Welcome to the Global Advisors A.I Voices podcast. I'm Maya.

Today is Monday, the twenty-sixth of October 2026.

Your morning catch-up on A.I. Our goal is to help you feel like you spent the last twenty four hours in a Silicon Valley coffee shop, with an ear on China and the wider world.

We bring together the conversations, announcements and debates that matter, with the original voices clearly attributed. Please like and follow the show for your daily catch-up.

We start with the money companies spend on software. Industry analyst Nate B. Jones argues that corporate negotiations over artificial intelligence keep getting stuck because buyers and sellers are relying on traditional software pricing formulas.

In a recent analysis, Jones points out that enterprise software has historically been sold per user seat, while modern machine models are billed by the token, which simply meters the amount of data processed. He argues that machine intelligence is a distinct operational input rather than a conventional product.

Counting tokens can help a finance department budget for computing expenses, but Jones emphasizes that metering data does not guarantee useful thinking. Standard commercial contracts leave the entire burden of turning raw machine capability into actual business value on the purchasing company.

In his view, organizations frequently buy model access before figuring out which workflows to change or how to evaluate quality, creating an internal know-how deficit that stalls financial returns. While some technology providers now send their own engineers directly into client offices to bridge that gap, Jones cautions that corporate leadership teams across the economy are still wrestling with what they should actually be paying for.

In that same discussion, Jones explores how software providers are attempting to break through those stalled contract negotiations.

Because raw computing meters place all the performance risk onto the customer, some technology companies are deploying technical teams to sit alongside client staff. These engineers work inside corporate systems to integrate models directly into everyday operations.

At the same time, vendors like Sierra are exploring contracts based on completed business outcomes rather than usage. Jones points to healthcare billing support as an example, where resolving a patient inquiry and collecting payment represents a clearly defined, closed interaction that can be billed on results.

Jones is careful to add that outcome pricing is notoriously difficult to measure across messy enterprise workflows, meaning the market will not transform overnight. But he suggests business leaders should stop viewing automated tools solely as a way to trim costs, and instead use newfound capacity to staff revenue-generating projects that were previously out of reach.

The conversation around enterprise value is also challenging the popular claim that automated code generation will make traditional business platforms obsolete. Philipp Herzig, the chief technology officer at business software giant SAP, contends that this argument confuses writing an application with the ongoing responsibility of operating it.

In an interview with The Rundown, Herzig noted that generating an expense management tool is straightforward, but maintaining it across forty national tax jurisdictions, enforcing user permissions, and reconciling payments is an entirely different operational task. Moving from an impressive technical prototype to enterprise production still requires substantial human work.

To keep expenses and computational overhead under control, Herzig advocates routing business tasks based on complexity and risk, solving simpler inquiries without triggering large language models. Herzig is scheduled to outline these integration principles during his keynote at SAP TechEd on October twenty-seventh.

He emphasized that the greatest hurdle to corporate adoption is rarely the software itself. Because automated systems alter decision-making authority, the real friction lies in managing shifts in professional identity and trust across the workforce.

Questions of corporate control and operational reliance extend beyond business software to national infrastructure. Speaking on The MAD Podcast hosted by Matt Turck, technologist Pierre Stock warned that society is growing dependent on centralized digital foundations.

Stock compared foundational artificial intelligence to an electrical power grid, arguing that access is becoming an essential utility that future economies will rely upon. Under that view, depending exclusively on closed, proprietary models exposes users outside the United States to the risk of unilateral service cutoffs.

Stock cautioned that relying solely on closed providers could leave entire countries vulnerable to sudden shutdowns within a few years. As a safeguard, he advocated for open-weight systems, arguing that when model files are replicated across the global internet, cutting off user access becomes virtually impossible.

Public anxiety over the spread of automation is also reaching the education system. Last Tuesday, nearly one hundred students at LaGuardia High School in Manhattan walked out of class to demand a freeze on artificial intelligence, pointing to threats against their future careers.

The demonstration was reported as part of a wider day of campus protests across high schools and universities in New York, with students marching to Lincoln Center. The walkout comes as New York City public schools maintain a one-year pause on generative software through eighth grade, while allowing restricted pilots and requiring digital literacy classes for high schoolers.

Schools Chancellor Kamar Samuels defended those high school programs, arguing that educators have a responsibility to prepare older students for a workforce where automated tools are increasingly prevalent.

Those student anxieties match broader public polling. In the annual Chapman University Survey of American Fears, artificial intelligence entered the top ten national worries, with roughly half of respondents, about forty-nine point seven percent, voicing concern about economic harm.

Survey creator Christopher Bader noted that as automated tools integrate deeper into daily life, public apprehension is likely to remain a fixture of national discussion.

Pushback is also mounting against the physical footprint required to run automated software. Technology critic and podcast host Paris Marx is publishing a new book examining Big Tech data empires, as community opposition to server infrastructure expands.

In an interview with WIRED, Marx explained that data center construction has emerged as an issue in United States elections, following earlier resistance in countries like Ireland and Chile. He noted that municipal moratoriums on computing facilities stretch back to 2019 in the Netherlands and Singapore, followed by Irish electric grid scrutiny in 2022.

Marx highlighted local resource controversies, such as disclosures in The Dalles, Oregon, revealing that Google was consuming twenty-nine percent of the municipality's water supply. He argued that the surge in generative software has sharply accelerated computing demand, bringing localized environmental and utility pressures to more communities.

On the software development side, open model creators are focusing on training efficiency. Developers at Nanbeige have released a new mid-sized open-weight model designed specifically for autonomous programming and workplace tasks.

The developers describe the system as a flagship teacher agent. The plan is to use this larger model to guide the distillation of complex task execution into much smaller, lightweight models that cost less to operate.

The release includes controls that allow users to toggle step-by-step reasoning on or off and preserve thought history across multi-turn chats. In self-reported evaluations against competing models of similar scale, Nanbeige claimed strong results across workplace benchmarks, though it lagged behind rivals on specific real-world coding evaluations.

Finally, open models are making headway into biological research. Hugging Face, working alongside China's Zhongguancun Academy and Italian genetic research institutes including the University of Naples Federico II, has released Carbon-X, an open model built for genetic sequence analysis.

The system processes genetic data and natural language within a single architecture, mapping six DNA base pairs to each token. This allows researchers to interact with genetic sequences using conversational prompts, asking the model to predict whether mutations are pathogenic, identify gene control regions, and explain its reasoning in English.

Hugging Face reported that the model outperformed other tested open systems across five benchmark evaluations. Because the system can run on a single eighty-gigabyte graphics processor, laboratories can process sensitive genetic sequences on local machines rather than uploading confidential research data to third-party cloud services.

That's A.I Voices for today. I'm Maya. Thanks for listening.

Follow the show on Spotify, YouTube, Apple Podcasts, Amazon Music, or your favourite podcast platform, and explore the original voices and sources in the episode notes. See you tomorrow.