When Alexandru Voica, head of corporate affairs at the prominent video-generation startup Synthesia, sent a digital link this past summer, the contents immediately caught me off guard. It was an interactive virtual avatar of Voica himself, trained meticulously to field and answer common press inquiries regarding Synthesia, its underlying technology, and its core business model. Just a day prior, I had been sitting on a conference panel where public relations professionals earnestly debated whether they minded pitches heavily reliant on AI-generated text. Yet, Voica’s interactive avatar bypassed that entire conversation completely. To me, it felt like encountering the final boss of artificial intelligence applications in public relations.
By September, Synthesia extended an invitation to visit its brand-new office space located in New York City. Originally launched in the United Kingdom, Synthesia stands as a high-valuation leader in the rapidly expanding digital avatar sector, sharing the competitive landscape with peers like D-ID, HeyGen, and Colossyan. The company reached a staggering $4 valuation earlier this year, following reports last year that it had successfully crossed the $100 million milestone in annual recurring revenue.
Synthesia’s core offering empowers enterprise organizations to construct interactive training videos featuring hyper-realistic AI avatars. Expanding beyond its traditional toolkit, the company recently introduced a product suite known as Roleplay Sessions. This innovative platform allows employees to practice high-stakes scenarios, such as complex sales pitches, with an interactive AI avatar that dynamically responds to their arguments and scores their performance in real time.
When the team at the New York office opening casually asked if I would be interested in creating my own AI avatar, I did not hesitate to accept. Naturally, the prospect of possessing a functional digital twin carried an undeniable appeal, especially on a day when my outfit felt particularly sharp and my hair was neatly in place.
Until that afternoon, my attitude toward avatars had largely hovered around casual indifference, though it felt increasingly obvious that they would eventually weave themselves into the fabric of everyday online existence. I had previously caught wind of independent creators on Instagram deploying digital likenesses to streamline their social media content creation pipelines. The entire phenomenon struck me as deeply compelling, which perhaps explains why I harbor no reservations about introducing my own digital twin to readers now. This marks a notable milestone: the first time Synthesia has ever constructed a digital avatar for a journalist, or indeed for anyone outside of internal personnel like Voica. The resulting interactive model has been specifically trained on my recent reporting regarding why venture-backed startups statistically commit more fraud than their non-VC-backed counterparts, and it is programmed to restrict its conversational scope strictly to that investigative piece.
The creation process required stepping inside a compact, professionally outfitted film studio nestled deep within Synthesia’s corporate offices. There, technicians captured a series of high-resolution photographs alongside a two-minute voice recording. After formally granting consent for my likeness to be utilized, the digital version of my persona came to life. The team generated both a personal avatar designed to read verbatim scripts provided by me—offered in variations with and without glasses—and two distinct interactive avatars capable of active listening and conversational response, also styled with and after my signature spectacles.
Once our team selected the investigative article to serve as the training baseline, Synthesia engineers constructed the interactive agent using a sophisticated architecture combining voice-to-text, video processing, advanced language models, and text-to-voice synthesizers. The technical stack powering my avatar heavily incorporates Synthesia’s proprietary video and voice models, though the enterprise platform simultaneously grants clients the flexibility to integrate third-party alternatives from leading artificial intelligence laboratories such as Cartesia, ElevenLabs, Google, and OpenAI. Furthermore, corporate customers retain the autonomy to host these deployed avatars across their preferred cloud infrastructure or contract Synthesia for dedicated hosting services.
Within this technical ecosystem, the voice-to-text module transcribes spoken user input into readable data, the agentic language model interprets the text and determines appropriate conversational actions, the text-to-voice engine converts the generated response back into spoken audio, and finally, Synthesia’s custom video model animates the visual avatar to synchronize fluidly with the speech.
Broadly speaking, Synthesia’s commercial offerings are divided into three distinct categories. The first is a classic video-creation and distribution platform utilizing standard avatars, where users type out a script for the avatar to recite. The second is an agentic platform called Sessions, designed for interactive surveys and roleplay scenarios. The third is an API platform enabling developers to blend Synthesia’s foundational video and voice models with external software services to engineer bespoke interactive agents and applications.
Building the complete suite of my avatars took the Synthesia engineering team a matter of days. My initial experimentation began with the personal avatars, where I drafted a relatively generic script to evaluate the fidelity of my synthesized voice. I directed the model to speak about the arrival of autumn in New York, my absolute favorite time of the year. The vocal reproduction proved remarkably accurate, and I felt secretly relieved that the algorithm managed to smooth over the slight hoarseness lingering in my voice during the initial recording session.
Reactions from non-tech-savvy friends upon viewing the personal avatar ranged from intrigued to thoroughly unsettled.
Reactions shifted further when I demonstrated the interactive model, which operates on a deterministic framework, meaning it strictly adheres to the boundaries of its assigned training data—in this instance, the venture fraud investigation. Deliberate attempts to test the boundaries by asking off-topic questions, such as my previous professional history before joining TechCrunch or the specific neighborhood I inhabit in New York, were met with polite yet firm redirections back to the core subject matter of the article.
While my friends remarked that the synthesized voice and visual likeness fell slightly short of the seamless perfection seen in the personal avatar, the resemblance remained close enough to elicit a distinct sense of unease. Conversely, my mother offered high praise, labeling the technology amazing. Both of my parents immediately attempted to trick the model by interrogating it with intimate questions reserved solely for family members, only to be repeatedly deflected back to the venture fraud article.
Reflecting on the experience, the encounter naturally prompts broader questions about the future trajectory of journalism. One must ponder whether audiences would eventually accept tuning into daily news broadcasts anchored entirely by artificial avatars. When posed this question, one venture capitalist dismissed the notion instantly. Undeniably, a fierce pushback persists today against the proliferation of automated content flooding social networks and information-sharing platforms. Yet, other industry insiders remain hesitant to issue a definitive dismissal. The lingering uncertainty asks whether digital avatars might eventually augment—or entirely supplant—traditional reporting roles, or whether corporate chief executives might one day prefer addressing an AI journalist over a human counterpart.
The most rewarding aspects of my career remain rooted in human connection, investigative writing, and deep-dive research into emerging topics. Yet, the absolute bedrock of journalism is trust, an intangible quality that feels exceedingly difficult to ever successfully outsource to an algorithm.
Outside the strict confines of media and reporting, the concept of self-cloning holds undeniable commercial appeal. The prospect of bypassing the daunting backlog of work following a holiday or extended vacation, secure in the knowledge that a digital proxy remains active to address inquiries, presents a compelling utility for corporate life.
Observing how avatar adoption matures across corporate environments will require patience, but holding my own digital twin leaves me with thoroughly mixed emotions. After the initial novelty faded, I found myself staring intently at the avatar as it concluded its remarks and entered a state of idle waiting—though waiting for what remains entirely unclear. Perhaps a subconscious expectation lingers for the digital entity to blink, to spontaneously utter something novel, to offer a reassuring smile, or simply to acknowledge that it possesses genuine awareness.
Because these digital twins remain strictly deterministic, such spontaneous behaviors will never materialize. Yet, it becomes entirely transparent how easily an individual could slip into a minor state of artificial intelligence psychosis when interacting with a nondeterministic counterpart powered by an unconstrained conversational chatbot free to pontificate limitlessly.
Sharing these observations with a venture investor, I ventured a prediction: my own generation, Gen Z, will likely struggle to grow fully comfortable with digital twins. They possess an undeniable science-fiction quality, as if the speculative tropes long confined to cinema screens have suddenly materialized in daily reality. Even so, digital avatars feel far less jarring than humanoid robotics. At least with a software-based avatar, if the interaction grows excessively strange, logging off remains an immediate option.
Until that future fully arrives, however, my personal avatar stands ready to deliver a brief rundown of the top stories spanning our site this week.