Reimagining Physical Work with AI and Robotics
An interview with Julien Seret, SVP of Technology and Business Integration
Helen Dwight: Welcome Julien! Could you start by sharing your background and tell us what excites you about your role at SoftBank Robotics America (SBRA)?
Julien Seret: I’ve been in robotics for over 15 years, beginning with a startup in France that was later acquired by SoftBank to create a bigger version of the humanoid robot we had at the time. The robot was the well-known humanoid, Pepper, and I was product manager. Since then, I’ve led many automation and robotic initiatives – most recently across industries like commercial cleaning and security. We are very disciplined in our approach to robotics at SBRA. My role is about analyzing all the fields of economic activity on the planet - to see if they’re ready for automation and whether emerging technologies make them viable. What excites me now is that the field of opportunities is so wide. Many use cases once considered futuristic are finally practical, scalable, and deliver strong returns on investment.
Helen: Let’s start by defining the basics: What is a robot, and what is AI in the context of reimagining physical work?
Julien: A robot is a machine that can sense and act on its environment, often autonomously. AI, meanwhile, is about making decisions and reasoning. The connection is clear: robots need AI to make intelligent decisions about how to act. What’s changed is the cost and capability—sensors, actuators, and computing are cheaper and more powerful, making sophisticated autonomy possible. When I first started in this field, a humanoid robot might cost $500K or $1M with a relatively limited capability including trying to pre-empt all possible responses to voice commands. Now we’re talking maybe $30K with capabilities far in advance like sensing, acting, reasoning and voice interpretation. The technology capabilities are accelerating rapidly and ripe for investment and widespread use. Early customers are already seeing enormous benefit.
Helen: What are some of the core problems robotics is trying to solve as it relates to the disruption of physical work paradigms?
Julien: Humans dislike repetitive, dull, or dangerous work, and quality drops when we’re forced to repeat tasks endlessly. Robotics helps in several situations:
- Repetitive tasks: Commercial robotic vacuums and scrubbers handle tasks such as floor cleaning where quality and consistency of the task matters.
- Dangerous environments: Robots can operate in fire zones, nuclear sites, or deep underwater.
- Labor shortages: In remote e-commerce warehouses, the labor pool is limited and/or staff are often unwilling to commute. Robots can fill the gap in many scenarios.
- Aging societies: Robots are being tested in taking on physical care tasks. There will be no other option here given the imbalance in population sizes as people age.
Robotics is a continuation of automation that began with the Industrial Revolution—giving us equipment such as washing machines to remove drudgery so we can focus on higher value work. Automation today is simply more intelligent, capable and advancing at a rapid rate.
Helen: Why is the environment more ready today for widespread robotic adoption?
Julien: There are several converging factors, for example:
- Hardware and AI maturity: Robotic hands—once a century-long engineering challenge—are now affordable and dexterous enough for nuanced tasks. Coupled with AI, robots can, for example, grasp a slippery container of liquid and adapt in real time.
- Affordability: Robots that once cost millions are now accessible at tens of thousands of dollars. This investment is now easily justified.
- Human-Machine Interaction: In Manufacturing environments, robots are caged off for safety. Now, collaborative robots work side-by-side with staff in warehouses, hotels, and offices. The evolution of LLM’s and interfaces (like natural language or vision systems) are making robots more approachable and able to understand human interaction. Their ability to sense obstacles in the environment makes them considerably safer to work with and around too.
- Expectation vs. reality gap closing: Previously, robots disappointed users because they failed outside scripted conditions. Today, capabilities are so much higher that it becomes easier to manage expectations. For example, manufacturing robots can detect misaligned objects on the assembly line and adapt, instead of freezing.
The result: automation has expanded beyond manufacturing into service environments like hospitality, healthcare, and logistics. Democratization of the technology is real and delivers business value.
Helen: How does AI accelerate adoption of robotic automation programs and further enable the adoption and use of robotics?
Julien: I think it’s first important to define the types of AI, before we talk about acceleration of adoption. They are:
- Narrow AI: Specialized for single tasks (e.g., object recognition, computer vision, speech recognition and translation), which empowers perception and task execution. Examples are autonomous vehicles – cars or delivery trucks.
- Reinforcement Learning AI: Robots learn optimal actions via trial and error, adapting to tasks dynamically. For example, robotic picking in warehouses: once unreliable because of object variability, now requires human intervention far less often (from every 10 minutes to perhaps once a day).
- Generative AI & LLMs: Enable new content creation from learned data patterns (e.g., LLMs, image generators). Robots are easier to instruct. Instead of programming, you can simply tell a robot what task you need completed.
- General AI: Hypothetical; would perform any intellectual task a human can.
This creates a virtuous cycle: robots generate data while performing tasks → data improves AI models → smarter robots enable new applications. ROI improves as robots become more autonomous, error rates fall, and the level of human intervention required drops from frequent to rare.
Helen: What is your view on the challenges AI faces relative to energy consumption and appropriate adoption of the technology
Julien: As with all innovation, there will be challenges to overcome:
- Energy & Compute Costs: Running advanced AI consumes enormous computing power and energy. To give you a sense, a single AI query can use 20x more energy than a Google search. In the book Abundance by E. Klein, the topic of supporting the power demands of AI (especially generative AI data centers) is discussed. It requires planning for carbon-free energy. He goes on to say that we would need to build a solar facility every 15 days to meet the rising energy demand. Because of this, I firmly believe companies must treat compute like a new “raw material” in their cost models.
- Hardware Limits: Dexterity is improving, but if we take the human hand as a benchmark, we’re still behind its capabilities.
- Design for Automation: To maximize ROI, processes and environments (like office buildings or landscaped gardens) must be designed and engineered to be compatible with robots to enable greater robot productivity.
- Education & Workforce Transition: Engineers, for example, can spend upwards of 30% of their time searching for information. AI can reduce this. Training workers for new, higher-value roles is essential for the future of the workforce. Education curriculums must evolve to include prompt engineering, critical thinking, and collaboration with AI in order to prepare individuals for the future. There are a plethora of tools, but many remain underutilized.
Helen: One last question. How fast and sustainably can we scale robotics and AI in the workplace?
Julien: First, it’s a question of how fast, not if we scale robotics and AI in the workplace. Companies that adopt could see 10x or greater EBITDA improvements. Those who don’t may struggle to survive, as competitors use robots to lower costs, scale faster, and expand into new markets.
- Industry-specific AI solutions and use cases: Already, autonomous vehicles in San Francisco, robotic pickers in warehouses, and commercial cleaning robots in hotels and senior living facilities show the breadth of adoption.
- Standards and Education: The key isn’t just regulation—it’s education. For instance, schools are teaching kids how to write AI prompts, turning creativity and precision communicating into new foundational skills.
In summary, I believe Robotics and AI will shift industries from mass production to mass customization. Workers will move into creative design, human interaction, and oversight roles. Education and training are central to ensuring this transformation benefits both today’s workforce and future generations. I’m excited about the future. There is an opportunity for every child to benefit from personalized education—think an AI tutor for every child—making knowledge more accessible and adaptive than ever before.
More Information:
For more information about solutions and services offered by SoftBank Robotics America, please visit the website.
