Optimizing Human-AI Collaboration
AI & HumanitySolving the AI Personality MismatchHave you ever felt a mismatch between yourself and your AI agent? Do you just not ‘get’ each other? Are the results fine–or even great–but the process feels awkward? Did you give it a second try or a third with a different model, but get the...
AI & HumanitySolving the AI Personality MismatchHave you ever felt a mismatch between yourself and your AI agent? Do you just not ‘get’ each other? Are the results fine–or even great–but the process feels awkward? Did you give it a second try or a third with a different model, but get the same result? There is significant work and research currently addressing these issues, and two of the two most common responses I see are giving the AI better context about the work, and training the person to better understand the AI. One idea that I’m not hearing anyone talk about is addressing the personality gap between the human and the AI. This is why I built The Cognitive Bridge to map and align these dynamics.Why AI Feels So Generic–and Why This is Bad News for AI AdoptionWe are currently witnessing a trend toward convergent development in AI. Conversational models increasingly sound identical, and there are a few reasons for this:They are based on the same corpus of human knowledge and writing.They are designed to be likable, engaging, and to please the user.They are trained by people who often share similar biases — tech-forward, individualistic, ambitious, and often quite wealthy.If you don’t get on well with the dominant personality in AI — and many people report that they don’t — you are often left with subpar results. The common instinct is to assume that’s just how it is. After all, we have a lot of options in the AI market. There are multiple different models available from multiple different providers with multiple harnesses. It’s not like the two-horse race that is desktop operating systems where there is nowhere to go if you don’t like Windows or macOS. If none of the AI options work for you, then there must be a fundamental barrier between you and the AI.As a trained counselor and a student of humanity, I resist the idea of fundamental barriers. I think that much that divides us can be overcome, but usually it requires first being understood. In this aggressive phase of AI adoption, where technological and enterprise-grade solutions are getting all of the attention, I find that we sometimes don’t slow down enough to understand the problem and the people we are dealing with. So let’s take a step back and think about this.Your AI is pretty cool. What happens when a real person uses it?As compute becomes scarce and tokens for AI reasoning get more expensive, anything that increases the efficiency of the model and the tools that produce your output is essential. See the attached chart from Artificial Analysis that measures AI models based on not only their raw intelligence but also their token efficiency in getting to a satisfactory final answer.Chart from Artificial AnalysisThe chart shows the native efficiency of the model in isolation, but there are situational/environmental/personality drags on efficiency, as well. In other words:Your AI is great, but how does it work when a real person uses it?Take your smartest employee and the best large language model that money can buy and put them together — they will probably produce amazing results, but it might take unnecessary rounds of fine-tuning, explanation, or correction if the user and the AI have personalities that aren’t aligned to begin with. Enterprise is spending billions of dollars on AI, and their great fear is that they might not get their money’s worth, that employees will resist, that competitors will leap ahead of them.Fortunately, we have a mental model firmly in place that addresses this dilemma: the hiring process.Hiring Your AIMost managers don’t hire solely for a skill set; they hire for a personality match with themselves and the existing team.While this is rarely official, I have seen it in the eyes of my teams during many post-interview discussions: they like, respect, or trust some candidates more than others. That vibe matters when you are planning to spend hours collaborating with someone.And yet, we have completely omitted this aspect of the hiring process with our AI. Does their personality click with mine? If it doesn’t, we are fighting an uphill battle. I’m not suggesting we just crank the “likability” scale up to 11. Friendliness, positivity, and cheerfulness do not work well for all people or all tasks. People have complex, multifaceted personalities. This project isn’t just about a “vibe”; it is about recognizing that, much like in an office, it is hard to get good work done when your team does not click.Introducing The Cognitive BridgeIf your AI’s personality doesn’t align with yours, I commissioned* a tool with Google’s Gemini AI to change it: The Cognitive Bridge. It does three things:Asks the user a short series of questionsAnalyzes the results to give the user a personality profile.Tunes an AI chatbot to be aligned with that user’s specific needsIt even allows you to chat with the aligned model right next to a deliberately misaligned one, so you can see the difference for yourself. You can also copy your “personality context” and plug it into your AI of choice.The Cognitive BridgeFor the personality model, I used the “Big Five” OCEAN model of personality traits:Openness to new experiencesConscientiousnessExtraversionAgreeablenessNeuroticismEach dimension in this model is assigned a value from 1 to 100. It is important to view this as a descriptive measure rather than a performance score; personality traits are not fundamentally positive or negative. Instead, they can be poorly matched to a situation or used at inopportune moments. These numerical values illustrate how varying intensities of these traits influence one another.Long-term, I can see updating this project to address differences of age and phase of life, religious beliefs, country of origin–there are many identity factors that influence how we think. But these aspects of human diversity should be approached with respect, with abundant research, and with multiple external opinions to identify and remove bias. Maybe I’ll work on those in v2.0. For now, I’m glad to have a tool that can remove a bit of the friction that can arise between people and their AI tools. Try it out for yourself and let me know how it shapes your AI team.Tech StackGetting an AI to listen to a human’s input, determine an OCEAN score, and then adopt an aligned and misaligned personality in response was an eye-opening experience in the fuzzy border between tech and life that is AI. I am not saying that I believe that AI is alive–I share the Pope’s belief that human life is distinct from machine intelligence because the former is embodied, incarnate, born out of experience, while the latter is simulated.But interacting with AI requires a paradigm shift from clear input/output determinism to a more relational, interactive conversation based on probabilities.So I gave my interviewer AI agent a baseline of five academic articles covering the OCEAN system of personality traits. Its mission is clear compared to its fellow agents: ask a series of narrow questions to determine the user’s scores. AI is generally good at sticking to clear systems. But tuning the aligned and unaligned personality result agents was more involved, as they had to take in what is known about the OCEAN system and extrapolate how untested combinations of personality scores would help/hinder interaction with an AI.The Cognitive Bridge’s multiple agents evaluate each other.I attached a robust system of evaluation and monitoring to keep the multi-agent framework consistent over future changes and new information. This included adding a psychometric validation agent to take in the interview transcript and evaluate how the diagnostic agent scored the user. On the other end of the process a distinct validation agent reviews the aligned and unaligned chatbot responses to evaluate if they are consistent with their expected results.I built my prototype in Google AI Studio, then exported it into Antigravity IDE for more control over the code. I graduated my local database to Firebase, and switched from the AI Studio API to Vertex for LLM orchestration. I used a local model of Gemma4:26b to keep down costs during the agent alignment and testing process–after I blew my first month’s Gemini token budget on the initial launch…ConclusionWe have always had to accept the software with which we work as they were built: in the image of their creators. Limited customization has grown over time, but this often invites confusion as users struggle to keep up with and find the hidden nuances of their advanced tech. The age of AI points to a new potential for truly bespoke tools that invite customization in the user’s own language. Everyone, from enterprise workflow managers to individual users, can benefit from considering AI-human personality alignment as a measurable and achievable goal.CitationsA-five-factor-theory-of-personality McCrae, R. R., & Costa, P. T., Jr. (1999). A Five-Factor Theory of Personality. In L. A. Pervin & O. P. John (Eds.), Handbook of personality: Theory and research (2nd ed.). New York: Guilford.Personality Pairing Improves Human-AI Collaboration Ju, H., & Aral, S. (2026). Personality Pairing Improves Human-AI Collaboration. arXiv preprint arXiv:2511.13979v2 [cs.HC].Personality-driven AI agents: Operationalizing OCEAN traits for human-AI collaboration in the coding domain — Amazon Science Garg, A., M, I., & DeLaPena, R. (2026). Personality-driven AI agents: Operationalizing OCEAN traits for human-AI collaboration in the coding domain. Amazon Science.Bhandari, P., Naseem, U., Datta, A., Fay, N., & Nasim, M. (2025). Evaluating personality traits in large language models: Insights from psychological questionnaires. arXiv. https://doi.org/10.1145/3701716.3715504.Stein, J.-P., Messingschlager, T., Gnambs, T., Hutmacher, F., & Appel, M. (2024). Attitudes towards AI: Measurement and associations with personality. Scientific Reports, 14. https://doi.org/10.1038/s41598-024-53335-2.Note(*): I’m using Ethan Mollick’s suggestion of referring to projects done with AI as ‘commissioned’. I did not ‘make’ this project in the sense of hand-coding it, but I did create the idea, the structure, the requirements, and deeply reviewed and refined the outputs.Note: These are my own words, reviewed and lightly edited with Gemini AI.This story is published on Generative AI. Connect with us on LinkedIn and follow Zeniteq to stay in the loop with the latest AI stories.Subscribe to our newsletter and YouTube channel to stay updated with the latest news and updates on generative AI. Let’s shape the future of AI together!Optimizing Human-AI Collaboration was originally published in Generative AI on Medium, where people are continuing the conversation by highlighting and responding to this story.Source: Generative AI Pub — Published — Category: Image AI