Good morning. We’re a little longer than normal today, because there are just a few things that we need to talk about.

Yesterday, I happened to run into one of the first startup CEOs I ever interviewed after jumping on the AI beat — Suman Kunaganti, of Personal.ai — and he asked me what I think is the single biggest story in AI right now. And my answer had to do with poor economic signals and the ongoing stock market rout; without even considering the impacts on startup funding, available capital, investment, etc., a recession seems likely to impact enterprise adoption of AI.

It’s not yet clear how it will do so, but in an environment where capital becomes restricted, GenAI experiments might well become lower priority for businesses, or, as he pointed out, adoption might surge in the opposite direction.

For your reference:

$SPX ( ▲ 1.56% ) 

$NVDA ( ▲ 2.82% ) 

$TSLA ( ▲ 5.42% ) 

— Ian Krietzberg, Editor-in-Chief, The Deep View

In today’s newsletter:

• 🌎 AI for Good: Mapping water on Earth

• 📉  Mag7 stocks routed amid heightened recession risk 

• 👁️‍🗨️ Global AI race intensifies with Manus, China’s answer to Deep Research

• 💻 Interview: MapQuest, but for business decisions

AI for Good: Mapping water on Earth

More than one billion people live without reliable access to safe, clean drinking water, according to the World Health Organization. 

In 2017, British researchers figured that if we can use satellites to find water on other planets, we could probably use satellites to find water on Earth. Thus, the social good startup DeepWaters AI was born. 

• The startup combined specially designed neural networks with satellite data to find groundwater on Earth. And this was back in 2019, so well before the ChatGPT craze. 

• The neural networks were trained “to identify the spectral signature of water and backscatter propagation signature of increased surface moisture,” according to the European Space Agency. 

Why it matters: The team was able to use this approach both to create a map of the world’s underground drinking water, as well as to track and identify underground pipe leaks. 

How to Operationalize AI

AI is only as powerful as the processes behind it. Without the right orchestration, AI initiatives can struggle to deliver real business value.

In this on-demand webinar, experts break down how to seamlessly integrate AI into your workflows with process orchestration.

You'll discover:

✅ Real-world use cases for AI-driven automation

✅ How to connect AI models with human and system-based workflows

✅ Best practices for scaling AI with process orchestration

Don’t let AI initiatives stall at the proof-of-concept stage—learn how to turn them into real, scalable business solutions.

👉 Watch Now

Global AI race intensifies with Manus, China’s answer to Deep Research

A week ago, a Chinese company called Manus launched onto the scene around a product called Manus AI, a system that the company described as the world’s first “general AI agent.” 

The launch marks another sally in the increasingly tense race between the U.S. and China for dominance, not just in technology, but in artificial intelligence — and if you get your news exclusively from Twitter, the launch marks China’s “second DeepSeek moment.” 

But, as is often the case, the reality is simply more nuanced. 

Here’s what we know: Very little information was published alongside the release of Manus AI; in an introductory video, Yichao Ji, the company’s co-founder and chief scientist, unveiled a handful of demos of the system at work, parsing through resumes and ranking job candidates, conducting property research and analyzing and comparing stocks. 

• Ji said that Manus has been solving “real-world problems” on freelance platforms including Upwork and Fiverr, and “has proven its capabilities on Kaggle competitions.” 

• But demos are not indicative of performance, since they can be engineered behind the scenes; the details of this real-world problem-solving, meanwhile, remain unknown. 

Manus AI did not respond to a request for comment on the details of the system’s architecture, the sizes and types of models that make it up, the training data that was leveraged, the training techniques that were applied, the hardware quantity and type used, the electrical intensity associated with that and the architectures at play here. Again, there are a lot of unknowns. 

What we do know is that, according to Manus, the system performs quite well on benchmarks. On the GAIA benchmark to assess the efficacy of generative, agentic systems, Manus significantly outperformed OpenAI’s Deep Research; on the first level of the benchmark, Manus scored an 86.5% to OpenAI’s 74.3%. None of this has been independently verified.

“This isn’t just another chatbot or workflow, it’s a truly autonomous agent,” Ji said. “We see it as the next paradigm of human-machine collaboration.” 

The system is fully cloud-based at the moment, with Ji explaining that users can shut their laptops down and the agent will continue to work asynchronously in the cloud. He specifically noted that “Manus wouldn't exist without the open-source community,” adding that the company plans to open-source “several of those models later this year.” 

Again, it’s not clear what Ji means by “open-source,” here — openly releasing model weights is not the same as open-sourcing the system, which would require the general availability of training data and source code as well. 

Why it matters: Manus is currently in an early preview mode, meaning it’s not yet widely accessible. Some early users have said the system far outperforms OpenAI’s systems. Ånd while we don’t know the costs associated with the construction or operation of this model, it is another reminder that, at this stage in the race, there is no technical moat, something that surely has some U.S.-based VCs at least a little nervous right about now. 

Your Data Made Them Billions—Now It’s Your Turn to Profit

🚨Open AI reached a valuation of $340 billion this year, all on the back of your data. 

The big data gold rush is here, but profits have been hoarded by AI and tech giants… Until now.

Instead of just collecting data, Mode Mobile is sharing the profits from it, all by turning your smartphone into an income-generating asset. Here’s what that looks like:

• Paid over $325M to over $45M users.

• Generated 32,481% revenue growth.

• Previously ranked #1 fastest-growing software company by Deloitte.

📲Their EarnPhone could be the biggest disruption to smartphones since the iPhone, and they’re gearing up for a potential IPO on the Nasdaq (ticker: $MODE).

But the real opportunity is now, and you can still invest in Mode’s pre-IPO offering at just $0.26/share.

Earn up to 100% bonus shares by investing in Mode now.

• Acquisitions: ServiceNow said Monday that it will acquire Moveworks as part of its push to expand its AI portfolio. To do so, ServiceNow said it will fork over $2.85 billion.

• AI in Government: Elon Musk’s DOGE has expanded its custom GSAi chatbot to 1,500 federal workers, with the hopes of expanding to the entire General Services Administration, according to Wired. An internal memo reviewed by Wired instructed workers not to “type or paste federal nonpublic information.”

• Meet the 21-year-old helping coders use AI to cheat in Google and other tech job interviews (CNBC).

• Stock rout picks up steam with recession warnings blaring (Bloomberg).

• Musk may still have a chance to thwart OpenAI’s for-profit conversion (TechCrunch).

• What’s behind the recent string of failures and delays at SpaceX? (Ars Technica).

• OPM watchdog says review of DOGE work is underway (Wired).

Mag7 stocks routed amid heightened recession risk 

Following several weeks of market chaos — during which investors and economists alike have been trying hard to react to the tariffs that President Donald Trump first threatened, then rolled back, then put in place — Trump openly addressed the risk of recession on Sunday, saying: “look, we’re going to have disruption, but we’re okay with that.”

What happened: The market heard Trump, and — following three weeks of market declines that have hit the tech sector especially hard — sold off. The Dow Jones dropped 2%, the S&P 500 declined 2.6% and the tech-heavy Nasdaq retreated some 4%, its worst day since 2022.

• This selloff was led by the “Magnificent Seven” coalition of Big Tech stocks, with Tesla dropping 15% and Nvidia falling 5%. Microsoft fell 3.8% to notch a new 52-week low, Apple fell 4.8%, Google slid 4.5% and Amazon 2.3%. 

• As Deepwater’s Gene Munster noted, this week marks the 25th anniversary of the dot-com bubble, a somewhat dark omen considering the fact that “bubbles have historically burst just ahead of a recession, and the probability of a recession has increased measurably over the past month.” 

Munster said that his “optimism for the next couple of years has taken a step back” given the current economic environment, adding: “the bottom line is that if we enter a recession, it will be extremely difficult for the AI trade to continue. However, if a recession is avoided, the market is highly likely to reach all-time highs over the next couple of years, driven by the revenue and earnings benefits that AI will provide to a wide range of companies.” 

Wedbush analyst and tech bull Dan Ives acknowledged this in a Monday note, saying that he “underestimated” the market’s reaction to the “Trump Policy Bazooka,” though he does not think “this dramatically changes the trajectory of the AI Revolution over the coming years.”

As Munster noted, with the Nasdaq down 11% in the past 14 days, it is “arguable that the AI bubble has already burst.” 

Interview: MapQuest, but for business decisions

David Shim’s problem with AI summaries is that they lack context. 

A large language model (LLM) can turn a 30-minute meeting into half a dozen bullet points, but the model doesn’t know which bits of information it ought to include in that summary, or which bits of information don’t really matter. 

In 2021, Shim co-founded Read AI, one of many AI startups whose broad goal is business optimization. 

But Shim’s idea with Read is centered around deeper insights coming from a broader array of data sources, something that, he says, enables the context that you need to make AI summaries actually usable. 

The news: The company today unveiled Search Copilot, a search engine that sits at the center of an information highway that includes meetings, emails, messages, presentations and other extraneous workflows. 

• The in-house models Read built that enable this were trained to produce a deeper level of cross-platform behavioral analytics. The model, for instance, that records and summarizes a video call, constructs that summary based on engagement metrics. 

• “If you're nodding your head only on certain points, it's going to go in and adjust that summary to say ‘these are the points that were most relevant in the conversation,’” Shim said.

Read then dips into third-party LLMs to make the insights its in-house models develop readable. 

And while meetings are highly relevant to it, Shim said that “it’s not just meetings.”

If I take, for instance, four hours on average to reply to an email, but I reply to an email — hypothetically — from Sam Altman in five minutes (because you can’t leave Sam waiting), Read’s copilot would identify that sender, and that topic, as higher priority than others.

“So now the models can go in and say, ‘these are higher engagement emails. These are topics that are more relevant to you. When this topic comes up, you pay more attention to meetings, you reply faster to emails,’” Shim said. “All of those things start to stack, where these strong and weak signals start to go in and learn and say, ‘hey, this is what Ian cares about. This is what I'm going to prioritize in this inbox. This is the prioritization of the summaries.’”

Armed with this level of depth and breadth — in addition to traceability, since summaries include specific links to original sources — Shim said that, today, he has fewer meetings than he’s ever had in his career; he can send his copilot in his place, trusting it to identify the pieces of information that, based on his personal data, is most important for him to know. 

“It's really gone in and made — I wouldn't say it's totally made a TikTok slash Tinder from a decision-making perspective — but it is more of an influencer than I would have expected,” Shim said. “And that's kind of crazy when you think about it; it's not to replace anybody, because there was nobody in that role, but it's going in and saying: ‘hey, did you know this client had this feature request? We've gotten it a lot, but this one is a very high-value client, based on what you've said.’ And I get those notifications and I'm like, ‘Yeah, we should probably release that.’”

At this point, it’s become a well-established fact that the key to artificial intelligence is data. Read’s copilot is evidence of that. But for it to have access to that breadth of data raises a mix of privacy and security issues, problems that Read is looking to challenge head-on. 

• On the security side, beyond data encryption at rest and during storage, the company has also secured a Service Organization Control (SOC) 2 Type 2 certification, a result of regular audits and tests to ensure its security measures are robust. 

• And on the privacy side, by default, Read does not train its models on customer data. Customers have to opt-in to that option, a markedly different approach compared to many other AI/social media platforms. 

Further, Read’s copilots — when in meetings — are always visible as a participant, with links to the company’s terms of service and privacy policy automatically uploaded into the chat. 

“If you want us to stop, type, ‘Read, stop,’ and we will stop and leave the meeting,’ Shim said. “If you type in, “opt-out,” we'll delete all the data so it doesn't even exist anywhere, because we haven't fully processed it yet … You know when we're there. We're not trying to be sneaky. We're trying to be transparent.”

When the pieces all come together — cross-platform information summarization based on engagement metrics, dynamic search through that information and recommendations based on those analytics — Shim said that you have a system that’s on the verge of turning into a sort of influencer, a MapQuest for decision-making, if you will.

“Just like MapQuest used to tell you directions and now it's Waze telling you to go left instead of right, it's that same thing,” he said. 

🤔 Your thought process:

Selected Image 2 (Left):

• “In Image 1 it is rather unrealistic for the window-cleaner to be holding hands with his pal, as if they were having a cozy chat, ha,ha!!”

💭 A poll before you go

Thanks for reading today’s edition of The Deep View!

We’ll see you in the next one.

Here’s your view on AI Overviews:

A third of you think AI Overviews are extremely helpful.

But a third think it wouldn’t even be necessary if Google just fixed its normal search algorithms …

20% just don’t like it.

Yes, super helpful:

• “Keep it going :-)”

If you want to get in front of an audience of 450,000+ developers, business leaders and tech enthusiasts, get in touch with us here.

Mode Mobile Disclosures

Mode Mobile recently received their ticker reservation with Nasdaq ($MODE), indicating an intent to IPO in the next 24 months. An intent to IPO is no guarantee that an actual IPO will occur.

The Deloitte rankings are based on submitted applications and public company database research, with winners selected based on their fiscal-year revenue growth percentage over a three-year period.

Please read the offering circular at invest.modemobile.com. This is a paid advertisement for Mode Mobile’s Regulation A Offering.