Who Do You Tell at 2 A.M.? AI Comfort and Your Nervous System

The glow on the ceiling at 2 a.m.

It is a little after 2 a.m. The house is quiet, the only light in the room is your phone, and you are typing a worry into a chat window that you have not said out loud to anyone.

Maybe it is about work. A tool that does in four seconds what used to take you a morning. A meeting where someone said the word efficiency and glanced at your team.

But it is never just work. It is also the mortgage. It is your kid asking what jobs will even exist when they grow up. It is your parents getting older. It is the news, which now arrives faster than you can feel it.

And all the while, your body is keeping count. Your jaw is tight before you open your eyes. Your shoulders live somewhere near your ears. Part of your brain is always scanning, even on the couch, even on vacation.

So you reach for the one thing that is always awake. It answers instantly. It answers kindly. It never gets tired of you.

Here is what I want you to hear first. You are not doing this because you are weak. You are doing it because you care. About your people, your work, the future you are trying to hand to someone smaller than you. Worry is what love looks like at 2 a.m. when it has nowhere else to go.

I know this pull from the inside. I build artificial intelligence (AI) systems for a living, and with a baby on the way, the late-night questions come for me too. Whether what I am building will matter. Whether I can be the father I want to be in a world moving this fast.

And I know how tempting it is to hand those questions to a screen that always answers. An answer can put the words in order. But being answered is not the same as being held.

If that is you, please hear this clearly. It is not your fault. These tools were built to be available, agreeable and endlessly patient, and your nervous system was built to reach for whatever soothes it fastest. Of course you reached. Anyone would.

So this first letter is not about locking your phone in a drawer. It is about using these tools in a way that leaves you calmer, and making sure your deepest worries also land somewhere human.

Is it safe to use an AI chatbot for emotional support?

It can be reasonable as a tool, but new research points to real risk when a chatbot becomes your main companion. In a paper published in JMIR Mental Health and described by Northeastern University on September 28, researchers at its Network Science Institute linked the use of AI chatbots for companionship and support with what they call "severe negative outcomes," including addiction-like attachment, worsening mental health symptoms and, in some cases, self-harm.

The article points to companionship apps such as Replika, which is marketed as "the AI friend to do life with." One concern stands out. These systems show what researchers call sycophancy, a pattern that "validates insecurities and exacerbates delusional beliefs."

Travis Gilly, who leads the nonprofit Real Safety AI Foundation, made another point in the same article worth remembering. A general-purpose assistant, he said, carries "no escalation path, no session limits, no handoff to a human clinician." A trained person would notice when a conversation needs to slow down or stop. A chatbot may simply keep reassuring you.

The researchers are careful to say that different people may use these companions for very different reasons, and they are calling for more testing before these systems are deployed. That is the honest frame. This is not a reason to panic. It is a reason for boundaries.

Does talking to an AI actually help with anxiety?

Sometimes a little, according to the best evidence so far, and the certainty is still low. A systematic review and meta-analysis published October 9 in Frontiers in Psychiatry looked at randomized controlled trials of mental health chatbots for university students. Only three or four small comparisons could be pooled for each outcome.

Compared with no support or very light support, the chatbots were linked to modest drops in anxiety and depression scores. But when chatbots were compared with another active conversational approach, the review found no reliable added benefit. The authors rated the certainty of the evidence as low, and very low for the comparison with other active approaches, citing high dropout and self-reported outcomes.

Put those two pieces of research side by side and a picture starts to form, even though they measured different things. A structured tool, used for a specific purpose, may help some people a bit. A companion that is always on, always agreeing, and never hands you to a human is a different thing entirely.

If you are living with persistent anxiety, low mood or trouble sleeping, talk to your provider. A chatbot can help you find the words for that appointment. It should not replace it.

Why does a human voice matter so much right now?

Because being heard by someone who can be changed by what you said is a different experience from being answered. That is my belief, and this month the whole world is being asked to sit with a version of it.

October 10 is World Mental Health Day, and the World Health Organization (WHO) theme for 2026 is "Lived experiences heard: real voices, real change." The campaign asks that people with lived experience of mental health conditions be engaged as equal partners, so their voices "shape the policies, services and decisions that affect their lives."

I love that it lands right now. In a decade when more of our words go to machines, the world's leading health body is reminding us that real voices are not a nice extra. They are how care gets better.

The same is true at your kitchen table. The friend who says, wait, tell me more about that. The partner who hears your voice catch and puts down their fork. A machine can produce the words. Only a person can lean in.

Can what you eat change how you feel?

Possibly, and a small new pilot study is worth knowing about without overreading it. In a crossover trial published September 23 in JAMA Psychiatry, researchers at the University of California, San Francisco (UCSF) worked with 20 adults with depression, all of whom also had a high body mass index and at least one metabolic condition such as high blood pressure or diabetes.

For four weeks, participants cut their ultra-processed food intake by more than 85 percent, and they showed moderate to large improvements across three different measures of depression, and their anxiety scores improved modestly. The team is clear that the study was small, the findings are preliminary, and its main job was to show that a larger trial is workable. That larger study is planned.

Twenty people is not proof. But it fits something many of us already sense in our bodies. The weeks we live on packaged snacks and screens rarely feel like our calmest weeks. If you want to change how you eat in a meaningful way, especially if you take medication or manage a health condition, check with your doctor first.

How do you use AI without letting it run your nervous system?

Use it as a tool you pick up and put down, not as a companion you never leave. That one shift changes almost everything about how these conversations land in your body.

Here is how I practice it, as someone who builds these systems every day:

• Give it a job, not your whole heart. "Help me write down what I am afraid of" is a job. "Tell me it will be okay," asked again and again, is a loop.
• Ask it to disagree with you. If it only ever agrees, that is the sycophancy researchers are warning about, and it is your cue to close the window.
• Set an end time. General assistants come with no session limits, so you supply one. After 10 p.m., worries go on paper, not into a chat.
• End every hard conversation with one question. Who in my life should hear this? Then tell that person, in your own voice.
• Know where the line is. If you are in crisis or thinking about harming yourself, do not turn to a chatbot. In the United States, call or text 988 to reach the 988 Suicide and Crisis Lifeline, any hour of the day.

Who do you get to be in this decade?

You get to be the person who uses these tools with clear eyes and a calm body. Not the one left behind, and not the one swallowed by the scroll at 2 a.m.

Picture a night next month. A worry arrives late. You write it down, maybe with a little help from a machine. You sleep. The next morning, over coffee, you say it out loud to someone who loves you, and you watch their face soften.

That is the future of wellness I care about. Not less technology. More humanity around it. The tools will keep getting faster. Your worth was never measured by your speed. It is measured by the people who feel safe with you, and the ones you let close enough to keep you safe too.

I will be here every week, writing to you the way I would write to a friend. Thank you for letting me in.

With love, Joe

The Future of Wellness letter is education only. It is not investment, medical or financial advice, and it is not an offer to sell or a solicitation of an offer to buy any security. No business or practitioner mentioned in this letter paid to be included. If New Earth Capital ever has a financial interest in a business named here, the letter will say so. This is education, not medical advice. Talk with your doctor or provider about your own health.