noticias ia para empresas pymes agentes de ia

Sage has opened a €3.5 million program in Spain to help small businesses, accounting firms and advisers put AI into their bookkeeping, cash management or payroll, and it's worth reading what it actually is before you apply. The big money is still buying agents: Salesforce is picking up Cresta, the company whose agents answer customer calls, for about $2.7 billion according to US press reports, and CrowdStrike is paying close to $500 million for two companies that monitor and stress-test agents, while rolling out a switch to shut them down that's already coming to Claude Cowork. And in Luxembourg, an adviser to the EU's top court has laid out when a chatbot that summarizes the news can infringe publishers' rights. Catch up on the Tuesday, November 3 radar.

Sage puts €3.5 million behind AI for small businesses: it's not a grant, it's a way into its products

On Tuesday Sage launched its "Fondos Transformación iA" in Spain, €3.5 million for small and midsize businesses, professional firms and advisers that want to apply AI to a specific problem. To get in, you describe a need, like speeding up an accounting process, cutting admin work or getting a real-time view of cash; Sage then assesses the fit and steers the company to the right solution. It covers five areas, from e-invoicing and Verifactu, Spain's new invoice-integrity rules, to cash management, ERP, HR and CRM, and it runs through Sage's own products: Sage Active and Sage 50 for small businesses, Sage for Accountants for firms, and Sage 200 and Sage X3 for midsize companies. Sage itself calls it a private initiative, subject to eligibility and availability, and warns that applying doesn't guarantee access. Its August survey, which included 500 Spanish small businesses, found the main barriers are a lack of skills (46%), data privacy (36%) and not knowing where to start (35%), while 29% said financial support would do the most to get them moving. The takeaway: this isn't public money you can spend however you like. It's terms for adopting Sage tools, so treat it like any vendor offer. Before you apply, write down the process you want to fix, how many hours a month it eats up and how you'll know it's better, then get in writing what's covered, how much, for how long and what you'll pay once it ends. If you already run Sage 50 or Sage 200, it's a sensible path from an assistant that answers questions to an integrated accounting process; if you don't, don't switch ERPs for an incentive. Source

Salesforce buys Cresta for about $2.7 billion: the agents that pick up the phone are moving into the CRM

Salesforce announced late Monday a definitive agreement to buy Cresta, a voice agent platform that handles complex calls end to end, coaches human reps in real time and analyzes every conversation; its customers include Marriott, Comcast and Alaska Airlines. Salesforce didn't disclose the price, but The Information puts it at about $2.7 billion, and the deal is expected to close between February and April 2027, pending regulatory approval. Cresta will be folded into Service Cloud to tie every call to the customer's history, and Salesforce says it will keep working with the contact center and phone platforms its customers already use. It's one more piece of a buying spree: Contentful, Fin (formerly Intercom), Listen Labs and Braze, which we covered on October 22; counting Cresta, CMSWire estimates Salesforce has spent more than $15 billion on acquisitions this year. The takeaway: voice, the channel where the most expensive complaints come in, is consolidating inside CRMs, and the value isn't that a machine can talk, it's that every call ends up logged on the customer record with its outcome. If you use an independent voice agent, get its roadmap and its promise to keep working with your phone system in writing, with a date. If you're considering one, first count how many calls you get a month and which ones keep repeating: those are the ones that can run unattended, always with a handoff to a person when things get complicated. Source

CrowdStrike pays close to $500 million to watch agents and launches a switch to shut them down, Claude Cowork included

At its Fal.Con Europe conference in Barcelona this week, CrowdStrike announced deals to buy Fiddler AI, which records what each agent does and why, flags when it drifts or makes things up and keeps an audit trail, and Dreadnode, which attacks models, agents and MCP servers with more than 100 known techniques, prompt injection among them, before they go live. It didn't disclose terms; The Wall Street Journal reported $400 million for Fiddler and $95 million for Dreadnode, mostly in cash. The same day it rolled out a kill switch in its Falcon Guardian product, available now: an authorized admin stops an agent where it runs, blocks it from restarting and cuts its network access, and every action is logged. CEO George Kurtz put it this way: it lives outside the model, "where an agent can't argue with it, rewrite it, or wait it out." CrowdStrike also said Falcon Guardian is coming to Claude Cowork, Anthropic's environment where agents run code inside a virtual machine on the user's own computer: it's in beta on macOS, with Windows expected later in November. The takeaway: before you let an agent work on its own, answer three questions in writing: who can stop it, from where, without relying on the agent itself, and what record is kept of what it did. You don't need to buy anything to start: a dedicated account for each agent that can be revoked in one click, plus a log of its actions, is already your kill switch. If anyone on your team uses Claude Cowork, add it to your inventory, because that's where the agent runs code on a company machine. Source

A chatbot that summarizes the news can infringe rights in the EU: what counts is how it gets in and how much it copies

EU Court of Justice Advocate General Maciej Szpunar delivered his opinion on Tuesday in the first generative AI case to reach the court: Hungarian publisher Like Company's suit against Google over answers from Gemini, which pulls news articles in real time to respond to users. His opinion separates two things. Training a model on protected works is, as a rule, allowed under the text and data mining exception. But when a chatbot retrieves an article, reuses it to build an answer and hands that answer to a user, it can infringe the publisher's right depending on two factors: whether it got the article unlawfully, for instance by getting around a paywall, and whether the answer just reports the facts or reuses the article's wording so recognizably that the reader no longer needs to go to the original. Facts belong to no one; the writing does. The provider is liable when that use is built into its business model or when it hasn't put credible measures in place to prevent it. The opinion isn't binding, but judges usually follow it, and a ruling is due in the coming months. The takeaway: if you run an assistant that reads news or other people's websites to put together reports, newsletters or answers, this test already works as a guide. Keep it out of anything behind a paywall or blocked by the site, have it summarize the facts in its own words and always link the original, and put those rules in writing, because having "credible and effective" measures is exactly what separates a liable provider from one that isn't. Source

What to watch next

Sage takes its program to Madrid Tech Show on November 4 and 5, where the actual terms should become clearer. Fal.Con Europe wraps up Wednesday with more agent announcements, and Anthropic still has to set the price range for its IPO.