Robotics after LLM: why the next question is not chat, but action in the world
LLMs changed the interface for interacting with computers. The next shift is physical systems that understand context and act in the real world.
For the past two years attention has been focused on language models - and understandably so. The interface changed fundamentally: instead of clicks and forms, conversation appeared. Millions of people began interacting with software differently than they had before.
But in parallel with this, something less visible is happening. The same approaches - large models trained on vast amounts of data, the ability to generalise across different tasks - are starting to be applied not to text, but to physical interaction with the world. That is a different scale of question.
What changed in robotics after LLM
Before large models, robots were a well-solved problem in only one sense: in a very controlled environment, with predictable objects and repeating motions. An industrial manipulator on an assembly line is exactly that kind of system. It does the same thing thousands of times and cannot handle anything unexpected.
The new direction is systems that can understand instructions in arbitrary form and generalise experience to new situations. A robot told "place the red item on the shelf next to similar ones" - not specific coordinates, but a semantic instruction. This is a fundamentally different class of tasks.
Progress here happens through using LLMs as the "brain" for action planning, with the robotic body as executor. The language model understands the task, breaks it into steps, responds to changes. The physical system executes.
Where this is actually being applied
Warehouse and logistics is the most mature area. Systems that can handle unfamiliar packaging, different object orientations, unpredictable arrival sequences - these are starting to produce real economics.
Equipment servicing. Inspection robots that can traverse territory, detect anomalies by visual signals, and make decisions about what needs attention and what does not.
Interaction with people in non-standard situations. For retail or hospitality - systems that can navigate a dynamic environment and respond to non-standard requests.
This does not mean robots will replace most operators tomorrow. It means that the applicability of robotics is starting to extend beyond very rigidly structured tasks.
What this changes for companies thinking about automation
If you looked at robotics before and got the answer "not suitable, we have too many variations" - that constraint is beginning to recede. The threshold at which a task becomes repeatable enough for robotics is lowering.
This does not mean you should immediately move toward physical systems. But a few questions are worth keeping in mind:
- Do we have tasks that require physical actions, high repetition, and at the same time contain too many variations for classical robotics?
- What operational processes are tied to physical labour not because it is necessary, but because there was no alternative?
- Where in our chain are physical actions a bottleneck for speed or precision?
The next shift in AI is not a better chat. It is systems that act in the real world and learn from those actions. For leaders who manage operational processes, this matters more than it might appear.