Tech Brew keeps business leaders up-to-date on the latest innovations, automation advances, policy shifts, and more, so they can make informed decisions about tech.

Tech Brew

Patrick Kulp

techbrew.com

Accenture’s chief AI officer thinks every C-suite needs a job like hers

Lan Guan is leading the consultancy’s efforts to harness a huge wave of AI interest from the business world.

A video of a sailboat moving down a river generated with Runway's Gen-2

Video generated with Runway's Gen-2 of a "person running on a track."

Video generated from the prompt "cat running on the street."

Is video the next frontier for generative AI?

AI startups want to make inroads in Hollywood, but the technology still has its limits.

Persistent AI hype fails to buoy startup market in Q1

PitchBook data shows a sluggish market for venture capital investing and fundraising.

Tricia Crimmins

In the nationwide treasure hunt to secure more geothermal power, some developers have pivoted from siphoning energy out of traditional geothermal wells to creating geothermal wells in parts of the country where they exist deeper under the surface. But where other developers zig, Zanskar zags.

The Utah-based geothermal developer uses AI to discover overlooked energy in geothermal wells that were previously thought to be tapped out, and thus far the company’s found success. Earlier this year, Zanskar announced it had used its AI models and new drilling techniques to harvest more power from a geothermal field in New Mexico that had been abandoned by the rest of the industry. In an interview with Tech Brew, co-founder and CTO Joel Edwards said Zanskar’s AI models have found more conventional geothermal hotspots in three years than the entire industry has been able to in the last three decades.

Edwards said the company predicts where geothermal resources are located and teams go out to test the areas for underground heat. Data pertaining to the area—whether it’s a hotspot or not—is then fed into an AI model, alongside datasets on rock types, fault lines, and the planet’s gravitational field. Eventually, the model creates a digital twin of earth’s geothermal system, learns how to identify geothermal characteristics, and can understand layers of physical data that can be tough for humans to grok.

“Our regional models are now better than humans,” Edwards told us. “Our models are now predicting where the hotspots are, and our field teams are out testing those predictions.”

And that tipping point, when the AI became better than humans at predicting where geothermal resources are located, happened in the last year, Zanskar co-founder and CEO Carl Hoiland told Tech Brew.

Because Zanskar is focused on traditional geothermal wells, or hot springs, the developer’s focus is in the western US. Hoiland said the magnitude of energy Zanskar will be able to access will mean the market will move “demand to where the cheap energy is, rather than the other way around.”

“As these sites get bigger, you can just bring a data center to that site,” Hoiland said. “Conventional [geothermal] should not be overlooked. There’s enough there to meet all of the future growth in energy demand.”

Zanskar is using AI to get more out of existing geothermal hot spots

The developer trains AI models on geothermal hot spot data, and AI directs humans to physical sites.

As Silicon Valley’s biggest city, San Jose is a ground zero for the AI industry. It produces double the machine learning and AI patents of San Francisco—as its mayor, Matt Mahan, seems to relish pointing out. Nvidia regularly 

But like the rest of the Bay Area, the city doesn’t build nearly enough housing to keep up with all this demand. Is it any surprise, then, that this tech-brained city sees AI as a possible answer to this dilemma?

“We have a huge shortage of housing, and we need to get to ‘yes’ faster,” Mahan told Tech Brew.

A former tech entrepreneur himself, Mahan has been adding AI to everything from bus routes to city services. The city’s latest mission is an AI pilot program that would cut down on the time it takes to approve building permits, starting with accessory dwelling units (ADUs), or smaller homes built alongside single family residences.

More than nine in 10 ADU permit applications are returned because they’re incomplete or incorrectly filled out. “It’s this game of ping-pong that often happens over email, and it’s very slow and inefficient,” the mayor said.

The city believes AI can speed this process by pre-checking applications before submission, cutting down on approval periods and sparing city employees from spending their time on email tag. Mahan said it will be a model for other types of building permit applications if it’s successful.

“That can shave days, if not weeks, off of ADUs, which are a little bit simpler, but you can imagine that extending to other forms of development—multi-family housing, commercial and industrial use, and a whole range of other types of permits,” Mahan said.

“If somebody is building something that complies with our general plan and our building codes, we need to get to a place where they have permission to go build it and they’re in conformance with all of our rules faster,” he added.

It might be particularly severe in places like the Bay Area, but the housing shortage is a national problem. Politicians in both parties have taken up the mantle of speeding development, and cutting red tape around housing construction is often seen as a key piece of this mission.

Cities like Austin, Honolulu, and Los Angeles have also embarked on pilots to test expediting various permitting processes with AI, according to 

 At least one agency in each of these cities belongs to the GovAI Coalition, a platform Mahan spearheaded for municipalities to review various tools and share notes on what works when it comes to implementing AI in local governments. The coalition counts around 700 public agencies as members, per Mahan.

“The whole idea is to leapfrog off of other agencies’ testing,” he said. “So while we’ve been leading the way in object detection by putting sensors on city vehicles, other cities should not have to start from scratch; they can leapfrog off of what we’ve done there. We hope to be able to do that for other cities.”

Having previously co-founded a startup with Napster and Facebook’s Sean Parker, and shared a dorm with Mark Zuckerberg at Harvard, Mahan said San Jose’s broad and diverse user base is the biggest difference from the startup world. He cited a program to use AI language translation to engage San Jose’s large Vietnamese community as an example.

“We need tools that make all of our services and processes as universally accessible as possible. It’s why language translation has been such a breakthrough for us,” Mahan said.

How the city of San Jose is tapping AI to tackle the housing shortage

It’s one of many ways the mayor has been implementing the technology.

Nearly three years after ChatGPT’s release, it seems like there’s new AI products everywhere you look.

But consumers have recently begun to latch onto some clear favorites. That was the biggest takeaway from Andreessen Horowitz’s 

 of the top 100 GenAI consumer apps, in which the venture firm ranked AI-native products with traffic data from Similarweb and Sensor Tower.

Among web products, this fifth edition of the list contained fewer newcomers than the same ranking assembled six months ago (11 versus 17). And in certain popular categories, like AI assistants and creative tools, the top handful of apps are starting to grow entrenched, Andreessen Horowitz partner and list author Olivia Moore told us.

“We are at the phase of seeing what I might call some early winners—or at least early leaders—in a couple of the categories where the models have been ready from day one,” Moore said. “And so that would be the general LLM assistants, like the ChatGPT, Claude, Perplexity group. And then also the creative tools—Midjourney, ElevenLabs, Photoroom.”

ChatGPT and Gemini topped both the web and mobile apps lists, followed by DeepSeek, Grok, and Character.ai on the web, and art app AI Gallery, ByteDance AI app Doubao, and Microsoft Edge on mobile. The venture firm also highlighted 14 apps that have appeared on all five of these lists that it has published, including ChatGPT, ElevenLabs, Hugging Face, QuillBot, Poe, and Character.ai.

 Those stalwarts tend to have one of two things going for them, Moore said: They’re either from a big-name lab, with the recognition and “aura” that goes along with that, or they’re startups that have come to own a specific use case, like (a16z-backed) ElevenLabs with voice generation.

“A lot of people thought…‘OpenAI is going to develop its own text-to-speech models and crush [ElevenLabs].’ And what we’ve seen is that OpenAI and Google don’t have time to make the best models for everything,” Moore said. “And so a lot of companies are able to get traction when they focus more specifically.”

“Then also, we are starting to see some moats. With Eleven, there’s a little bit of a brand moat in that they’re kind of the name in text-to-speech. And so every other product that pops up gets compared to them, which helps them continue to keep the crown. And then some data network effects—as you have more users, you can train the model to be better, which then gives you more users.”

Here are some other trends Moore noticed in assembling the lists:

 Google’s consumer AI products have come into their own with Gemini, Google AI Studio, NotebookLM, and Google Labs all cracking the top 50 web products. But Moore said the search giant is still figuring out how to best market its AI apps to consumers.

“For such a historically great consumer company, they still seem to be struggling with exactly how to package each of the products in a way that normal people find them usable,” she said. “There’d be more of them, and they’d be ranked even higher, if they can figure out the UI.”

 Blocks on American AI apps in China, as well as Chinese teams moving abroad, have helped boost a number of Chinese apps into the rankings, including ByteDance’s Doubao, Alibaba’s Quark, and Moonshot AI’s Kimi, according to Moore.

 apps like Replit and Lovable have grown in popularity this year, while AI-powered personal productivity—answering emails, generating slide decks—has also attracted more consumers, according to Moore.

“We’re finally starting to see AI productivity emerge,” Moore said. “We’re still early, but I would expect to see even more of those names on the list next time, and we’ll definitely be investing there.”

OpenAI, Gemini, DeepSeek and Grok icons on pedestals. 

Who’s winning the consumer AI race?

An a16z partner tells us what it takes to stay on top.

 recently confirmed what economists had already suspected: The labor market has not been doing so hot lately.

Whether that big summer hiring slowdown has anything to do with AI disruption is still somewhat of an open question, however. Some recent data suggest that the technology’s impact may be disproportionately hitting younger 

. Other reports aren’t clear on whether AI has much of a detectable effect.

“It’s very difficult to disentangle, you know, what is the impact of AI, compared to what is the impact of the uncertainty in the economy in general,” Columbia Business School professor Stephan Meier said.

Some recent studies on AI and jobs attempted to tease it out:

 from Stanford University traced a 6% decline in employment for early-career workers—those between the ages of 22 and 25—in fields considered most exposed to AI since 2022. Those jobs included software developers, customer service reps, and accountants. The findings are “consistent with the hypothesis that the AI revolution is beginning to have a significant and disproportionate impact on entry-level workers,” the economists wrote.

The above study was based on an analysis of ADP payroll data. ADP itself also issued a 

 a jobs slowdown in August, with “AI disruptions” pinned as one possible culprit for “whipsawed” effects.

Before too much panic sets in, however, other reports paint a different picture. A regional survey from the Federal Reserve Bank of New York 

 in the number of businesses using AI—40% this year versus 25% last year. But only 1% of firms had conducted layoffs driven by AI in the last six months, and just 12% of the companies using AI had hired fewer workers because of it.

 of the labor market released last month by the Economic Innovation Group found no meaningful impact on the labor market when it comes to AI. Unemployment among recent graduates has been “creeping up,” but not among AI-exposed jobs in particular.

If you don’t know exactly what to make of all of that, you’re not alone. Meier, who has been studying AI workplace transformation, told us that while AI will definitely automate some skills that are needed in offices, it’s unclear whether that will amount to offices needing fewer people overall: “That I don’t know yet,” he said.

It is plausible, though, that more of those replaceable skills might fall to entry-level employees, which would make it more difficult for young workforce entrants to get a foot in the door, he said. That would not only leave a generation with diminished job prospects, but also eventually harm the talent pipeline for companies.

“If you look at what positions within a hierarchy AI probably affects the most—or most immediately—it’s those entry-level jobs that are going to be affected probably sooner than later,” Meier said. “I’m not sure we see massive, massive effects of that yet, but we’re seeing the early signs that entry-level jobs are going to be affected.”

Meier said he is optimistic that the labor market will weather AI transformation in the long run with the creation of “new jobs, new industries, and new demand.” But in the shorter term, things could get messy as educational institutions and organizations struggle to keep pace with the change.

In that environment, it can be especially hard for young workers to set themselves up for a steady career.

“I see it with my teenagers—they’re as uncertain as everybody else, and obviously it hits them really, really hard,” Meier said. “When you just start a career and you don’t know what skill is going to be marketable, that’s scary. That uncertainty for sure creates a lot of anxiety. And I think organizations and society as a whole have to think about what to do with that.”

Close up of a briefcase with AI abstract connected lines displayed over the key lock.

Is AI actually hurting entry-level jobs?

It’s hard to tell.

Amid Trump administration cuts to federal science funding, one government program is still pushing forward with a mission to democratize AI research.

The National Science Foundation (NSF) plans to spend $35 million annually for up to five years to establish an operations center for the National AI Research Resource (NAIRR), in a move that could further cement what’s currently a pilot program. The NSF 

 early last year in a bid to give researchers more access to computing resources through partnerships with top tech companies.

For decades, many of the biggest breakthroughs in AI came from academia. But in the ChatGPT era, the cost and resource demands of creating a massive cutting-edge model have ballooned, and the scales have tipped decisively toward big, closed-off companies that can better afford such undertakings.

“Today, no university in the world can build a frontier AI system on par with industry,” the authors of a Stanford University 

That disparity might limit AI developed in the public interest and close doors to research questions that can’t be readily commercialized, the authors said. Academics like “AI godmother” and Stanford computer science professor Fei-Fei Li have been 

 for a national research cloud of this sort for more than half a decade now.

, a California Republican, and Rep. Don Beyer, a Virginia Democrat, reintroduced a bill called the CREATE AI Act that would formally establish the NAIRR, but it hasn’t passed despite bipartisan support. President Trump also supported the NAIRR as part of his 

Since the NAIRR pilot began, the NSF said it has supplied more than 400 research teams with resources like “computing platforms, datasets, software, and models,” according to the announcement. The Operations Center is intended to help grow the NAIRR through a new “operations strategy and management structure” and a centralized web portal to make it easier to access.

“The NAIRR Operating Center solicitation marks a key step in the transition from the NAIRR Pilot to building a sustainable and scalable NAIRR program,” Katie Antypas, director of the NSF Office of Advanced Cyberinfrastructure, said in a statement. “We look forward to continued collaboration with private sector and agency partners, whose contributions have been critical.”

 Brandi Geurkink, executive director of the Coalition for Independent Technology Research, said government-sponsored computing resources like NAIRR and similar state-level programs can be helpful for researchers as long as measures are put in place to ensure the independence of the work. The CITR recently released a report on the state of independent tech research, in which 85% of respondents said lack of funding was a top barrier.

“When I first saw these pilots, the question that came to my mind was really about how independence is still ensured, because we talk a lot in the report and in the coalition writ large about independence from corporate interests, but independence from governmental interests is another really big part of researcher independence,” Geurkink said. “And so when pilots, for example, are funded by the federal government, funded by state governments, then it still raises that question around independence. Which isn’t to say that no funding should ever come from state or federal governments, but there need to be safeguards in place to ensure that that independence can be preserved.”

A government resource for AI research will get a new operation center

The NAIRR could help academia better keep up with industry.

Extreme weather is becoming a more frequent part of day-to-day life, and long-range forecasting can help governments and businesses alike prepare for severe weather.

Planette’s subscribers are usually businesses with physical assets that might be environmentally vulnerable. But now, the company is working with NASA’s Goddard Institute for Space Sciences to forecast extreme weather using AI and quantum computing algorithms.

In an interview with Tech Brew, Planette CEO Hansi Singh said the company trains AI on existing models of Earth, like those used by the National Oceanic and Atmospheric Administration, that contain the physics of how the ocean and atmosphere interact with each other. Then, its AI is able to forecast extreme weather up to six months in advance.

“If you invest in the training [the AI], then the actual inference where you produce the forecast, either daily or weekly, is super, super, super cheap,” Singh said. “That is the advantage of the AI, because you can learn the physics, but you can run it extremely efficiently.”

Through the company’s partnership with NASA, Singh said they’re working on QubitCast, which will forecast “one-off events”—like extreme weather—using “quantum algorithms that are even more efficient” than AI algorithms.

“That analysis would really help increase the time horizon over which you can see that an extreme event has a certain probability of occurring,” Singh said. “My goal is to make it so that every agency, every business that has some environmental vulnerability is looking at forecasts like this, knows what’s ahead, and is preparing adequately.”

QubitCast is still in the research and development stage, as is another AI forecasting model the company is working on with the National Science Foundation. Singh said she’s excited about the progress on both projects, especially because they’re focused on adapting to the effects of the climate crisis.

“Depending on how fast we get our act together—or don’t get our act together in terms of the energy transition—that adaptation is going to be huge and expensive and potentially painful,” she said. “To minimize that painfulness, we really need to be investing in adaptation technologies.”

How Planette helps NASA use quantum principles for weather forecasting

The long-range weather forecasting company trains AI on existing physics-based models of Earth.

Jordyn Grzelewski

Back in 2023, BMW was widely panned for its plan to offer heated seats as a paid subscription.

Steve Basra, Google Cloud’s global head of automotive, believes that the tides have turned since then on what consumers are willing to pay extra for, upending automakers’ plans to create new subscription-based revenue streams.

“Subscription fatigue is definitely a real thing after Covid,” Basra said. “Customers are less willing to pay for things that they believe they should just get when they buy the second most expensive thing in their life.” After a home, that would be a new vehicle, which costs an average of nearly $50,000 in today’s market.

But Basra is bullish on automakers using AI to personalize and improve the in-vehicle experience, thereby creating a value proposition consumers can’t deny.

“A lot of the transformation hasn’t been achieved, and I think AI is the one technology that is going to transform the industry,” Basra said of the industry’s long-hyped transition to EVs and 

Tech Brew recently caught up with Basra, an auto industry veteran who spent 25 years at Toyota, at Google’s Detroit office, where he explained how Google Cloud is helping automotive clients implement AI across their value chains.

The auto industry’s product planning and development cycles are notoriously long, historically taking several years to bring a new vehicle to market. This is one area where Basra pointed to the potential for AI to help automakers save time and money, citing EV battery research as an example.

“When you get a direction from the company that says, ‘We want to start working on the next generation of EV batteries,’ the first step [is] a lot of research. What are the Chinese doing? What are other competitors doing? What are the latest technologies? What are the latest patents? That can take a lot of time,” he said.

Google Cloud offers its Gemini Deep Research 

, which Basra said can collect, collate, and summarize data in ways he claimed could cut research time down by months.

The US auto industry collectively spends tens of billions of dollars per year on warranty claims. This nagging issue is another opportunity, as Basra sees it, for AI to help reduce costs.

Google Cloud’s Manufacturing Data Engine tool connects factory floors with the cloud, which Basra said can help automakers catch issues like battery defects and paint anomalies before vehicles get shipped out.

“Using vision systems at the point of manufacturing to make sure that the batteries are manufactured in the correct way is one area that we’re investigating with an OEM,” he said.

, manufacturers like GM say they’re implementing AI to make their plants safer and more efficient, and to improve quality. One representative application uses AI to inspect weld and paint jobs.

Then, of course, there are the customer-facing AI use cases. Google Cloud’s Automotive AI Agent is being used by automakers including Mercedes-Benz to improve voice control and act as in-vehicle assistants.

Google Cloud revealed the product in January, describing it as “a new way for automakers to create helpful generative AI experiences” and to let automakers “create highly personalized and intuitive in-car agents that go beyond current vehicle voice control.”

And last year, Volkswagen announced that, in partnership with Google Cloud, it was introducing a GenAI-powered virtual assistant designed to give drivers “intuitive access to critical vehicle information and services.”

Basra posited that an AI agent could improve the in-vehicle experience for drivers, for example, by automatically opening the trunk when they approach with shopping bags in hand or notifying someone they’re meeting for lunch that they’ve hit traffic and are going to be late.

“You’ve got an assistant that sits with you in the car, that takes care of all that hassle,” he said. “That is the experience, I believe, that people want, and that’s the experience they expect from these agents.”

The transition to EVs and software-defined vehicles may be taking longer than expected, but Basra said he views AI as something that will force legacy automakers to catch up.

“I think without AI, they would have really struggled against the Chinese OEMs and the new OEMs that are coming in,” Basra said. “This gives them the opportunity to compete with these companies.”

A photo composite of a manufactory worker, a car panel with a screen on the dashboard and a table of people planning overlaid with binary code. 

AI