out of regulated markets. I heart regulatory moats, you should too.
So, but… Funny story. I just did (most of) this traditionally-high-paid-special-consultant-work in two weeks. Part-time. For $20.
Recently I wrote The Horse Job Cope, where I argued that AGEI (artificial good-enough intelligence) collapses whole business functions, not individual jobs. A single FTE with agents replaces a team.
The reason I felt compelled to write that piece is what I’m going to talk about here. This is the story behind that essay.
I’m the incoming (currently part-time) COO at SmallCorp, a small aerospace and defense firm with two sides to the house: an engineering services consultancy (good unit economics, poor scalability, needs operational freedom) and a nascent capital-P Products division (needs big-boy process maturity and discipline). We need AS9100D compliance for the products side. The budget for a consultant or a dedicated quality hire is, to use a technical term, nada nunca niente.
So I did what any reasonable (foolhardy?) person with an LLM subscription and a tolerance for ambiguity would do. I skimmed a book on surviving AS9100 implementation, opened a chat window, and started asking stupid questions. Worth noting that my last gig, at BigCorp, was in the FDA-regulated space, so I’m more than passingly familiar with Quality Management Systems, having suffered worked under one.
First question to the LLM: “Give me every reason I should, and shouldn’t, DIY this.” The answer was clarifying. The documents are writable. The process architecture is designable. The gotcha review by someone who’s actually been through an audit? That still needs a human. Also, having a framework is one thing and actually walking the talk is a different thing. Good. Now I know where the boundaries are.
Second prompt: “Build me a plan.” The LLM said: “Start with the quality manual and the two core procedures (design controls and configuration management), then build everything else from those pivot points.” Not unreasonable. I studied the LLM’s proposed plan, agreed with that logic, and started.
Then a few days in the cycle: draft, review, revise. I told the LLM what was special about this business. That conversation (a lot of context-loading) mattered. The dual nature of SmallCorp, engineering services and products under one roof, required fundamentally different process weights. The services side demands flexibility, while the products side absolutely requires rigor. Already I could see the beginning of a payoff for “Screw it, DIY”: a canned template from a consultant’s filing cabinet would’ve gotten this wrong. An off-the-shelf QMS would’ve tried to impose uniform process density across both, and the engineering services team would’ve ignored it within a month. (Rightly so.)
That business-specific calibration drove repeated revision cycles. While I lather-rinse-repeated, I learned something important: there’s a goldilocks level of specificity in quality process documentation. Too vague, and the procedures aren’t auditable, because there’s nothing concrete for an auditor to verify you’re actually doing. Too specific, and your QMS runs the show instead of reflecting it. Updates become a tax. Projects with different needs are denied the flexibility that makes them work. Every procedure ended up going through several rounds of tightening and loosening before it was goldilocks juuuust right.
One by one, I built all 16 procedures. Each one built from the core two as structural pivot points, so the system has internal cohesion. Along the way, I kept asking the LLM to review the growing body of documents as a set: check for auditability and internal consistency across the whole system. I had it run web searches for specific guidance on tricky clauses. I cross-referenced against the book, the standard, and regulatory guidance.
What did I get for my prompting labor? A quality policy, a quality manual, 16 procedures, and roughly two dozen forms, registers, and regularized evidence documents. An Approved Supplier List. A full document control framework. The whole spine.
Two weeks. Part-time. No quality background.
Of course we still need to “Do what you say,” and also, we’ll bring in a consultant for a gap review before we go for certification. Maybe $10-15K. That’s the part that still needs a human with audit experience, someone who’s seen where registrars actually dig and where companies actually stumble. I’m not pretending I replaced that.
But I did replace the other 90%.
You might be nodding your head. “Of course an LLM can help with a QMS. AS9100 is a published standard with enumerated requirements. It’s basically a template job. The rules are written down. You just systematized what was already systematic.”
Exactly. But like the old commercial for Ginsu knives: Wait! There’s more.
SmallCorp is a 28-person company that has been running on tribal oral tradition. High trust, zero hierarchy, zero process. It worked “ok” at 15 people. It’s actively breaking at 28 and will absolutely be fatal at 50. SmallCorp needed a Team Operating System: a leveling framework, competency matrices for every engineering discipline, a compensation architecture with pay bands, a structured interview methodology, a performance management system, promotion criteria, AS9100 Clause 7.2 compliance wiring, change management and rollout sequencing. Basically the whole enchilada of people infrastructure that a company needs before it can scale.
The obvious move was to hire a fractional CHRO or a people ops consultant who understands aerospace, hardware manufacturing, and the specific weirdness of a company transitioning from boutique services to production. I tried. For weeks. I looked, trust me, and it turns out: that person doesn’t exist. The intersection of aerospace domain knowledge, hardware/manufacturing experience, and willingness to do a bounded 90-day engagement is, as I wrote in the document itself, “a null set.”
So silly me, I just built it. Same process as the QMS. Interrogate the LLM about what a team operating system needs to contain. Study the plan. Draft, review, revise. Feed it the specific context of the business. Iterate until the document reflects reality instead of imposing a generic template on it.
The result is a comprehensive Team Operating System. Job architecture with career tracks. Universal level definitions from IC1 through Principal and M3 through Director. Core competencies (ownership, quality mindset, collaboration, communication, technological leverage, safety & stewardship) with behavioral indicators at every proficiency level. Functional competencies for firmware, electrical, and mechanical engineering, each with sub-competencies and proficiency scales. A full manufacturing competency spine covering production operators, manufacturing engineers, and quality engineers, with explicit distinction between the proficiency models because manufacturing competence is not engineering competence. Manager competencies. Product maturity stages mapped to talent expectations, because a Senior Engineer in prototyping operates under different process demands than a Senior Engineer in full-rate production. Role-to-artifact mapping for audit evidence. Compensation bands benchmarked against market data for every role family. A performance management system with scoring methodology, calibration protocol, and a script for the actual review conversation. Promotion criteria. A structured interview process with behavioral methodology. AS9100 Clause 7.2 compliance documentation woven throughout. A change management plan that sequences the rollout so you don’t trigger a mutiny. An appendix that names the failure modes the system will hit first.
This is not a template. Nothing remotely like this can be had “off the shelf”. It’s organizational design for a specific company at a specific inflection point, and it required judgment about dozens of things an LLM can’t know on its own: how much process the engineering services side can tolerate before the good people leave, where the real compliance risks live versus where the documentation theater is, which competencies actually matter for safety-critical avionics versus which ones are HR kayfabe, what “good enough” looks like for a 28-person company that needs to be audit-ready without pretending to be Lockheed Martin.
I directed. The LLM executed. One person. Part-time. No fractional CHRO consultant.
Here’s what should keep you up at night: nobody got fired, but much worse: nobody got hired.
There was no QMS consultant on payroll to let go. No quality manager who got a RIF notice. No fractional CHRO who lost a client. The $150,000 quality engagement never happened. The 90-day, $150,000 CHRO contract never happened. The six-month timelines never even started.
These jobs didn’t disappear. They never got created.
This is the version of The Horse Job Cope that doesn’t make LinkedIn. Nate gets a RIF and posts “Excited for the next chapter!” and the comments fill with performative sympathy. That’s visible. That’s countable. A Bureau of Labor Statistics analyst can track it.
But when a startup that would have hired a quality manager just… doesn’t? When the consulting engagement that would have been $150K just… isn’t? When the fractional CHRO engagement just… never materializes? When the functions that required specialists collapse into one person’s two weeks of part-time work before the specialists were ever assembled?
Nothing imploded. No headline. No sad LinkedIn post. No comment thread assuring anyone the future is bright. Just an absence where employment would have been, invisible and uncountable, repeated across thousands of companies making the same discovery right now.
This is the ghost org chart.
When I looked at the finished work, both stacks of documents sitting there ready to go, I felt three things.
First: “Damn, this is awesome.” The giddy, almost dumb thrill of having made something real for about 1% of what it would have cost otherwise. Forty-plus aerospace-grade quality documents and a comprehensive people infrastructure for an actual company. That’s satisfying in the way building things is always satisfying. Plus: OpEx win.
Second: “Holy shnikies, the white collar myth is over.” I wrote The Horse Job Cope as an argument. I believed it. But there’s a difference between believing your thesis and watching it play out on your own desk in real time. Quality management consulting is a legitimate profession. People operations is a legitimate profession. Both are staffed by competent people who spent years learning their domains. I just did a supermajority of both deliverables in a month without their training.
Here’s the thing: QMS is legible. Published standard, enumerated requirements, published guidance. You can dismiss that. “Basically a template job.” But the Team Operating System? That’s organizational design. Compensation architecture. Behavioral interview methodology. Change management. The QMS is the easy case. The TOS is the hard case. And the hard case fell too.
Third, and this one had some wryness in it: “At least I’m equipped to make this transition. Most people aren’t.” I’ve been building shit for decades: teams, products, companies. I know how to decompose a problem, interrogate an LLM’s output, calibrate specificity, question premises, see when something’s wrong, backtrack and pivot, etc. Both builds worked because I brought decades of hard-won judgment to the process. The goldilocks specificity didn’t come from the LLM. The business-specific calibration didn’t come from the LLM. The decision to structure two operational cultures under one quality umbrella didn’t come from the LLM. The insight that manufacturing competence is fundamentally different from engineering competence didn’t come from the LLM. I directed. The LLM executed.
But “I directed, the LLM executed” is exactly the ratio that Horse Job Cope predicted. One person with agents replacing a team. The many are gone. I’m the one person. For now.
So what happens to the professionals, the “knowledge workers”, the whole white collar labor pool?
AS9100 is a published standard with enumerated requirements. “Say what you do, do what you say, produce evidence that you do.” That’s the entire punchline of quality management. (Yes, of course, I’m oversimplifying. Deal with it.)
It’s also, if you think about it for more than a few seconds, a precise definition of the work that LLMs eat. Any job reducible to “say what you do, do what you say, produce evidence that you do” is a job whose rules can be expressed. Expressible rules are legible. And legibility is a death warrant where Judge Dredd is an LLM.
Remember the Bobs in Office Space? “What would you say… you do here?” That question was a legibility test. If you could answer it clearly, you were cooked. The Bobs just didn’t have the technology to act on it at scale. But hello 2026, now we do. The LLM is the Bobs, running that interview on every expressible function simultaneously, and it doesn’t even need the Hawaiian shirt.
And yes, there are still gaps in legibility, which still require a human. The certification audit still needs a registrar. The ongoing internal audits still need judgment about what’s actually happening on the shop floor versus what the documents say. The difficult conversations with employees about level assignments still need a human in the room. Those are residual work: contextual, embodied, dependent on having been in the room when things went wrong before.
The 90% in the middle? That’s a commodity now. The consultants who survive will be the ones who reposition around the 10% that still requires their hard-won judgment. The rest are horse breeders in 1908.
In Office Space, Peter Gibbons survives the Bobs by cleaning a fish at his desk. It’s utterly illegible. You can’t reduce “guy filleting a bass during a staff meeting” to “say what you do, do what you say.” The Bobs can’t even categorize it, let alone replace it. If you want to survive the next round of Bobs, maybe figure out what your fish is.
(I’m re-reading that paragraph and thinking “What in the actual fuck?” I can’t believe we’re here. So weird.)
Is there a bright side? If there is, it’s this. SmallCorp can now compete in aerospace. A small firm, early on the products side, with no quality department, no HR department, and no six-figure consulting budget, now has both a robust QMS and the people infrastructure to scale. The hurdles to entry that keep small companies out of regulated markets just got shorter. Pied Piper vs. Hooli.
That might be the best thing in this whole story.
Just as in medtech, pharma and med device, every year, small defense and aerospace companies expire in the gap between “we have a product thing and it kinda works” and “big boy company” The compliance overhead is a moat that protects incumbents. (The primes really love this.) If LLMs shrink that moat, more small companies survive to compete and more innovation reaches the market. The industrial base gets broader instead of narrower. More competition. I believe that’s a good thing.
I care about this because I am a fan of SmallCorp, obvs. But I also type 2 care about it because the defense industrial base is super consolidated and everyone knows it. Anything that lowers the barrier for small, capable firms to participate is a national security win, a commercial aerospace win. That’s not nothing.
So yeah. I vaporized a quarter million dollars of professional services in a couple weeks of part-time work and I have mixed feelings.
A year ago, the limit of machine competence was on the near side of “build an aerospace-grade quality management system” and “design a company’s entire people infrastructure.” Now it’s on the far side of both. Nothing dramatic happened, and there was no RIF and no headline. Just me, a guy with a chat window and a tolerance for ambiguity, doing in a couple weeks what used to require a team of specialists.
No fanfare. But yeesh. The implications.
The Horse Job Cope is the belief that new jobs will emerge fast enough to replace the ones that vanish. But the jobs that vanish loudly are the small problem. The big problem is the jobs that never appear at all.
The ghost org chart grows.
Colin Steele is part-timer COO at SmallCorp and writes about coordination failures, automation, and the cognitive economy at colinsteele.org. He does not have a quality management or HR background. He does have an LLM subscription.