AI for Humanity · Northline Advisory
Communications Support Agent
Help understanding others, without losing your own voice.
People do not all interpret words, intentions or social expectations in the same way. A message that seems clear to one person may be ambiguous to another; a direct response may be misunderstood even when no offence was intended. Communication is a shared responsibility, not a test of whether someone behaves conventionally.
For autistic and other neurodivergent people, these differences can be particularly consequential in settings built around unspoken expectations. But the need for clearer understanding extends beyond any diagnosis.
Northline's Communications Support Agent is being developed to help people explore unclear messages, distinguish facts from assumptions, consider plausible alternative interpretations and prepare for difficult conversations. It is intended to support a person's own judgment and communication style, not prescribe how they should think, speak or respond.
The approach began with the founder's lived experience and draws on research into communication differences, accessibility and person-centred support. Through a guided calibration interview, individuals are intended to choose, correct and approve the kinds of assistance they want before using personalized instructions with a compatible AI service. The resource is in development, not a diagnostic or clinical tool, and its effects have not yet been established through broader evaluation.
Four ways your agent can help you communicate
- Understand others. Explore possible meanings of an email, text or conversation while keeping the difference between what someone said and what you think they intended clear.
- Recognize potential blind spots. Identify assumptions, overlooked context or ways your own words might be interpreted differently than you intended. Misunderstandings can happen on either side.
- Reflect before responding. Separate evidence from interpretation, consider reasonable alternatives and clarify your purpose before replying—particularly in a sensitive or upsetting exchange.
- Communicate your intentions. Prepare messages and conversations, ask for clarification or work through a misunderstanding while preserving your natural voice.
You decide when to ask for support, what you share and whether to use any suggestion. Your agent applies the communication strategies you chose during calibration.
The same message. Different people. Different support.
Imagine your manager sends you this message:
“Can we talk tomorrow about what happened in the meeting?”
Illustrative agent response
You may be reading this as criticism, but what do we actually know? Your manager wants to talk about the meeting. The message doesn't establish dissatisfaction or blame. Is there other context supporting that interpretation? If not, we could prepare a clarification rather than respond defensively.
Illustrative responses showing different communication strategies. This is not a live agent or a validated assessment.
The message hasn't changed. What changes is the support the person wants. Northline's research-informed calibration method helps you choose which approaches your agent should use and when, rather than applying the same approach to everyone.
See how your personalized agent could help
Suppose you want your AI assistant to acknowledge uncertainty rather than guess what others are thinking, and you prefer direct wording. You could paste an unclear message and ask what it might mean.
Illustrative everyday interaction—not a live service or a real participant transcript.
You
My manager wrote, ‘We need to talk about how things have been going.’ Does that mean I'm in trouble?
Your personalized AI assistant
There isn't enough information to know. It might be an ordinary check-in, feedback, or a concern. Because you asked me to separate facts from assumptions, I wouldn't treat any one explanation as certain. Would you like help asking what the meeting will cover?
You
Yes, but I want to sound direct, not confrontational.
Your personalized AI assistant
You could write: ‘I'm happy to meet. Could you let me know what you'd like to discuss so I can come prepared?’ This asks for clarity without assuming the purpose of the meeting.
This example shows both sides of communication. The agent does not assume what the manager intended. It helps the person separate facts from inference and prepare a direct reply that seeks clarity without creating unnecessary confrontation. Someone else might prefer a different approach—that is why calibration matters.
Everyone is different. And so is the support they need.
There is no single communication strategy that works for everyone.
Some people want help examining an assumption before responding to an upsetting message. Others want several possible interpretations, a short clarification question or help understanding how their own words could be received.
The same person may need different support in different circumstances. A routine text, a difficult workplace conversation and a disagreement with someone close can call for different approaches.
Your agent should adapt to your needs—not expect you to adapt to it.
Northline is developing a personalized, research-informed method of calibrating the Communications Support Agent. The guided interview helps you identify strategies that work for you, the circumstances in which you want them and how actively the agent should intervene. You are free to correct or reject its proposed interpretations and instructions.
How we personalize and calibrate your agent
The guided, research-informed calibration interview explores the kinds of communication challenges you experience, the support you find helpful and when you want particular strategies applied.
You answer a combination of practical questions and situations. An AI guide then explains what it believes you mean; you can confirm, change, correct or reject it. A choice is not automatically turned into a rule. You review and approve the actual instructions that will guide your agent.
Illustrative calibration questions
These examples are illustrative, not the complete or verbatim interview.
- When someone sends an unclear message, what would help you most? Several possible interpretations, distinguishing facts from assumptions, or help asking for clarification?
- When a message upsets you, would you want your agent to question your first interpretation? Only when you ask, when evidence is weak, or not at all?
- Before sending an important message, should your agent point out ways your words might be understood differently than you intended? Should it explain the issue, suggest optional wording or simply flag it?
- When a conversation becomes difficult, what helps you communicate more effectively? Clarifying your purpose, considering the other person's perspective, examining assumptions or preparing a response in your own voice?
See how the calibration interview would interact with you
Illustrative personalized calibration interview—fictional, not a live application, verbatim research instrument or validated assessment.
AI guide
When a message upsets you, would you like your agent to question your first interpretation?
Participant
Yes. I sometimes assume the other person is criticizing me.
AI guide
I understand that you want your agent to challenge that assumption whenever a message feels critical. Is that right?
Participant
Not every time. Only when I ask for help with a difficult message.
AI guide
So when you ask for help with a difficult exchange, you'd like the agent to distinguish what was actually said from what you inferred, consider other explanations and tell you directly when the evidence doesn't support your conclusion. For ordinary messages, it would not intervene unless asked. Is that correct?
Participant
Yes. I want the challenge to be direct, but not dismissive.
AI guide
Proposed instruction: ‘When I ask for help with a difficult interpersonal message, separate facts from assumptions. If my interpretation is unsupported, explain why and offer plausible alternatives without dismissing my feelings. Do not automatically challenge routine messages.’ Approve, edit or reject?
Participant
Approve.
The person can qualify their preference by situation. Only the reviewed, explicitly approved instruction is intended to guide the personalized agent.
The diagram below illustrates the planned conversational pattern. The equivalent interaction is provided in the HTML transcript above.
An AI interpretation is not automatically an instruction. This diagram shows the separate review and approval needed before a proposed preference becomes part of the participant's configuration.
The interview, reflection process, scenarios and proposed output are in development. These examples describe intended functionality, not a working public service or evidence of proven outcomes.
Using your personalized agent
When you finish calibration, you will be able to review the communication strategies you have chosen and when you want your agent to apply them. The planned outputs are a readable description of your approved support, personalized instructions for a compatible AI service and setup guidance.
You then use the configured assistant through that service whenever you want help communicating with another person. Your agent could distinguish a fact from an assumption, offer alternative explanations, help you consider how your words may be received or suggest a reply while preserving your voice. Which strategies it uses depends on your approved calibration, not a fixed approach for everyone.
Northline is not developing a new trained AI model or hosting your ongoing agent conversations. Compatibility, configuration options and behaviour vary across providers. The resource does not automatically connect to every provider.
Who this is for
Designed for adults aged 18 and older who would like more control over how an AI assistant supports their communication with others. The resource may be particularly relevant to individuals on the autism spectrum and others who benefit from different communication strategies. A diagnosis is not required. The goal is not to impose a conventional communication style.
It is not an autism test, diagnostic assessment, psychometric instrument, therapy or a measure of whether someone communicates “normally.”
Why I developed this approach
This began with a narrower problem: I wanted my own AI assistant to work better with me.
I am on the autism spectrum. That is part of my experience, but it did not tell my assistant how direct I wanted it to be, when I wanted it to challenge me, or how I wanted uncertainty handled.
Over time I repeatedly corrected how the assistant responded. I did not want it agreeing too easily, smoothing away my voice, or treating uncertain interpretations as facts. I wanted it to explain how my words might be received when that mattered, without assuming it always knew what another person meant.
I didn't want my assistant simply to agree with my first interpretation. I wanted it to distinguish what was known from what I assumed, consider how someone else might understand my words and challenge unsupported conclusions when I wanted that support. The calibration method grew from making those behaviours explicit and reviewable.
Those corrections became useful working rules, but they were scattered through conversations and accumulated instructions. I wanted a way to make them explicit and reviewable.
That led to the research question: could those preferences be understood, checked and translated into instructions that someone else could approve for themselves?
An initial design case suggested that repeated informal corrections could be recovered through a more structured approach. That is not evidence that the resource works for everyone.
That experience identified the problem and prompted development of a structured approach that other people could tailor to themselves. People should be able to understand and control how AI adapts to them.
How research informed the interview
Personal trial and error identified the problem; established research informed the interview questions, the reflection and correction process, and how personal answers become approved guidance for an AI assistant. I examined whether the practical behaviours I wanted were consistent with established work in communication, accessibility and individualized support, and how to ask others about their own needs without assuming my answers would apply to them.
I examined work in autism support, communication, accessibility and person-centred decision-making. The Autism Healthcare Accommodations Tool and Helen Sanderson Associates' Decision Making Profile show how individual preferences and needs can be converted into practical guidance. Autistic SPACE highlights the role of predictability, communication and context in support.
Research on the double empathy problem examines how misunderstandings can arise between people with different experiences, rather than always belonging to one person. Work on masking and camouflaging cautions against automatically changing a person's voice to fit conventional expectations. Communication passports offer precedents for recording individual needs, while studies of ambiguity, implied meaning and conversational certainty help distinguish what someone experiences from the support they want.
The research informed four principles:
- Start with the individual, not a diagnostic label.
- Recognize that context can change the support someone wants.
- Make AI interpretation visible, correctable and subject to approval.
- Produce practical instructions that the person can review, keep and use with compatible AI systems.
These sources informed the design; they did not supply a ready-made interview, endorse this resource, or establish that it works for everyone. Its usefulness, fairness, limitations and possible unintended effects remain to be evaluated with individuals who bring different experiences, backgrounds and communication needs.
These research foundations informed questions about how people interpret others, express themselves, experience ambiguity and prefer to receive support. The proposed calibration method translates a person's own clarified choices into specific agent behaviour. The research does not establish that any one strategy is right for everyone or that the agent has been clinically validated.
The complete research paper, figures, findings and supporting intake instrument are available through Northline Research. The published studies use synthetic cases and AI judging; they do not establish human communication benefit or clinical validity.
Our approach to responsible AI
AI should support your decisions about how you communicate, not make those decisions for you. Its suggestions are interpretations, not conclusions about your personality, abilities, motives or intentions. You can question, correct or reject them.
There is no single correct way to communicate. The goal is not to make you sound more conventional, encourage you to mask who you are, or change your natural voice. It is to help you understand options and choose what support, if any, works for you.
The agent should neither automatically validate a person's first interpretation nor routinely contradict them. If the individual chooses reflective support, it can ask what is known, acknowledge missing information and offer alternatives without dismissing the person or claiming certainty about someone else's intentions. Communication difficulties can be reciprocal, and the agent must not treat an individual communication style as a defect.
An AI assistant may help you consider how words could be received or explore alternative explanations for a misunderstanding. It cannot reliably know what another person thinks, determine whether a relationship is safe, or guarantee how a conversation will turn out. Its suggestions should not replace trusted professional or personal support when situations involve abuse, threats, coercion or serious distress.
We intend to make assumptions visible, minimize unnecessary personal information and ask people to approve only instructions they understand. Participating in feedback or future research will be voluntary and require its own consent, separate from release notifications and use of the resource.
As we evaluate the resource, we will look not only at usefulness but at mistaken interpretations, unwanted changes to a person's communication style, accessibility barriers and possible harm. Concerns and corrections can be raised through Northline's existing contact channel.
This approach is early and research-informed, not clinically validated. We will document limitations and make responsible changes as evidence develops.
Privacy and your information
Your communication preferences are personal. This resource is being designed to help you understand and use those preferences without requiring Northline to maintain a permanent personal profile about you.
Release notifications. While the guided interview is in development, you can sign up for occasional updates about its release. We ask for an email address and explicit, voluntary consent. You can unsubscribe using the link in our emails. Your email will be used for updates about this resource, not unrelated marketing, and will be managed separately from any future interview information.
Mailing-list processing and retention. Release-update subscriptions use Brevo to collect and manage the email address, opt-in record and confirmation status, while Google reCAPTCHA v2 provides bot protection on the signup form. These services may receive technical information when the form is used. Before launch, Northline will disclose the actual processing/storage regions, relevant provider terms, retention periods and how to request deletion. Signing up does not reserve a place in the free pilot.
Future guided interview. The application is intended for adults 18+. It will not require your legal name, home address, employer, medical records or identifying details to construct a communication profile. We encourage generalized descriptions of third parties instead of names and sensitive details.
How the proposed application processes information. When the guided interview becomes available, information you choose to enter will need to pass through Northline's web application and an AI-processing service for interpretations, follow-up questions and proposed support rules. The proposed architecture currently anticipates a dedicated OpenAI API project, but the actual AI provider and processing arrangements must be verified before launch. This is not solely on-device processing.
Intended retention approach. The design objective is to avoid a permanent Northline database of your interview answers, conversation transcripts, approved profiles and generated AI instructions. Exact treatment of session data, service-provider processing, application/error logs, security monitoring, backups, retention and deletion must be verified and disclosed before the application launches. This page does not promise zero third-party retention.
Using another AI provider. If you choose to use approved profile instructions with another service, that service's privacy terms, account settings and available features apply. Review what you share. Northline cannot control how third-party AI providers process information once you transfer it.
Your choices. The planned resource lets you review, correct, skip or reject provisional AI interpretations and approve only the instructions you understand and choose to keep. No diagnosis or therapy is provided. You should be able to access this section from the future application before participating. Requests to withdraw participation or delete information must be supported to the extent technically possible, with any limits disclosed.
Website-wide privacy. Northline's existing Privacy Policy continues to cover unrelated website-wide processing.
Frequently asked questions
What is the Communications Support Agent?
A personalized AI assistant in development under Northline's AI for Humanity initiative. Originating in lived experience on the autism spectrum, it is intended to help people navigate differences in how they interpret messages, express themselves and understand others without imposing a conventional communication style. A guided, research-informed calibration process is intended to let individuals choose and approve the strategies they want. The resource is not clinically validated.
Who is this resource intended for?
It is intended for adults aged 18 and older, particularly individuals on the autism spectrum and others who may benefit from different communication strategies. You do not need a diagnosis to participate.
Is this only for individuals on the autism spectrum?
No. The idea began with the founder's experience on the autism spectrum, but communication preferences and misunderstandings are not unique to autism. People have different needs depending on their experiences, circumstances and relationships.
How is the Communications Support Agent different from ordinary AI personalization?
Rather than relying only on assumptions an AI system accumulates about you, the calibration interview lets you choose and approve communication strategies—for example, distinguishing facts from interpretation, exploring alternative meanings or reviewing how your words may be received. You also decide when the agent should use them.
How is my agent personalized?
A guided, research-informed interview calibrates the agent to you. It asks about real communication situations, checks its understanding of your answers and helps you choose when you want particular kinds of support. You review and approve the resulting instructions.
What will the interview ask me?
Questions explore ambiguous messages, interpretations of other people's intentions, how your own words might be understood, and whether you want an initial assumption challenged. They also establish when you want particular support strategies applied. Examples on this page are illustrative, not a complete validated assessment.
What if the AI misunderstands me?
AI can misunderstand what you mean. The process is designed to let you correct, reject or skip interpretations rather than treating them as facts about you. The method still needs evaluation with a wider range of participants.
Will it tell me how I should communicate?
No. Everyone is different, and so is the support they need. You decide which strategies your agent should use, when to apply them and how actively it should intervene. The agent is intended to help preserve your own communication style, not impose another one.
Can it help me prepare for difficult conversations?
It is designed to help you clarify what you want to communicate, examine assumptions, consider how your words might be understood and prepare possible responses. It cannot know anyone else's intentions or replace trusted support in situations involving abuse, threats or serious distress.
Will Northline store my conversations or communication profile?
The design objective is to avoid permanently retaining participants' conversations and completed profiles. The application will still need to process information through its services, and exact logging, retention and deletion details must be verified and explained before launch. The separate release-notification list holds the email address you choose to provide.
Will I have to share my diagnosis or personal identity?
No diagnosis is required, and the planned interview does not require your legal name, address or medical records. Avoid sharing unnecessary personal or identifying details about yourself or others.
What will I receive when I finish?
The planned outputs include a readable record of your approved communication support strategies, personalized agent instructions and guidance for applying those instructions through a compatible AI service.
What can my personalized agent help me do?
When you ask, it may help interpret a message, distinguish fact from inference, consider plausible alternative explanations, identify assumptions or potential communication blind spots, or prepare a response in your own voice. It does not automatically agree with or challenge you: its approach follows your approved calibration.
Can I use the profile with ChatGPT or other AI assistants?
The intended instructions can be applied to compatible AI services using the configuration features each provider offers. You use your assistant through that provider, not through an ongoing Northline conversation. Available features and results vary.
Is the resource free?
The initial release is intended to be free and personally funded, with limited capacity. Signing up for updates does not reserve access or guarantee a place.
When will it be available, and how can I receive updates?
There is no confirmed public release date. You can join the optional Communication Support — Release Updates email list below. You'll need to confirm your email subscription and can unsubscribe from future messages.
Is this a diagnostic test or a form of therapy?
No. It is not an autism assessment, clinical or psychometric test, therapy or medical advice. It is an early, research-informed approach whose usefulness and potential unintended effects still require evaluation.
How can I contribute, share concerns or collaborate?
Northline welcomes contributions from people with lived experience, communication and accessibility professionals, researchers, technology practitioners and organizations interested in responsible, human-centred AI. If you would like to contribute or explore collaboration, please complete the Collaboration Inquiry form on this page. General inquiries and concerns can be submitted through Northline's Contact page.
Help shape this initiative
Interested in Contributing or Collaborating?
The Communications Support Agent is being developed as a free public-interest resource. While its initial direction is established, there is important work ahead in evaluating how well it supports different communication needs, identifying limitations and improving its accessibility and usefulness.
Northline welcomes perspectives from people with lived experience, communication and accessibility professionals, researchers, and organizations interested in responsible, human-centred AI.
Collaboration could include sharing practical insights, reviewing aspects of the approach, contributing relevant research, participating in evaluation or exploring opportunities to support future development.
The initial development is funded by Northline. Any future collaboration or partnership would be discussed separately, with clear expectations about its purpose and scope.
Get notified when it's available
We're developing the Communications Support Agent and plan to make a limited free first release available. Leave your email address if you would like occasional updates.
If the form does not display, use the accessible hosted signup form. The form uses Brevo, double opt-in and Google reCAPTCHA v2.
Signing up does not reserve a pilot place or guarantee access. Initial free capacity will be limited. Access arrangements will be communicated transparently and should not depend solely on how quickly someone responds to an email. Your email is used only for the updates you request. Read Privacy and your information above.
For research or professional collaboration, use the Northline Contact page.