Day: November 19, 2025

• Cohere CEO Aidan Gomez said enterprise AI hasn’t really even scratched the surface.
• “We’re still doing the super foundational,” Gomez said.
• He said that once models improve, greater disruption will follow.

AI has already disrupted coding. Finance could be next, according to Cohere CEO Aidan Gomez.

“Finance will come,” Gomez told Kleiner Perkins partner Joubin Mirzadegan on a recent episode of the firm’s “Grit” podcast.

Cohere, a Toronto-based AI startup focused on enterprise applications, is already doing some work in the financial sector through partnerships with RBC and TD Bank. Both partnerships are built on Cohere’s LLM models. After it closed its latest round of funding in September, Cohere was valued at roughly $7 billion.

Gomez, who co-authored the famous 2017 paper “Attention Is All You Need” while he was an intern at Google Brain, said that AI’s power lies in its ability to augment white-collar work, a view widely shared throughout Silicon Valley and the AI industry. Anthropic CEO Dario Amodei has said that he’s worried that AI could wipe out up to half of entry-level, white-collar jobs in the next one to five years.

“You have to pay them a lot because there’s not a lot of these people for the world,” Gomez said. “And so there’s tons of demand for these people, but there’s not enough of those people to do the work the world needs. And it turns out that these models are best at the types of things those people do.”

AI’s effects are already being felt in software engineering, where powerful tools like Cursor, along with frontier models like Claude and ChatGPT, allow users to “vibe code” software without any specialized coding knowledge. Other industries, including finance, legal, and publishing, are still in varying degrees of the early days of AI disruption.

“It’s still so early for the enterprise,” Gomez said. “We’re still doing the super foundational, super — Summarize this email for me. Summarize these meeting notes for me. It’s so basic, low-level. I just think there’s so much to be done.”

Mirzadegan said the next step of AI is the ability to “augment people.”

“I mean, isn’t that the whole point of this? We’re eating a different part of the market where it’s not seats and licenses,” he said. “It’s being able to actually augment people, actually do the jobs of people in many ways, and that’s so much more transformative.”

• Pat Santiago moved from Pittsburgh to San Francisco to launch an AI startup with his cofounder.
• They ended up managing hacker houses and focusing on physical community-building.
• Despite challenges such as moving locations, fostering real-world connections drives his success.

This as-told-to essay is based on a conversation with Pat Santiago, a 28-year-old cofounder of Accelr8, based in San Francisco. The following has been edited for length and clarity.

Last year, I moved to San Francisco to cofound an AI startup. Living in a hacker house seemed like the best way to make as many connections as possible.

My cofounder and I had an idea for a no-code AI workflow builder and planned to stay in San Francisco for three months, but we decided that running our own hacker house could be more fun. It would also help us build a strong network.

Soon after we decided to move from the separate cities we were in, we began searching for potential hacker houses that we could rent and run ourselves. The operational stress can be intense at times, but I’ve now got a better grasp on managing it.

We found our first hacker house on Airbnb

I met my cofounder in 2023 because we were both working at Decentralized Autonomous Organizations. Those projects fizzled out, and we went on to do other things.

I moved from my hometown of Pittsburgh, and he was in Dallas. San Francisco was always a place I wanted to live because it seemed like the kind of place where people who thought like me were.

We planned to share a bunk bed in one room, and fill the rest with people we found, possibly securing free rent for ourselves in the process.

We looked on Airbnb and found a place that had been converted from a five-bedroom house to a 15-bedroom house. After speaking with the owners, it seemed like it would work. They said we had two weeks to close on it if we wanted a master lease for all the units.

We messaged people on LinkedIn until we filled all the spots in the house

We were frantically sending messages to people on a San Francisco housing directory, LinkedIn, and X. We thought we could definitely convince 15 people to do this in two weeks, and it would be worth it if we did.

We scheduled around 100 calls to fill the first cohort. We made the decision to start the house at the beginning of June. By July 4, we had filled every spot, but some people dropped out, so we continued interviewing until the first move-in day on July 15.

One of my most memorable experiences was when the first group arrived at the house last year.

The pros of running the hacker house

What surprised me the most about moving here was the level of optimism; people are generally optimistic by default.

For San Francisco, the money I’m making from this isn’t great, but it’s livable. Ideally, as we expand to more buildings and locations, I can create a better financial situation for myself. But for now, it’s nothing to complain about because I’m so inspired by doing this.

The biggest motivator right now is seeing the impact our alums make in the world. They’re doing some really cool stuff, and I’m constantly getting updates that make me smile.

At the end of summer, we dropped to 12 residents, but I wasn’t really panicking because I now know this is just the natural flow of the year, and we’ve picked back up into the 40s.

The biggest challenge was when we were given a 2-month notice to vacate the first house

Our first house was taken over by a new management company, and they gave us a two-month notice to leave. When I looked for our next spot, I procrastinated because I wasn’t even sure if I would commit to managing this house for another year. But I had the hypothesis that a lot of hotels here are vacant, and if I called them, we could get a good deal and fill one.

After we found a building that would work with us, within a day of posting the first advertisement, we had 80 applicants. It was the reassurance I needed because I thought, ‘Okay, this is going to work. People love it.’

We priced a private bedroom at $1,500 a month, and that base price has remained consistent, except that we now offer larger rooms or suites as additional options.

Building a strong community is the key to success here

The house does a lot of community events because when we got here, we were going to networking mixers that droned on. Then, I was sitting in the backyard, complaining about just that, and I had the idea for a Mark Zuckerberg-themed rave.

Everyone was cracking up, but then we threw a Mark Zuckerberg-themed rave, and there were no name tags in sight. No one networked, and it was a blast. I was very proud that I’d sparked that.

People try to make community a digital phenomenon, but I think those who are literally shaping the future should be coming together and being physically present with each other.

Community used to mean gathering in a bar, a social club, a church, or somewhere physically present with other people; however, there’s a lot less of that in the world now, with the rise of AI and technology.

My major piece of advice is just to be nice to the people you meet and make friends. Doing community events is what keeps me engaged. Leading a group of people out for the night, to the park for a day, or on a hike, I can see the magical feeling I felt when I first moved here through their eyes, and I really love that.

Do you have a hacker house or founder story to share? Contact this reporter, Agnes Applegate, at aapplegate@businessinsider.com.

• AI is changing the nature of search and rewriting the rules for brands.
• It has spawned a cottage industry of AI search experts, though skeptics warn of overpromising.
• The old rules of search still apply in this era, major platforms say, but there are key differences.

This summer, the PR agency Bospar was preparing a major announcement for its client, the AI computer vision company RealSense, when it discovered an awkward problem.

When Bospar asked ChatGPT, Claude, Gemini, or Copilot for information about RealSense, each one provided a version of the same answer: RealSense was no longer in business.

In reality, the company was on the verge of announcing a spin-out from its parent, Intel, and a $50 million funding round. How could RealSense get top-tier press coverage if the leading AI tools had already written its obituary?

“There’s no 1-800 number for ChatGPT if there’s an error,” said Curtis Sparrer, principal at Bospar.

About half of US consumers are using AI-powered search to evaluate and discover brands, per a recent McKinsey report. In this new era, Google — which has introduced its own AI search features — still dominates the overall search market, responsible for roughly 90% of global search engine traffic, according to Cloudflare.

As the nature of search begins to shift, businesses from travel companies to clothing brands are racing to ensure they show up prominently — and accurately — inside these answer engines.

The trend has spawned a cottage industry. Former experts on SEO (search engine optimization) are now declaring themselves gurus of GEO (generative engine optimization) and AEO (answer engine optimization).

LinkedIn and Meta feeds are flooded with ads from startups, agencies, and consultants saying they have a formula to boost visibility on top AI platforms. There’s a huge market for the taking: only around 16% of brands systematically track their AI search performance, according to McKinsey.

The search shifts have also sparked a fierce debate: Can GEO and AEO experts actually deliver on their promises?

“Everyone is going crazy about becoming the next agency — all the side hustlers and snake oil sellers with their tools already on the train and riding the hype,” said Kai Spriestersbach, an applied AI researcher, web scientist, and SEO veteran.

Business Insider spoke with SEO experts and three of the top AI search platforms — Google, Microsoft, and Perplexity — to investigate what content optimization looks like in the age of AI. OpenAI did not respond to a request for comment.

The takeaway: Some GEO techniques can help brands boost their visibility in AI-generated answers, but these gains may be fleeting in a fast-changing space where models are constantly retrained and updated. In some ways, GEO is a new label for companies’ existing brand-building strategies, which involve a mix of maintaining good website hygiene, a solid PR strategy, and traditional paid advertising to help boost consumer awareness.

One AI visibility tracking tool, Lorelight, recently shut down, with its founder saying that there was no such thing as a “GEO strategy” separate from brand building, at least for large companies.

“There’s going to be a lot of people who benefit from implying that they’re very good at GEO. And then there’s going to be a lot of tech companies, specifically AI companies, who benefit from saying that it’s all hogwash,” said Jesse Dwyer, head of communications at Perplexity. “The reality is always somewhere in between.”

GEO vs. SEO

SEO, which is largely focused on Google search, refers to the practice of designing web pages and securing links from quality sites to help your brand rank highly on search engine results pages.

The SEO industry is built on more than 25 years of study, experimentation, and some disclosures from the search giants themselves about how the ranking algorithms work. But there are key differences between GEO and SEO.

For starters, the GEO world has far less historical data to build its assumptions on. For traditional search, practitioners can draw on tools like the Google Search Console to see how their sites appear in search results and Google Trends to identify trending terms. AI platforms keep this sort of data much closer to the vest, beyond the occasional blog post, such as a recent study from OpenAI that found people were largely using ChatGPT for everyday tasks, including help with writing and tech support.

“You don’t know how people are looking for brands and services right now,” within AI chats, Spriestersbach said.

There’s also the personalization factor. The front page of traditional search generally looks the same for everyone, though Google offers options to tailor results. Now, companies must try to make their brands shine in unique and deeper conversational searches. Answers can vary wildly for different users based on how they’ve configured their preferences and their chat histories.

Further complicating things: Platforms are regularly retraining their models, which can dramatically change how a brand shows up from one day to the next, said Tim de Rosen, cofounder of GEO auditing firm AIVO Standard.

While many GEO companies offer services like dashboards that analyze a subset of user prompts, most of what’s being measured isn’t reproducible. The different ways users ask questions and how models respond — plus the data they’re drawing from — are all in a constant state of flux, de Rosen said, which makes results “inherently unstable.”

The platform view

Major platforms say the key principles of SEO still apply in the age of AI. Similar to SEO, AI systems rely on fresh, highly ranked, and trustworthy content, said Krishna Madhavan, principal product manager for Microsoft Bing.

“Be skeptical of shortcuts,” he said.

Danny Sullivan, director at Google Search, said any GEO tools that advise designing content solely for rank and visibility purposes lose “track of the big picture.”

“Are you doing things that are useful for human beings?” he said. “That’s what we want to reward.”

Sullivan shared an example of popular advice from SEO/GEO experts that could soon go stale: that large language models favor bite-sized content.

“Maybe they’ve seen that this seems to work in some edge cases in some places,” but the model will inevitably change, he said. “All that work you did to please the system may not carry through to the long term.”

What is new, Madhavan of Bing said, is that companies must think about optimizing for inclusion in a synthesized answer, rather than simply a list of links on a search results page.

“Think beyond keywords to user intent, question‑answer structure, and machine‑readable cues that make your content easy to parse,” Madhavan said.

How companies are navigating GEO

As the GEO industry takes shape, companies are proceeding with caution and seeking guidance from multiple practitioners.

Vineet Mehra, CMO of fintech company Chime, said marketers often make the mistake of working with one tool or agency.

“We try to use multiple companies to create a little bit of competition,” he said. “You see who’s going to customize their product road map for you, who is innovating faster than the other.”

The PR agency Bospar also determined that bringing together a patchwork of solutions was better than relying on a single consultant or GEO platform. They ultimately traced RealSense’s premature “death” to a 2021 news article that had misrepresented a restructuring, which then snowballed when it was discussed on platforms like Reddit — an important source of content for training large language models.

Bospar sought a correction from the original publication and made changes on RealSense’s website, including an FAQ section addressing the closure rumor. The agency also encouraged RealSense executives to actively participate in trending robotics and AI conversations on social media to demonstrate their thought leadership.

While AI visibility tools like dashboards are in their infancy, Bospar’s tactics should be familiar to anyone handling a company’s reputation in the digital era.

“It’s pretty early days in GEO and AEO for anyone to raise their hands and declare themselves as an expert,” Bospar’s Sparrer said. “I think that’s a little rich at this point considering how new the science is.”

Here are some key pieces of GEO advice from Google, Microsoft, and Perplexity

Google:

Sullivan said the core principles of SEO generally apply to new forms of AI search.

“Plenty of sites succeed because they don’t do SEO, or hire SEOs,” Sullivan said. “They don’t think about it because they’re just focused on making great content.”

General website and structured data hygiene always make sense, ensuring Google’s search crawlers can actually get to the relevant content, Sullivan said, especially as AI answers still have a lot of traditional search results at their core.

AI formats are also getting better at multi-modal queries — a user might ask “what’s in this video?” for example.

“If you’ve still been a text-only kind of player, more images and videos may help you. But they would have helped you anyway,” Sullivan said.

Microsoft:

Madhavan said the fundamentals of SEO are still critical, including structure and freshness signals that make content easier for AI to consume. This includes using Q&A sections, sitemaps, and schema, a code that helps search engines understand your site, as well as adopting IndexNow, a protocol that lets search engines know when your site changes.

Stylistically, Krishna also suggests lists and tables instead of long walls of text, and advises keeping punctuation simple, including avoiding em dashes and symbols.

Perplexity:

With the shift from SEO to GEO, “the biggest mistake you can make is to just try and transfer your understanding apples to apples,” Dwyer said — and a lot of the companies offering GEO services are doing just that.

Dwyer said he’s also been advising marketers that AI search will shift budgets toward old-fashioned brand marketing. AI removes the “friction” of search and lets people buy things just by asking for them. As a result, building a strong brand will become increasingly important, he said.

• Two MIT dropouts have raised $2.7 million for police tech startup Code Four.
• Code Four uses AI to generate reports from bodycam footage.
• Here’s the pitch deck the Y Combinator graduates used to raise their seed round.

Two 19-year-old MIT dropouts have raised $2.7 million in seed funding out of Y Combinator to arm police officers with AI.

CEO George Cheng and CTO Dylan Nguyen cofounded Code Four, which is police radio lingo to indicate that a situation is under control. They say their technology reduces paperwork, allowing officers to spend more time in the field.

The startup uses AI to generate reports from bodycam footage that can be used in court or for record-keeping purposes. Code Four can also redact footage and reports for records requests and generate transcriptions and summaries from video interviews and security footage.

“The public safety and the government space is always kind of behind the curve in terms of getting the latest technology, even though these folks are some of the hardest working and most passionate about their mission,” Nguyen told Business Insider.

While the duo knew they wanted to build in an impactful sector, they are aware of the concerns surrounding the use of AI in policing. Cheng said that while Code Four uses AI to generate preliminary drafts of reports, officers review and edit them for accuracy.

Cheng and Nguyen met during high school on the international science fair circuit, and both ended up at MIT, where they decided to build a company together. Ultimately, they dropped out as freshmen to attend Y Combinator.

They also plan to participate in the second cohort of the Palantir Startup Fellowship next year.

AME Cloud Ventures led Code Four’s seed round, with participation from Pathlight Ventures and Webb Investment Network.

Cheng and Nguyen plan to use the funds to grow their team, which consists of four employees, split between engineering and sales.

Cheng said Code Four works with 25 police departments and makes money via a subscription model, which starts at $30 per officer a month.

Here’s a look at the pitch deck Code Four used to raise its seed funding. The deck has been edited so that it can be shared publicly.