Changing Shapes – Episode 6
In this episode, I sit down with Alex Platonov, a UX engineer and former technical artist who spent eight and a half years at DeepMind, to unpack how simulated worlds train real capabilities, why agents find exploits humans miss, and what “values in product” actually looks like when you’re shipping. Today he’s building an AI-driven call-centre platform at a stealth startup.
Our conversation explores:
A self-taught path from architecture and graphic design into front-end engineering, then DeepMind, then technical art building 3D worlds for agents.
What front-end and simulation tooling enable in research: experiment dashboards, organisational visibility, and environments that both agents and humans can interpret.
Agents in simulated worlds: reward design, emergent “bug-finding”, and why sim speed and scale matter before taking skills to robots.
Sim-to-real and the near-term horizon for embodied AI, plus a candid take on uncertainty, timelines, and what to teach our kids now.
Craft and culture in an AI era: why offline making, woodcarving, and tangible artefacts feel more valuable amid ubiquitous generative media.
The new venture: standing up AI call-centre agents with prompts, instant iteration, human hand-off, and why agility beats static scripts.
Values in practice: reducing user frustration, creating small moments of joy, and choosing work with real impact over “pretty for pretty’s sake.”
Breaking into AI now: bring curiosity, learn the basics, use LLMs as leverage, and understand what they produce so you can correct and direct.
It’s a conversation about building worlds for machines and meaning for people — speed and safety, tools and taste, and where human touch still sets the standard.
Alex on Linkedin: https://www.linkedin.com/in/platonovs
🧠 Subscribe for more conversations at the intersection of culture, values, and technology.
🔗 Past episodes and articles at tomhorak.substack.com
