Will AI Hurt or Improve Communication in Relationships?
A Couples Therapist Reflects
I am a therapist with 15 years of experience providing couples therapy to couples in Silicon Valley. In engineering terms, I’m an expert at human-to-human user interface: communication. Because I’ve been centered in Silicon Valley my whole career, I’m also familiar with how one member of a couple being an engineer affects their communication patterns, and how those inputs shape the outputs they receive from their partner. So, while still knowing almost nothing about engineering, I think I’m well-positioned to offer some insights on how the AI revolution may or may not affect human-to-human interactions.
Let’s start by identifying common fears.
The Fears: Will AI and Communication in Relationships Suffer?
The fear about AI is that it will harm our ability to connect with each other[i],[ii]. I think this fear comes from some other specific fears:
- A fear that sycophantic AI agents will erode people’s ability to tolerate the real needs and annoyances of other people.
- A fear that we’ll be so focused on communicating with bots that we won’t seek human communication.
- A fear that using AI will affect our problem-solving[i] or critical thinking skills.
The Opportunity: How AI Prompting May Improve Communication in Relationships
I’ve been using AI recently for website tech support, blog post editing to maximize SEO and AEO, and small-business strategizing. As I used it more, I became less convinced of the validity of the fears listed above. I was encouraged that there are some great opportunities embedded in the AI revolution.
Nate B. Jones has a YouTube video/Substack post describing a Four Disciplines Framework for how AI prompting is evolving. He suggests it actually mirrors good human communication and might inadvertently teach you good communication skills as you use it. I agree. The positive effects on communication may not be limited to the business application Jones describes. They may also apply to relationships.
The Four Disciplines Jones describes are: Prompt Craft, Context Engineering, Intent Engineering, and Specification Engineering. I’ll review each and describe what I think they will teach AI users about human communication.
How Could Prompt Craft Improve Communication Skills?
Jones describes prompt craft as including the following key elements: clear instructions, examples and counterexamples, guardrails, an explicit output format (e.g., what file format do you want this in), and ambiguity resolution.
This sounds to me like really good consent communication. No, I don’t mean just sexual consent. Let me explain. A common problem I see in couples comes from the fact that they love each other very much. But they help each other without consent. A classic example is the empathy-versus-solutions response when your partner is complaining about a problem. It’s only a problem if they didn’t want what you offered. And you gave them what they didn’t want because you didn’t ask. They didn’t prompt you correctly to generate the desired output. Consent begins with knowing what you want and asking for it.
So, when you’re telling your partner what you want, you need to become skillful at prompt engineering. You need to tell them in unambiguous terms how to do what you want, give examples of what to and not to do, explain how they’ll know if they’re doing the wrong thing, specify what form to deliver in, and explain how to check in with you if they aren’t sure.
Human Prompt Engineering Example
Please clean the kitchen thoroughly. That’s the dishes, counters, stovetop, kitchen table, the floors, and taking out the trash and recycling. I don’t expect you to wipe out the fridge. Please do check the cabinet door and drawer faces to make sure there’s no food on them, but otherwise, leave them alone, too. When you’re done, look over the kitchen. If you see any messes, anything out of place, or anything obviously dirty, please clean that too. I need this done before Sunday at 6pm. If you aren’t sure whether something needs to be done, how to do it, or where something goes, please text me and ask.
Let’s put all three of these things together so you can see it all at once:
| Jones’ Description of Prompt Craft | My Description of this Skill in Human Communication | Human Prompt Craft Example |
| Clear instructions | Tell them in unambiguous terms how to do what you want | Please clean the kitchen thoroughly. |
| Examples and Counterexamples | Examples of what to do and what not to do | That’s the dishes, counters, stovetop, kitchen table, the floors, and taking out the trash and recycling. I don’t expect you to wipe out the fridge. Please do check the cabinet door and drawers faces to make sure there’s no food on them, but otherwise, leave them alone, too. |
| Guardrails | How they’ll know if they’re doing the wrong thing | When you’re done, look over the kitchen. If you see any messes, anything out of place, or otherwise obviously dirty, please clean that too. |
| Explicit Output Format | What form to deliver in | I need this done before Sunday at 6pm. |
| Ambiguity Resolution | How to check in with you if they aren’t sure | If you aren’t sure whether something needs to be done, how to do it, or where something goes, please text me and ask. |
How Could Context Engineering Improve Communication Skills?
In Jones’ framework, Context Engineering is a way of managing tokens, the currency of processing power in AI engineering, to get the job done by the AI agent as efficiently as possible. The sub-skills of context engineering he identifies include system prompts, tool definitions, retrieved documents, message history, and memory systems.
In human terms, those translate to the how and the why of the task, which tools to use (and where they are), guides, and past experiences that can be referenced to understand the request or how to complete it.
Human Context Engineering Example:
It is really important to me that the kitchen is clean by 6pm because I’m having friends over that night. This is the first time these people are coming over, and I want to make a good first impression. Because I’ll be volunteering with them all day Sunday, I won’t be able to do any of this cleaning, so I need you to do it for me. The dish soap and dish rag are at the sink. The mop is in the hall closet. The scrubbing powder, if there are any stains on the counter, is under the kitchen sink. New tablecloth and fabric napkins are in the drawer under the kitchen utensils drawer. If you need help remembering where things go in the cupboard, remember that there are labels in each cabinet about what goes there. Thank you, you know from past conversations how important acts of service are to me. You did such a good job the last time I asked, and it made me feel really loved. [This references the partner’s memory that acts of service are a love language. Specifically, the most important love language for the speaker.]
| Jones’ Description of Context Engineering | My Description of this Skill in Human Communication | Human Prompt Craft Example |
| System Prompts | The how and the why of the task | It is really important to me that the kitchen is clean by 6pm because I’m having friends over that night. Because I’ll be volunteering with them all day Sunday, I won’t be able to do any of this cleaning, so I need you to do it for me. |
| Tool Definitions | Which tools to use (and where they are) | The dish soap and dish rag are at the sink. The mop is in the hall closet. The scrubbing powder, if there are any stains on the counter, is under the kitchen sink. New tablecloth and fabric napkins are in the drawer under the kitchen utensils drawer. |
| Retrieved Documents | Guides | If you need help remembering where things go in the cupboard, remember that there are labels in each cabinet about what goes there. |
| Message history/Memory Systems | Past experiences that can be referenced to understand the request or how to complete it | You did such a good job the last time I asked and it made me feel really loved. |
How Does Intent Engineering Improve Human Communication?
Intent Engineering, according to Jones, is “The practice of encoding organizational purpose, goals, values, trade-off hierarchies, and decision boundaries into agent infrastructure.” It “tells agents what to want.” This is how you get the AI to do the job in a way that supports the organizational objectives, not just fulfill the task. In one example, Jones describes how failed intent engineering leads to lower customer call times but higher customer dissatisfaction.
In human relationships, failures at the intent-engineering level lead to deep trust issues. In my experience, more likely, nothing was actually communicated. Humans fail each other at the intent engineering level frequently in the following ways:
Assuming we have the same intent.
The first problem is a common form of self-oriented thinking. It is neither selfish nor narcissistic. It just doesn’t grasp at a deep level that other people are fundamentally different from me. It is an error to assume that they value, feel, or perceive the same things I do. Children develop the theory of mind[iii], the capacity to know that other people experience things differently, at a young age. Some research suggests that capacity continues to develop into adulthood[iv]. Indeed, in my practice, many adults have difficulty remembering that their partners feel, value, and experience things differently inside.
“Any decent person would want the same thing,” and “I think it’s reasonable to expect these things” are both examples of this kind of error. This leads people to assume that they don’t need to, or shouldn’t need to, explain why they are making the requests they are making.
Believing you should know my intent, and if you don’t, that’s your failure, not mine.
People in long-term relationships often make this error. As the song goes, “If you don’t know me by now, you will never ever know me.” (Harold Melvin & The Blue Notes, 1972, and covered by Simply Red in 1989). But what people fail to realize is that human intent engineering needs regular updates. As long-term couples age, they need to take the time to recalibrate their mental models of each other. So, yes, you need to tell him that, “Yes, I used to like fried chicken, but now I prefer grilled chicken.”
This is because of the reminiscence bump[v]. That is, the fact that our memories are better for early developmental experiences. In addition, we pay much closer attention when falling in love. So, it is a lot harder to remember something new your partner says than something they told you when you were together in college or when you first got together.
Regular updates are also required because your partner is growing and changing, too. Their internal values, feelings, needs, and emotional landscape are evolving. So how they interpret you changes over time, even if you stay the same.
So, the reality of human communication is that in order to communicate well with our partner, we have to know ourselves well. This allows you to provide complex requests that are resilient to the unexpected.
Intent Engineering Examples:
“Please get red roses at the flower shop to have on the table tonight for dinner.” Is different from, “Please get red roses, but if they don’t have that, any red flower will do. I want to match the napkins.” Is different from “Please get red roses, but if they don’t have red, get white, or yellow in that order of preference, because roses are mom’s favorite. If they don’t have roses, don’t get any.”
Let’s see what intent engineering can add to our example about cleaning the kitchen.
I’m having these people over because I want to deepen my friendships with them. I want to make a good first impression. I appreciate situations like this where we can function as a team. Oh, if you’re running short of time, sweeping the floor but not mopping it could be okay. And if you feel like it is getting close to 6pm and you could start something new, but might not get it done in time, then it’s better for me that you not start it. Cleaning chaos while people arrive is counter to the impression of a neat and orderly team that I want to give.
| Jones’ Description of Intent Engineering | My Description of this Skill in Human Communication | Human Prompt Craft Example |
| Purpose | At this depth, we don’t need a human translation of the communication process. This is just good communication, and we use very similar terms for communicating with an LLM as with a human. | I’m having these people over because I want to deepen my friendships with them. |
| Goals | I want to make a good first impression. | |
| Values | I appreciate situations like this where we can function as a team. | |
| Trade-off hierarchies | Oh, if you’re running short of time, sweeping the floor but not mopping it could be okay. | |
| Decision boundaries | And if you feel like it is getting close to 6pm and you could start something new, but might not get it done in time, then it’s better for me that you not start it. Cleaning chaos while people arrive is counter to the impression of a neat and orderly team that I want to give. |
How Does Specification Engineering Improve Human Communication?
In Jones’ model of AI engineering, specification engineering provides a larger scaffolding for all work done by agents for the company. It allows smaller pieces of AI work to fit into larger production contexts (AI or human) seamlessly, even if different AI agents, or different people, pick them up.
If you’re lost, don’t worry. This is about to make so much sense when we bring it down to the human household level.
One of the big problems of a modern family, the reason that this time feels so stressful, is that our culture is the most complex in human history. Let’s just take one tiny slice, and you’ll see it really easily. How many different types of insurance do you have?
- Health insurance
- Dental insurance
- Eye insurance
- Car insurance
- Home/Apartment insurance
- Life insurance
- Long-term/Disability care insurance
- Business liability insurance or professional insurance
- Pet insurance
In addition to this, you may have a variety of other protection services you pay for as add-ons to other services that are essentially additional kinds of insurance, like car warranty, computer care plan, antivirus plan, and credit protection services. The list goes on. Running a household, even a household for one, requires the skills of running a business. I didn’t say “small business” on purpose.
Specification Engineering is the solution to the stress caused by all of this complexity, and it allows a household to run smoothly. Let me give you an example you’ll be able to read more about if you want to: Family Meetings. A good family meeting doesn’t just process some information for an hour once a week. A good family meeting coordinates each family member’s capacities with the family’s needs, so that everything flows together. A family meeting helps maintain an existing routine when disruptions occur.
Household Specification Engineering Example:
- A family meeting decides what the meal plan is, so that
- When grocery shopping happens, the shoppers know what to buy, so that
- The cook has the food they need on the days they need it, so that
- Everyone has the food they need to do their homework, childcare, chores, and play.
A good family meeting, weekly routine, and chore assignments are the specification engineering of a well-run household. Even if someone else has to take over dinner one night, the meal plan said what dinner was, the grocery shopper already got the food, and the recipe box or the cookbooks describe how to make it.
If you’re thinking that’s a lot of work, you’re right. But not doing that work is a lot of work and creates a lot of stress, too.
That’s why, sometimes, the best couples therapy interventions I do are to help a couple have good family meetings, have family values conversations, and create a good family routine. Sometimes, relationship problems result from household chaos. Yes, that chaos is often caused by an impossibly challenging world. It isn’t the couple’s fault.
Conclusion: Are There Positive Consequences of AI on Human Communication?
To summarize, based on what I’ve seen, I’m actually excited about some of the ways AI will teach us to communicate. Because we know an AI isn’t human and only knows what we tell it, we’ll be forced to develop conscious self-awareness of what we want, how we want it, why we want it, the broader context in which the request sits, and what it serves. Then we’ll become practiced at communicating that awareness. If we let them, those skills will also make us better at communicating with other people.
My Real Fears About AI’s Impact on Relationship Communication
With that said, there are three fears that I still have about AI’s impact on human communication. I’ll present them briefly, along with an opportunity.
Fear 1 : Sycophancy Fear Updated
None of the skills above includes vulnerability. If bots teach us to be vulnerable by opening up to them, will we stop opening up to each other?
Opportunity: In practice, I’ve already seen some of my clients practice opening up with a bot. They use that training ground to refine their communication, and later bring it into therapy or to their partner.
Fear 2: Skills/Class Gap Acceleration
I fear that people who use AI agents and who already have critical thinking and emotional humility will be able to use and transfer these skills to their human interactions. But will people without these skills be trained by the agents to develop them? Or will they be relegated to using bots to confirm their own biases?
Opportunity: AI agents can be trained to help people develop these skills. Companies may even be motivated to make sure the agents do this as much as possible. The more unskillful you are at prompting the agent, the more iterative passes you’ll make trying to complete your tasks. That’s inefficient for your time and tokens, uses provider compute capacity, and decreases your satisfaction with the product. The more an AI product becomes humble and asks you for all these pieces of information before executing, the higher your satisfaction will be.
That might teach us what questions we need to ask of each other. Using our kitchen-cleaning example from earlier, the cleaner could also ask questions to elicit the information that my examples filled in.
Fear 3: Human attention is a valuable resource.
We can’t communicate with each other without human attention. And AI bots are hungry for our attention. It is their purpose to exist. Will our economy free our attention for each other or just devise more ways for our attention to be drained to fuel corporate bottom lines?
Opportunity: Major industrial changes can and should free workers to enjoy more of life. If the AI revolution, which allows workers to accomplish more in less time, also precipitates a reduction in the workweek length and an increase in workers’ time off, then it may earn its keep.
Is AI Going to Be Good for Human Communication?
The answer is ultimately up to us. AI, like other tools, is what we make it. Industrialization created machines that could be dangerous to workers and communities. Unions gave workers the strength to bargain for their safety. Government regulations further proscribe how companies could use machines to make a profit and how much risk to workers, the environment, and people is and isn’t tolerable. That’s going to be no different for AI and communication in relationships.
If you want to work with a therapist who thinks deeply about human communication, has practiced in Silicon Valley for 15+ years, and can use technology-speak to teach human connection, please contact me.
[i] Kennedy, B., Yam, E., Kikuchi, E., Pula, I., & Fuentes, J. (September 17, 2025). How Americans View AI and Its Impact on People and Society. Pew Research Center. https://www.pewresearch.org/science/2025/09/17/how-americans-view-ai-and-its-impact-on-people-and-society/
[ii] McClain, C., Kennedy, B., Gottfried, J., Anderson, M., & Pasquini, G. (April 3, 2025). How the U.S. Public and AI Experts View Artificial Intelligence. Pew Research Center. https://www.pewresearch.org/internet/2025/04/03/views-of-risks-opportunities-and-regulation-of-ai/
[iii] Wellman, H. M. (2011). Developing a theory of mind. In U. Goswami (Ed.), The Wiley-Blackwell Handbook of Childhood Cognitive Development (2nd ed., pp. 258–284). Wiley Blackwell.
[iv] Carlson, S. M., Koenig, M. A., & Harms, M. B. (2013). Theory of mind. Wiley Interdisciplinary Reviews. Cognitive Science, 4(4), 391–402. https://doi.org/10.1002/wcs.1232
[v] Munawar, K., Kuhn, S. K., & Haque, S. (2018). Understanding the reminiscence bump: A systematic review. PloS one, 13(12), e0208595.
AI Use Disclosure: I created all the sections, headings, concepts, examples, and tables in this blog entry unless otherwise noted and cited. I used AI to copy-edit the original human-created manuscript (3,386 words) into a new version (3,491 words) to meet SEO/AIO/GEO/AEO needs (letting the bots serve the bots). Please let me know how you feel about my using AI in this way in the comment boxes below. I read all the comments myself.
