noticias ia para empresas agentes de ia automatizacion de procesos

Gartner delivers the day's uncomfortable number: only 13% of technology leaders say their company has seen significant value from AI, and what sets the winners apart isn't the model, it's knowing when to pull the plug on a project. Money keeps flowing into tools teams already use: Otter.ai, the meeting notes app, has closed $320 million to sell harder to businesses, and Yelp will charge $30 a month for an agent that responds to quote requests. And OpenAI shows that its new models are already migrating COBOL and Visual Basic code, backed by tests that demand exactly the same output. Picking up from the Tuesday, October 20 radar.

Only 13% get real value from AI, Gartner says: the winners decide up front when to walk away from a project

Gartner surveyed 2,002 technology leaders between May and July, and only 13% say their company has realized significant or transformational value from AI. Use cases have multiplied in two years, but the odds of any single one succeeding are still one in five. "Volume is not value," said Alicia Mullery, a VP analyst at Gartner, at the firm's Orlando conference. What sets the winners apart is having rules for stopping: companies with guidelines for deciding when a pilot shouldn't move into production are 2.3 times more likely to be in that group, and those with guidelines for dropping systems already in production that don't justify their cost are 2.6 times more likely. Only 31% have the first in place, and just 17% have the second. Meanwhile, 58% say they're under more pressure to deliver cost savings from AI, and 51% expect it to raise the total cost of ownership of their technology. For your business: before your next pilot, write down in one line what it has to improve (hours, errors, days to get paid), by how much and by when, and what signal will tell you to drop it. Measure it against a set of real cases, like the Santander test bench we covered yesterday and every quarter, review what's already running: if it doesn't deliver, switch it off. A pilot without criteria to stop it won't get you to INTEGRATED: it leaves you stuck at ASSISTED, paying for licenses. Source

Otter.ai gets $320 million it only pays back if it sells: expect your meeting notes app to sell you harder

Otter.ai, the app that joins meetings to transcribe them and write up the notes, has closed $320 million from General Catalyst's Customer Value Fund. It isn't a typical funding round: the fund fronts the sales and marketing budget and gets paid only out of what the customers won with that spending pay, up to a cap; if the sales don't come, Otter owes nothing. Otter says it has more than 40 million users and has transcribed over a billion meetings, and it wants to become the "voice system of record" for businesses, with agents that act on what gets said. The space is crowded with rivals like Fireflies and Granola, and it's already consolidating: Superhuman bought Fathom in September. For your business: your meetings are exactly what these tools want to capture, and Otter just got the money to chase more customers. Before each employee installs whatever note-taker they like, pick one for the whole company and get it in writing: where the recordings are stored, whether they're used to train models, and how you take them with you if you switch vendors or the vendor gets acquired. And always tell everyone in the meeting that it's being recorded. Recording meetings and doing nothing with the notes is ASSISTED; having the agreed tasks flow into your CRM on their own, with someone checking them, is INTEGRATED. Source

Yelp will charge $30 a month for an agent that answers quote requests: whoever replies first wins the job

Yelp has rolled out more than 30 new AI features and updates. For service professionals, such as movers, roofers, plumbers, or cleaners, it's launching Smart Lead Reply next month, an agent that instantly replies to quote requests, gathers project details and qualifies the lead; the professional configures it with their calendar links, prices, and promotions, and it costs $30 per month. In the second quarter, about 10% of quote requests already came in through its AI assistant. For consumers, it's debuting Check for Me: a voice agent that calls the restaurant, asks about the wait time, and texts you the answer; it rolls out in the coming months in the US, on iPhone. It's the flip side of what we reported on Monday, October 19, when Yelp was letting in only the agents that pay for its data. For your business: Yelp may not count for much in Spain, but the trend is coming anyway. More and more requests will arrive through assistants, and whoever replies first gets the job. Check how long you take today to answer a quote request, write down answers to the questions you get most (ballpark prices, service areas, lead times), and get whoever picks up the phone ready to talk to a machine. Replying by hand whenever you get a minute is MANUAL; an agent that answers with your prices and your calendar, and leaves the final quote to a person, is UNATTENDED with guardrails. Source

OpenAI migrates COBOL with 86.6% accuracy, up from 5.4%: the old program nobody dares touch has a way out, if you keep its outputs

On Monday, OpenAI published how its new models handle legacy languages like COBOL, Fortran and Visual Basic, which have far less code to learn from but still run banks and essential services. In its migration test, which requires the program translated into Java, Python, or another language to provide the exact same output as the original, GPT-6.1 Sol hits 86.6% accuracy at low reasoning effort, compared to 5.4% for GPT-5.6 Sol, at 39% lower cost; at higher reasoning effort, it reaches 97.9%. In another test, building a database engine in 20 languages, GPT-6 Astra passes 89.4% of the tests in Visual Basic, versus 27.5% for GPT-5.6 Sol. OpenAI flags the risk itself: COBOL drops excess decimal digits unless told to round, so a rewrite that rounds will run without errors and still produce different numbers. That's why it says production use still needs expert review and testing. For your business: if your invoicing or calculations run on an old Visual Basic or Access application that only one person understands, as we discussed on Monday, October 12, migrating it is no longer a lost cause. The groundwork is on you: keep real examples of inputs and outputs, rounding included, and require the new version to match them to the cent before you switch off the old one. A program nobody dares touch keeps you stuck in MANUAL; with parity tests, migrating it is the first step toward integrating it. Source

What to watch tomorrow

Whether Anthropic and Nscale confirm the second $45 billion compute deal that Reuters and Bloomberg say they're negotiating, which would mean more capacity for Claude. And the rest of Gartner's data from Orlando, where the firm has already said 44% of CEOs struggle to show a financial return on AI.