AI adoption splits large and small firms wide open in Spain
A new Spanish survey shows big companies racing ahead on AI while micro-businesses barely start, plus a specialised document-extraction model, agentic security concerns, and Santander's innovation awards.
Every line links to a primary source in the full briefing.
01Spanish survey: large firms race ahead on AI, micro-businesses barely move
A new survey reported by El Español finds 74% of large Spanish companies already use AI in some form, against only 21% of micro-businesses. The gap mirrors earlier findings from Spain's central bank and other surveys this month showing deep integration remains rare outside big firms.
This is a direct measure of competitive risk for a small Spanish business: your larger competitors are several times more likely to already have AI embedded in daily work.
Source: El Español
02Santander awards six companies for AI and quantum computing projects
Santander recognised six companies for projects spanning artificial intelligence and quantum computing, part of an ongoing push by the bank to spotlight applied innovation among its business clients. Coverage was limited to naming the winners and the categories involved.
Awards like this are a useful place to scout what AI projects are actually working for comparable Spanish businesses, even if the business case isn't detailed here.
Source: PressDigital
03Architecture firms flagged as behind on AI training for staff
Spanish coverage closing out the year for architecture studios highlights a recurring gap: firms have access to AI tools but have not invested in training their teams to use them well. The piece frames this as unfinished business heading into next year.
The warning generalises well beyond architecture: owning a tool licence without training staff on it tends to produce low usage and wasted spend, a pattern worth checking in any small firm's own software stack.
Source: Demócrata
04New open model pulls structured data straight out of documents
Hugging Face released a small, specialised model that reads any document type and returns only the specific fields requested, such as invoice line items or ID data, as structured JSON, rather than raw OCR text. The developers say it outperforms general-purpose frontier models on this narrow task at much lower cost.
Manually keying data from invoices, forms or IDs into a database is exactly the kind of repetitive task this targets, and a smaller specialised model means the cost of automating it keeps falling.
Source: Hugging Face
05Report warns AI agents create data-tracking blind spots
MIT Technology Review's sponsored content arm describes how chains of AI agents pulling data from multiple internal systems can produce outputs with no clear record of where sensitive information originated. It argues that securing agentic AI requires tracking how data moves, not just securing individual tools or systems.
If your business lets AI agents touch pricing, customer or financial data, this is a reminder to ask whether anyone could actually trace a mistake back to its source before switching more processes over to agents.
Source: MIT Technology Review
06Hugging Face policy chief on who actually profits from open-source AI
In a TechCrunch interview, Hugging Face's chief policy officer discussed how open-source models are becoming more mainstream for enterprises and startups seeking lower costs and more control, while Chinese open models continue gaining ground. The conversation focused on what businesses should weigh before building on open models.
Choosing an open-source model over a paid API is increasingly a real option for cost-conscious businesses, but the trade-offs around support and control are worth understanding before committing.
Source: TechCrunch
Check whether your team actually knows how to use the AI tools you already pay for
The architecture-sector story and the Spanish adoption survey point to the same gap: many businesses have AI access but little training, which is a likely reason small firms lag so far behind large ones. Spend twenty minutes this week asking staff what they actually use, and what they don't understand about the tools you've already bought.
Sources
6 links- 01Santander premia a seis empresas por sus proyectos de IA y computación cuánticaPressDigital
- 02Las grandes empresas se comen a las pequeñas en el uso de la IA: el 74% ya la emplea, frente al 21% de las micropymesEl Español
- 03Los estudios de arquitectura cierran 2026 con una asignatura pendiente: formar a sus equipos en IADemócrata
- 04Dharma Field Extractor: Any Document In, Structured Fields in JSON OutHugging Face
- 05Open source AI is booming. Hugging Face explains who profits.TechCrunch
- 06Securing the agentic enterprise with data lineageMIT Technology Review