Blog›Automatización

One agent for the business, one for the tech team: what is reasonable to expect

AI agents shorten the distance between the business and the tech team on specific tasks. What they do well today, what they do not, and how to split the work.

Equipo Neural Fabrik2 min de lectura
One agent for the business, one for the tech team: what is reasonable to expect

In a small or medium business, business information and technical information tend to live apart. Whoever keeps the books does not see what happens in the systems, and whoever maintains the systems does not see what each decision costs. The promise of AI agents is to shorten that distance. It helps to know how far it goes today.

What they do well

Agents perform on bounded, repeated tasks: sorting what comes in, preparing a draft, watching a service, matching a technical fact with its cost. In Neural Fabrik each agent has a scope (marketing, customers, finance, service monitoring) and shares context with the others, so a technical alert can arrive together with its effect on the business.

What they cannot do yet

An agent does not replace a department. In a Carnegie Mellon University test with a simulated company, the best agent completed 30% of the tasks on its own, and long tasks remained out of reach (Xu et al., 2024). It is a measurement with 2024 and 2025 models and the frontier moves fast, but it works as a reference: the sensible expectation is help with part of the work, not full autonomy.

The frontier is jagged

An experiment with 758 consultants showed that AI does not help equally with everything. On tasks within its capability, consultants completed 12.2% more tasks and did so 25.1% faster. On a similar-looking task outside it, they were 19% less likely to produce a correct solution (Dell’Acqua et al., 2026). The difficulty is that the two cannot be told apart from the outside.

What to do with this

  • Give each agent specific, checkable tasks.
  • Measure the result of each one before expanding.
  • Keep a person on anything that has consequences.

An agent that connects the technical side with the business is useful. One that is asked to do everything ends up costing more than it saves.

Sources

  • Xu et al. (2024), «TheAgentCompany: Benchmarking LLM Agents on Consequential Real World Tasks», arXiv:2412.14161.
  • Dell’Acqua et al. (2026), «Navigating the Jagged Technological Frontier», Organization Science, DOI 10.1287/orsc.2025.21838.

Quieres ver como funcionaria en tu empresa?

Te preparamos una demo personalizada con datos de ejemplo de tu sector.

Solicitar propuesta