LLMs as a Parenting Copilot: Using AI to Translate the World
I want to be upfront about something: I use AI tools daily, both at work and at home. Not as a replacement for judgment, but as an accelerant for the translation work that parenting neurodivergent kids requires constantly. The world doesn't come pre-formatted for how my children process information, and a significant portion of my job as their parent is serving as a real-time adapter between the world's defaults and their needs.
The Translation Problem
Neurodivergent kids often need information delivered in a specific format, at a specific density, with specific framing, before they can engage with it. A teacher's written instructions might be technically clear but formatted in a way that overwhelms (e.g. a paragraph of text when a numbered sequence would land better). A social scenario might make perfect sense to a neurotypical peer but arrive as pure noise to a child who processes social cues differently.
This is translation work. I've been doing it manually for years: re-reading assignment sheets and rewriting them as step-by-step lists, pre-briefing social situations with "here's what will probably happen and here's what you can do," scripting conversation starters before birthday parties. It's constant, it's energy-intensive, and it's the kind of task where LLMs genuinely help.
Where AI Fits
The use cases that have worked for us fall into a few categories
| Use Case | What I Ask the LLM | What I Do With the Output |
|---|---|---|
| Assignment reformatting | "Rewrite this paragraph of instructions as a numbered checklist for a 6-year-old with ADHD" | Review, adjust reading level, add our specific language, then share |
| Social scripting | "Generate 3 ways a kid could join a group conversation at lunch" | Pick the ones that match my child's personality, rehearse together |
| Emotion vocabulary | "Give me 5 words between 'fine' and 'angry' that describe frustration" | Use during debrief conversations when my child is stuck on binary descriptions |
| Routine building | "Draft a morning routine with 8 steps, each under 5 words" | Validate against what actually works, iterate over a few days |
The common thread is that I'm not outsourcing the parenting, I'm outsourcing the first draft of the formatting. Every output gets reviewed, adjusted for our specific context, and often modified significantly before it reaches my kids. The LLM doesn't know my children, I do. It's a tool for generating options quickly, not for making decisions.
Where I Draw the Line
There are places where I deliberately don't use AI in this context.
I don't use it to generate emotional responses. If my kid is upset and I need to figure out what to say, that comes from me, not from a prompt. The warmth and specificity of "I remember last time this happened and you handled it by doing X" is something an LLM cannot produce because it doesn't have the relationship context. And even if it could approximate it, the authenticity matters. Kids (especially neurodivergent kids who are often hyperaware of inauthenticity) can tell when language doesn't come from a real place.
I don't use it as a diagnostic tool. "My child is doing X, what does this mean?" is a question for their therapist, not for a language model trained on internet text. The risk of confirmation bias alone makes this dangerous, and the stakes are too high for pattern-matching without clinical expertise.
I don't hide that I use it. My older kid knows that sometimes I ask "the computer" for ideas about how to explain something. Transparency here is the same principle as transparency everywhere in our household, it reduces volatility and builds trust.
The Copilot Framing
The reason I keep using the word "copilot" is that it maps to exactly how I think about AI in my professional life (not just the Microsoft product, but they did pick a good name for it). At work, I use LLMs to draft communications, generate Ansible playbook skeletons, and brainstorm troubleshooting angles. In every case, I'm the pilot. The AI is handling the mechanical parts (first drafts, formatting, option generation) while I make the judgment calls, provide context, and own the outcomes.
Parenting is the same model. The mechanical translation work (reformatting, vocabulary generation, script options) is a legitimate use of the tool. The judgment, the relationship, the attunement to what this specific child needs in this specific moment, that stays human. That has to stay human.
What I'm Watching
The thing I'm most cautious about is dependency creep. Not my kids' dependency on AI (they barely interact with it directly), but my own. The risk is that I lean on generated content so heavily that I stop doing the effortful thinking that actually builds my understanding of my children's needs over time. If I always ask the LLM for a first draft, do I lose the muscle memory of constructing those translations myself?
I don't have a clean answer for that yet. For now, the pattern is if I could do it myself in under five minutes, I do it myself. If it's a longer formatting task or if I'm genuinely stuck for options, I'll prompt for a starting point and then make it mine. That boundary might need to shift over time, and I'll revisit it when it does.