AI Radar for Businesses — Friday, September 11, 2026
noticias radar ia agentes de ia automatizacion whatsapp

AI Radar for Businesses — Friday, September 11, 2026

· CompaniesAutomation

Meta buys Swedish startup Stilla.ai to reinforce its Business Agent, already used by over a million businesses for support and payments within WhatsApp, Messenger, and Instagram. In Madrid, HeyDiga raises 5.5 million with 200 clients and 78% of conversations automated, achieving an 85.3% resolution rate. Salesforce is negotiating 2 billion for Listen Labs, its third AI acquisition of the year after Fin (3.6 billion). OpenAI introduces the ChatGPT Work Data agent, which respects permissions by table, row, and column. And Harvey closes 550 million at a 15.5 billion valuation with over 400 million in recurring revenue.

Yesterday's radar was about who controls your agents; today's is about who owns the channel through which you speak with your customers. Meta has bought the Swedish startup it was missing so that its Business Agent—already used by more than a million businesses—can handle support and payments exclusively within WhatsApp, Messenger, and Instagram. Meanwhile, in Madrid, a company doing exactly that in Spanish has raised 5.5 million with two hundred clients, 78% of conversations already automated. In parallel, money continues to buy entire layers: Salesforce is negotiating 2 billion for a customer research platform and Harvey is closing 550 million at a 15.5 billion valuation. And OpenAI is putting an analyst inside ChatGPT Work that reads your data warehouse while respecting row and column permissions.

Meta buys the missing piece to handle support and payments within your WhatsApp

Meta has acquired Stilla.ai, a Stockholm-based startup founded by former Shopify executives that emerged from stealth in January with $5 million in pre-seed funding and lasted eight months on the market before being acquired; terms were not disclosed. What it's buying isn't a chatbot: Stilla built agents that work across a company's tools with shared context and permissions, and it goes directly toward reinforcing the Meta Business Agent, launched in June and already used by over a million businesses to respond and close transactions on WhatsApp, Messenger, and Instagram. For your company: the channel where your customers already message you is going to bring its own agent as a standard feature, and whoever activates it first owns the conversation—and the history. Before plugging it in, decide what it can do alone (inform, book, charge) and what it cannot, and ensure that conversations are exported to your CRM and don't just stay in Meta's inbox: if the history lives at your messaging provider's house, your second step is rented. Source

5.5 million and 200 clients: the same agent, but in Spanish and with a provider's name

HeyDiga, founded in Madrid in July 2025 by David Zafra and Sergio Espeja, has closed a 5.5 million euro seed round led by the British fund 6 Degrees Capital, with Kfund, Italian Founders Fund, Decelera, and the founders of Jobandtalent; in December 2025 it had raised 1 million. Its agent handles phone calls, WhatsApp, and chat, has more than 200 clients in automotive, health, beauty, and hospitality—Audi, Seat, Toyota, Northgate, Marco Aldany—and claims to automate more than 78% of conversations with an 85.3% resolution rate. Yesterday we mentioned that the technical cost of automated voice had plummeted; today the new data point is the other side: there is now volume, large Spanish clients, and a published resolution figure. For your company: that 85.3% is the benchmark to negotiate with. Ask any agent provider for their resolution rate without escalation and the percentage of automated conversations based on your own history, not their demo, and set in the contract what happens to recordings and transcripts if you leave. Source

2 billion to ask your customers: Salesforce buys its third layer of the year

Salesforce is negotiating the purchase of Listen Labs, a platform that automates customer research by conducting AI interviews, for about $2 billion, according to Business Insider; talks are ongoing and may not end in a deal. The figure is four times the $500 million valuation the company had in January after raising a $100 million Series B, and comes after Listen Labs walked away from a signed $125 million term sheet. It would be its third AI acquisition of the year, following Fin—formerly Intercom— for 3.6 billion in June and m3ter in July. For your company: the takeaway isn't the figure, it's the pattern. If your CRM suite is absorbing customer service, consumption measurement, and market research, in two years your five providers will be just one, and your negotiating power will be whatever you put in writing today. Review the change of control clause in your contracts and the right to export data and configurations in an open format. Source

OpenAI puts an analyst in ChatGPT Work that respects row and column permissions

OpenAI introduced the Data agent within ChatGPT Work yesterday: it connects to approved data sources—Snowflake, BigQuery, Databricks, Redshift, ClickHouse, MongoDB, Datadog—drags files from Google Drive and SharePoint, investigates why a metric moved, and builds dashboards in Power BI, Tableau, Sigma, Oracle BI, ThoughtSpot, or Omni. What's relevant is not the conversation but the control: the administrator decides which sources exist and which roles see them, and the system respects existing permissions at the table, row, and column level, applying business definitions from dbt, Databricks Genie, or Snowflake Horizon. Alpha clients include NTT Data, Thermo Fisher, and ServiceTitan. For your company: this turns a management question into a query without going through anyone, and there the bottleneck stops being technical and becomes data governance. Before opening it up, write down who approves each source and check that your current row-level permissions—the salesperson who should only see their portfolio—actually exist in the warehouse and not just in the report. Source

Harvey: 550 million more and 400 in recurring revenue in professional services

Harvey has raised $550 million at a $15.5 billion valuation, co-led by Lightspeed and Diffusion, less than six months after closing 200 million at 11 billion. It exceeds $400 million in annual recurring revenue and more than 3,000 clients, including 80% of the world's 100 largest law firms, 20% of the Fortune 500, and half of the Fortune 10, and is using the money for proprietary legal models that keep client documents under the firm's control. For your company: if you work with lawyers, management agencies, or auditors, the question this quarter is no longer whether they use AI, but what tool they use to process your contracts, where they are hosted, and if your documents are training anything. Ask for it in writing in the next fee renewal, along with the subrogation of your data if they change providers. Source

What to watch for tomorrow?

Whether Salesforce confirms or breaks off talks with Listen Labs, because it sets the price for consolidation through the end of the year, and whether Meta sets a date for the arrival of the reinforced Business Agent to European businesses, where the AI Act mandates notification that the one responding is not a person.

Watch the 1-minute video