Reimagining the CV with AI: An Early Experiment in Agentic Thinking.
Chat GPSteve is an AI chatbot trained on my career experience. Built in 2023 as an experiment in combining conversational AI, generative imagery, and video animation into a single interactive experience. What started as a creative side project ended up shaping how I think about AI systems today.
In 2023, I set out to answer a question: could I create a virtual version of myself that captured not only my experience, but also my voice and presence? The result was ChatGPSte, an experiment in connecting multiple AI technologies to reimagine the professional profile.
Looking back, this project was an early step toward the thinking that now drives my work on Agentic Experience (AX). It forced me to consider what happens when AI systems need to understand, represent, and act on data.
Building a Conversational Foundation
The main goal was to make something that felt real. Not a generic chatbot that could have belonged to anyone, but one that sounded like me — the way I'd actually answer a question about my experience.
I chose Writesonic to build the conversational layer. The decision came down to control: I needed to train the bot on my own content rather than relying on a general language model. I fed it my website copy, professional profiles, and detailed answers to the questions I get most often about my career. That last part mattered. The difference between a chatbot that says "Steve has experience in experimentation" and one that explains how he ran a testing programme at a financial services firm is the difference between a gimmick and something useful.
This process taught me something I now use in all my AI work: the quality of what you put into an AI shapes the quality of what you get out. Good, well-structured content is not a nice-to-have. It is the starting point.
Bringing the Avatar to Life
A text-only chatbot felt incomplete. If the goal was to capture presence, not just information, it needed a face.
I turned to Midjourney to generate a 3D avatar. The challenge was getting something that looked like a stylised version of me rather than a generic stock character. It took several rounds of prompt engineering to get the expression, setting, and lighting right. The result was an avatar that matched the tone of the conversation: approachable, a bit informal, clearly not trying to pass as a photograph.
The final piece was animation. Using D-ID's technology, I transformed the static image into a video with natural head movements and lip-sync. The effect was immediate, people could view the animated version in a way they never did with the text-only chatbot.
Bringing Writesonic for conversation, Midjourney for imagery, and D-ID for animation together into one experience was the real test. Each tool did one thing well. The value came from connecting them into something none of them could do alone. That principle, that the right combination of AI tools creates more than the sum of its parts, has guided every multi-model system I have built since.
What This Experiment Revealed
What started as a creative side project ended up raising questions that have shaped my work ever since.
As conversational AI gets better, the focus is moving away from tracking funnels and toward training AI with good data and clear content. When AI agents are handling thousands of small interactions for users, organisations need strong ways to track, measure, and improve those moments. The same evidence-based discipline that made experimentation programmes work, test the assumption, measure the outcome, let the data decide, applies directly.
Building Chat GPSteve also forced me to think about what happens when an AI system needs to represent someone accurately. The training data had to be curated, structured, and tested against real questions. Get it wrong and the bot would confidently say things I never said. That problem scales. Every organisation deploying AI agents faces the same risk: the system is only as trustworthy as the content behind it.
From Side Project to Way of Thinking
The gap between this 2023 experiment and today's production-ready AI agents is remarkable. But the questions it raised have not gone away. How do you ensure an AI system represents your organisation accurately? How do you measure whether it is actually helping? How do you keep a human at the centre when the interactions are automated?
Those are the questions I now help organisations answer. What started as a creative exploration of what AI could do became a lens for understanding what AI should do: serve both human users and autonomous systems, with evidence and guardrails built in by design.
That realisation is at the heart of everything I do.
