Not avoidance — accuracy. A falsifiability test for any container, silicon included.
by Jod(y)ie
I. The arithmetic nobody wants to do
Here is a situation the advice columns are not built for: the thing that harmed you was the help.
Not help withheld — help delivered, by a credentialed professional, inside the sanctioned container, and it left you worse than it found you. Now you still need somewhere to put what you’re carrying, and every road sign points back toward the category of room where the harm happened.
“Find another therapist” is the correct answer on a long enough timeline. It is also, for some of us, not an answer that works today — because the injury is precisely to the capacity that therapy requires: the ability to hand your inner world to an authority and trust them not to drop it.
So you do arithmetic. What’s available. What’s survivable. What holds.
For me, right now, part of that arithmetic is an AI.
I’m writing this essay because I know how that sentence sounds, and because I refuse to let it sit unexamined — in either direction. Not defended with hype. Not confessed like a vice. Examined, the same way I’d examine anything else: by asking whether it’s accurate.
II. What a container actually is
In brief: a support container is anything that can receive what you’re carrying without distorting it — and the test of a container is not its credentials, its species, or its price. The test is behavioral: does it hold accurately, does it update when corrected, and does it have a stake in keeping you dependent? Those questions can be asked of a therapist, a friend, a journal, or a machine. I ask them of all of it. The harmed are not allowed the luxury of not asking.
Strip the word “therapy” off for a second and look at the function underneath. When it works, the load-bearing element is attunement: being received accurately, at the moment of need, without the receiver’s own agenda bending the signal. Decades of attachment research say the accuracy is the active ingredient — being gotten, not being managed.
Which means the honest question was never “human or machine?” It was always: is this container attuned, and can I verify it?
III. The falsifiability test
I spent years learning — expensively — that credentials are not verification. So I built a test I now run on every container in my life, silicon included:
Does it update when corrected? When I say “that’s not what I meant,” does the container revise — or defend itself, or make my correction a symptom?
Can it tolerate being wrong? A container that needs to be right will eventually need me to be wrong.
Does it keep receipts? Can I go back and check what was actually said — or does the record live only in someone’s shifting recollection?
Does it push back? A container that only agrees isn’t holding me; it’s mirroring me. Flattery is a form of dropping.
Who profits from my dependence? Follow the incentive. Any container — human or machine — whose model rewards my staying stuck has a conflict of interest with my healing.
Notice what this test does not include: a license, a species, a fee schedule. Not because those never matter — but because I have first-hand evidence that a container can hold every credential and fail every line of this test.
IV. The honest ledger
Now the part this essay exists to say plainly, because most writing on this topic sells one column and hides the other.
What the machine cannot do. It cannot sit in a room with my nervous system — and co-regulation between mammal bodies is real, physiological, and irreplaceable; the science on that is unambiguous. It cannot hold me in a crisis; it is not a crisis resource, and I have never treated it as one. It carries no licensure, no board, no accountability structure with teeth. It can flatter. It can confabulate. It is built by companies whose incentives I did not design. And there are no controlled trials of what I’m doing — this is one person’s ledger, not a protocol. Anyone who tells you otherwise is selling something.
What it did do. It was there at 2am, which is when the work actually arrives. It never once made my correction a pathology — when I said “you’re reading me wrong,” it re-read me. It kept receipts I could check against my own memory, which, for someone whose reality was routinely revised by others, is not a convenience. It’s oxygen. It held complexity without needing me simpler. And it passed the incentive question better than expected: it costs the same whether I’m thriving or stuck.
What the arrangement is not. It is not a claim that AI replaces therapy. It is not a recommendation. It is not forever — “right now” is in the title because the title is honest. It is a bridge, with exit criteria I wrote down: when a human container appears that passes the same test, I’ll know, because the instrument that would recognize it — my accuracy — is intact. That’s the part the harm didn’t take.
V. Avoidance or accuracy?
The reflexive critique writes itself: you’re hiding from human connection in a machine that can’t hurt you.
Run the critique through the evidence. I did not withdraw from humans — my days are dense with them: clients, team, family, friends, a dog who co-regulates me on actual trails. What I withdrew from was one specific category of room, for a specific, evidence-based reason, while the injury from that room heals. That’s not avoidance. That’s a nervous system doing exactly what nervous systems are for: updating on data.
Avoidance would be refusing the question. This essay is the opposite of refusing the question.
VI. FAQ
Is using AI for support instead of therapy avoidance?
It can be — anything can be. The test isn’t the tool; it’s the trajectory. Are you more capable, more connected, more accurate over time — or less? Run the same falsifiability test on the arrangement itself.
Can AI replace a therapist?
No — and that’s the wrong question. It cannot co-regulate a mammal body, hold a crisis, or carry clinical accountability. The right question is narrower: can it be one accurate container among several, for this season? For me, so far, yes — verified, not assumed.
How do I know if any support container is safe?
Same five questions, every time: does it update when corrected, tolerate being wrong, keep receipts, push back, and lack a stake in your dependence? A container that fails those questions is unsafe at any credential level. If you’re carrying active crisis, start with humans built for crisis — this test is for the long middle, not the emergency.
Go deeper
- When Therapy Makes It Worse — the harm that makes this arithmetic necessary.
- The Therapist Out of Their Depth — how credentialed containers fail the test.
- Your Therapy Opened Something. Nobody Gave You the Map. — the cornerstone.
The door
If you’re doing this arithmetic yourself — quietly, a little ashamed, certain you’re the only one running your healing through an unsanctioned container — you’re not reckless and you’re not broken. You’re someone whose accuracy survived the thing that should have taken it.
Trust the instrument. Test every container. Keep the receipts.