From an LLM with tools to sector intelligence: what remains to be proven
Tools expand what an LLM can do. The next challenge is understanding the business objective, adapting to new problems and working with the autonomy the company…
A language model connected to a database, a calculator and a planning program can do much more than hold a conversation. It can query a facility, compare costs and prepare a work plan. The interesting question starts there: how far does it understand the problem, and how does it respond when conditions change?
In a company, conditions change often. An investment is postponed, a different offer arrives or a machine becomes unavailable. Sector intelligence must be able to relate those facts to the objective it is trying to achieve.
The capability belongs to the whole system
When an assistant uses an optimizer to find a solution, the result comes from the combination. The model interprets the request, the optimizer solves a formal problem and the data define the starting situation.
Needing external tools does not invalidate that capability. What matters is describing the system accurately and understanding what each part contributes. In energy, research already combines language agents with stochastic optimization to study the balancing of a system.
It also matters what happens when a tool returns an incomplete result or data are missing. Coordinating calls is part of the work; recognizing that there still is not enough to reach a conclusion requires an additional criterion.
Solving one case and adapting to another
A demonstration usually shows the case that works best. To assess a broader capability, I want to see what happens with a problem that was not prepared in advance.
Suppose the system knows how to compare tariffs. Now we ask it to plan an energy purchase, account for a power constraint and review its proposal because production changes. The challenge is transferring what it knows to that new combination.
Levels of AGI proposes distinguishing performance, generality and autonomy. The separation is useful: success on one task tells us about its quality; solving different tasks speaks to breadth; acting independently describes another aspect of the system.
Within a sector, discussing generality requires explaining the range of problems covered and where the limits appear. I use “vertical AGI” to study that ambition within a specific domain.
The business gives tools their meaning
The same mathematical operation can support different decisions. Minimizing a factory’s bill, securing a hospital’s supply and planning energy for a mission involve different priorities. The system needs to understand what it is meant to protect or improve.
A sector-specific world model connects business assets, conditions and relationships. Predictive models let us explore how they might change. On that basis, calculation tools can compare alternatives that serve the right objective.
This connection can also explain why an apparently cheap option is rejected: it may interfere with production, introduce a risk or depend on an unreliable assumption.
The organization decides the autonomy
Finding a suitable alternative and having permission to contract for it are different matters. A company can let the system prepare scenarios while reserving the final decision for a person. It can also authorize specific actions within defined limits.
SARA Scale addresses the properties of that governance. The ability to adapt needs other kinds of evidence. Both questions matter, but each tells us something different about the architecture.
At TheryOS, I am interested in building that relationship between capability and business context. A system deserves trust when it can explain what it proposes, review its plan when something changes and stay within the responsibility the organization has given it.
Related reading
References
Alejandro Díaz
Founder of TheryOS, business owner and entrepreneur in the energy and financial sectors.
