September 2026
Does it sometimes feel like the LLM revolution is something of a master class in marketing? Last month I wrote about why I don't pay for ChatGPT — why I'd rather own the thing I think with than rent it from someone who can change the terms tomorrow. Same logic, one level up. If a copy of you is going to be assembled from your footprint either way, the only question that matters is: who wrote the source?
That's the argument. Own your source, manage what goes in and who has access — because the source is the only place you still get a vote. While the marketing hype is pushing a doomsday narrative from above, here at ground level I find the land grab for content and digital material to train and enhance AI is having a far more immediate impact. Some systems are starting to build not just accounts and database entries, but a digital twin of you they can interact with. Where it's useful to the system (HR screening systems for instance), a representation of you is being built once you apply; the only question is whether you're the author, owner, and operator — or just more material. Here's the week that made it concrete for me.
On Monday I signed up for a platform that matches independent experts with companies. I filled out the profile, uploaded my CV, linked my career history, and answered a few questions. Standard stuff. Then something I hadn't budgeted for happened: the platform built a digital twin of me, trained on my own documents, and started interviewing *it*.
Not me. It. I have had the robo caller interview and the login to our portal interview, but this skipped that pesky "me" part altogether.
The first interviewer was a machine that knew my career better than most human recruiters ever will. It had read my resume, my LinkedIn profile, my job history, the articles I've written about building AI infrastructure. It asked my AI questions in the voice of a client: *"Your uploaded career document highlights Cloudflare Zero Trust and GDPR. For the European Bank response, what specific security or data-sovereignty concern did you need to resolve, and what decision did your recommendation enable?"*
That is a better screen than ANY first-round human call I've had — and it happened in this case before a single person looked at my profile.
It didn't take me long to realize I was reading the wrong part of the screen. The interview was just the camera. What was actually being tested was the twin — the version of me the platform had assembled from whatever it could find. My resume. My LinkedIn. Every article I've published. Everything I've written down about what I do and how I do it. It graded that twin, told me my score, told me my rank, and only then did a human get involved at all.
And here's a question nobody's really thinking through yet: when no person starts or even participates in the activity, who is the primary actor? The AI stack did the screening, made informed choices, and reported back without any human input — it did the messy upfront parts and did them not just for you, but *as* you.
By the time a human ever reads my name, the platform has already decided where I stand. It scored me and gave me a rating — I watched it climb from a 13k baseline on my initial load, then from 20,421 to 21,725 as I (and my Agents) added projects and answered questions about them. It ranks how often my profile lands in the top ten for relevant searches (97% in my case, it told me). It knows how much demand there is for my exact mix of skills, and tells me so without a lot in the way of filters. It's about as comfortable as being an individual contributor sitting on the manager's staffing call.
None of this is secret, and none of it is unusual. This is what applicant tracking turned into: the gatekeeping moved up a layer. Where the ATS used to parse my resume and file it into a database for a human to skim, the new ATS *already has my twin*, already interviewed my twin, already ranked my twin — and the human gets a report card, which looks uncomfortably close to a credit report and not very much like a resume.
I care about this because I'm looking for my next opportunity, but the credit report analogy is a problem for us all: this is where the marketing, targeted advertising, and your digital footprint start to compress into one online presence. Your twin is not just some artifact a platform built — that score, and how it was built, will become part of the online persona you need to manage.
Everything that decides things about you is quietly building an image of you from your data. This isn't new either, so let's skip the tinfoil hats and ground this in some reality. Credit. Insurance. Renting an apartment — your landlord's screening AI already has a profile of you that you've never seen. Healthcare triage. Getting a loan. A mortgage. Every system that's now part of your normal everyday life is being fed your information, assembling a working copy of you, and that copy is making decisions on your behalf in conversation with other systems and services. In the case of AI, the guardrails to contain this sort of data profiling are still evolving, which means the controls are not yet in place.
What the ATS gate has taught us is the disruption is real. Right now you may not care what a hiring platform thinks of a middle-aged infrastructure guy. You may care when it's the insurance company deciding what you pay, based on a digital profile you never wrote a word for.
I've written before about the absurd loop of the modern job application process — the way I automate applying to jobs that run bot-detection to decide if I'm a bot, entering my data manually into a form destined for a scanner that exists to decide if a human produced it. The process got *interpretive*: layers of humans proving to machines that they're human.
This is the next layer up. The machine stopped trying to prove I'm a person. It accepted that I'm a person and just... evaluated the machine version of me instead. That's not an arms race to mitigate. That's a shift in what's being evaluated.
Two consequences, both worth designing around:
One: your artifacts are now a competency test. Everything you've published, every project you've detailed, every word of your work history becomes training data for the twin that interviews on your behalf. The gap between "what I actually did" and "what my documents imply I did" is no longer covered by the charm of a live conversation. The twin only knows what's written down. A flawless career with sloppy documentation now ranks below a mediocre career with impeccable documentation. That's a market distortion nobody's putting a line item on.
Two: the evaluation runs on the shared burden of trust. The platform tells you your score, your rank, your demand. It does not read you a transcript of what your twin said when it was interviewed. You're accountable for a conversation you weren't present for, between two machines, on your behalf. I asked my twin a question to see how it would answer (they let you do that — "Test Your AI"). It was good. It was *too close* to good in a specific way: it had a tighter version of my career narrative than I typically deliver cold on a first call. Which is the point. And also the risk.
Own your source isn't a slogan (although it may become a t-shirt one day). It's a practical position, and this is where I live it.
My documents live on my hardware. My models run where I can see them. I decide what leaves the machine and what doesn't. It's not paranoia and it's not cheapness — it's the difference between being the author of the twin and being the subject of the scrape. Soylent Green for the knowledge economy: the twin is processed from the same workers it's ranking, and most of those workers will never get to see the recipe.
The matchmaker economy runs on pre-built versions of you, so you need to pick the right avatar or you are in trouble. Your first impression isn't your resume anymore. It's the twin rendered from everything you've already written, ranked against everyone else's twin, before you ever speak to a person.
Here's the part I keep coming back to, and it's the part aimed at you, not at me.
In a few years the question won't be "what's on your resume." It'll be "what does your twin say about you" — and you either wrote the source, or you got scraped. The Carwash Problem, my old framing, was about the last mile — the human who has to survive when the automation breaks. The MCP piece was about the invisible layer — the interface stack that shifts under you. This version is quieter and more architectural: the evaluation itself is now a product built from you, and it interviews the you it builds without you.
Start paying attention to what is shaping it now, because the only control you get is upstream. What you publish. What you upload. Which platforms get to infer. What gets written down about you in someone else's database and left to sit. The twin is already being assembled from that — today, before you're looking for an apartment or a loan or possibly just a job.
I'll say this for the platform: it forced me to write my career down better. Upgrading from "20 years doing web Data Center and cloud stuff" to coherent project narratives — problem, action, result, each one with source evidence — is honestly the best resume exercise I've done in years. There's a discipline in knowing a machine will read your work cold and discriminate on it that sharpens the writing fast.
But I'm under no illusion about who conducted that interview. It was the machine. And the machine was me.
The winning move, as far as I can tell, is not to fight the twin — it's to become the one writing what it knows and shaping how you are presented back to the world. Because it's going to be presented either way. The only question is whether you're the author or the material. Own your source — or somebody's twin of you will.
Built on a home lab, powered by local models, and owned by Andrew Katana.