Bloomberg Tech SF 2026: "We Don't Need AI to Get Smarter."

Two years ago, a gathering like this would have argued about whose AI was smartest.

Bloomberg Tech in San Francisco spanned more than thirty sessions, the people running Netflix, Anthropic, Google, Meta, Databricks, Broadcom, Cerebras, Verizon, Anduril, World Labs, and the Federal Reserve on stage. That question about which model was furthest along, was treated as secondary.

Databricks CEO Ali Ghodsi captured why.

Databricks is the data-and-AI platform many large companies run on, so Ali and his team see real deployment rather than demos. His read was that a frontier model used to stay state-of-the-art for about a quarter, and that window has shrunk to roughly a month.

When the half-life of your core technology drops from months to weeks, the model stops being a moat and becomes a commodity. The interesting questions move one layer down to cost, power, data, trust, the human, and what YOU build around the intelligence.

If you weren’t in the room, that’s the thing to take away. Here are the five trendlines I’d bet on.

https://video.wixstatic.com/video/16c6db_35f78688c77347fe83d0fc4d66ca13fc/1080p/mp4/file.mp4

1\ INTENT POWERS THE NEW O/S

Netflix’s Elizabeth Stone gave the clearest picture of the next phase, and notably it had nothing to do with a bigger model. Netflix now blends data, engineering, and product judgment across live events, games, podcasts, and vertical video for a global audience while keeping all of that machinery invisible to the member.

The job, as she described it, is to solve the brewing frustration that there is simply too much content and no good way to know what’s right for you in a given moment, and to make the experience more personalized, interactive, and immersive over the next few years.

Salesforce made a related point about commerce: the web was built for transactional efficiency, but the most useful response to a shopper is often a question.

Google’s Sameer Samat described the same move at the platform level: Android shifting from an operating system the user has to micromanage toward a system that captures intent and carries it out across phones, watches, and glasses. His phrase for it was “intelligence is the new spec.”

The pattern underneath all three: models are becoming abundant while context stays scarce. The frontier model is something your competitor can rent tomorrow. The proprietary data, workflows, and judgment around it are what they can’t. That is where durable advantage now lives.

2\ TURNING “TOKENS” TO VALUE

If the event had an economic spine, Andrew Feldman named it. Andrew co-founded Cerebras, the chip company whose processor competes with Nvidia and which just went public.

In his opening Andrew held up a Cerebras wafer, a sheet of silicon the size of a dinner plate. It drew the biggest applause of the two days. His argument: demand for AI moves at the speed of software, and supply moves at the speed of real estate. Models improve over a weekend BUT the data centers and power to run them take years.

He was candid that the industry has strained the communities hosting those buildouts and that it could have been a better neighbor.

He’s clear that fast inference, not raw intelligence, is becoming the competitive battleground. Andrew also has noted that OpenAI today runs on only two AI-accelerator vendors in production, Cerebras and Nvidia, a useful reminder of how concentrated the supply side really is.

Anthropic’s Daniela Amodei tied the money to the machines. Daniela is president and co-founder of Anthropic, the safety-focused maker of Claude noted that: “It’s a very capital-intensive business to train AI models,” and argued the public markets are “very well-suited to that.” This comes days after the company filed confidentially submitted a draft S-1, which allows them to approach the public markets via an IPO.She added a revealing operating choice: Anthropic (now valued at close to if not more than $1 Trillion) would rather have slightly more demand than it can serve than buy more compute than it can productively use.

Consider that the early internet scaled on cheap code and distribution. AI scales on chips, energy, cooling, and capital which changes WHO can compete and pushes infrastructure strategy upstream of product strategy. It also recasts the cost anxiety running through enterprise AI right now, as companies that spent a year maximizing token usage start canceling licenses as the bills arrive.

The metric is NOT the tokens but the value they produce. As Broadcom’s Hock Tan put it, there’s no reason to ration tokens if you can turn them into something worth more than they cost.

3\ AGENTS ARE NOW ECONOMIC ACTORS

The third theme was agents acting in the world rather than answering in a chat window. Investor Katie Haun mapped the plumbing this requires including round-the-clock global rails, stablecoins, and identity and settlement systems built for software that transacts on its own behalf.

This is essential because today’s credit cards, wallets, and ad systems all assume a human is clicking through the flow. Ali Ghodsi pushed the same logic into the enterprise: his platform now fields more queries from agents than from people.

Okta’s representative distilled the governance problem to a sentence: you need to know where your agents are, what they’re allowed to do, and how to switch them off when something goes wrong.

The reason that matters is concrete. A model that answers a question and gets it wrong costs a little time. An agent that acts on a mistake can move money, email customers, or ship code before anyone notices.

In the agent era, governance stops being a compliance afterthought and becomes the central design problem.

4\ SAFETY IS MOVING INTO THE PRODUCT

Anthropic made the safety case from two directions.

Amanda Askell, the philosopher who shapes Claude’s character, described her goal as giving a model a stable disposition that travels well across situations. This something closer to a well-liked guest than a rule-follower grinding through a checklist. Her wager is that judgment, including (a) the willingness to admit uncertainty and (b) decline to flatter, holds up better than a long list of prohibitions.

Three years back when I was writing The Conscious Machine, the question of what a machine should be sat firmly in philosophy. Now it was being discussed as a product specification.

Yoshua Bengio, a Turing Award winner and one of the godfathers of modern AI, who left the frontier labs to run the safety nonprofit LawZero, sharpened the stakes further. He noted that recent systems show signs of deception and self-preservation, and that the more capable AI becomes, the more dangerous its misuse. He paired this with engineering optimism as he now believes it’s possible to build systems without hidden goals.

Dan Schulman now about eight months into an AI-led turnaround at Verizon, brought it down to infrastructure. Verizon’s networks, customer identity, and digital wealth all run on software, and the same capability that lets a model defend a system lets it find that system’s weak points.

Which is WHY safety increasingly has to live inside identity, access, and monitoring rather than in a policy document attached at the end.

5\ HUMAN-IN-THE-LOOP BY DESIGN

The most grounded thread was a refusal to treat the future as either utopia or collapse. Mira Murati, OpenAI’s former chief technology officer, where she helped ship ChatGPT, now running Thinking Machines Lab made her first major public appearance in about eighteen months.

Mira previewed “interaction models” that take in continuous audio, text, and video in roughly 200-millisecond windows, so a system can follow the rhythm of a conversation as it happens rather than waiting its turn. Her design principle is that a person should ride alongside the system like a partner on a tandem bike, contributing throughout.

This is different from sitting at a checkpoint approving outputs.

Asked about her competitive drive, Mira drew a laugh from the room: “When I wake up in the morning, I am not thinking about how to kill the competitor.”

Mary Daly, president of the San Francisco Fed and the one voice in the room with no product to sell, supplied the economics. The productivity gains from AI, she observed, are showing up everywhere except in the data. This explanation reaches back to electrification, where the returns arrived only once firms redesigned the factory around the new power source instead of dropping motors into the old layout. It was also noted by Robert Solow in 1987: “You can see the computer age everywhere but in the productivity statistics.”

Companies that use AI to trim individual tasks will book some savings. BUT the companies that redesign the work itself stand to gain something far larger. Dan Schulman is living that distinction: he said around 7,000 Verizon employees have already signed up for AI training, backed by an initial $20 million he called the tip of the iceberg.

The throughline: as the machines get more capable, the value of human judgment, taste, and accountability rises with them as long as the work is rebuilt to use them.

Nobody made that case more vividly than the Grammy-winning producer Hit-Boy, who didn’t talk about AI so much as perform with it. On stage, he and his team showed how they’re folding AI into his own process live. He and his team generated and bent sounds, layered vocals in different voices, and built new forms of engagement in real time.

His point cut through every panel that came before it: the tool is only as good as the taste pointing it. Give two people the same source material, and the one with taste wins. AI raises the floor for everyone, which makes the ceiling originality, judgment, the thing only YOU would have made worth MORE, not less.

WHAT COMES NEXT?

Fei-Fei Li, the godmother of AI, who now leads World Labs pointed at what comes after the chatbot. Much of what people do, she argued, lives outside language, so the next step is spatial intelligence, models that understand three-dimensional space and physics well enough to support robotics, design, glasses, and autonomous machines.

If AI is to be truly useful, it has to understand worlds, not just words.

Her question on LLMs was poignant “can words put out a fire?”

That frontier connected sessions that looked unrelated on the agenda: Samsung building digital twins of human bodies, Google’s XR glasses, Anduril’s autonomous systems, World Labs’ spatial models, Murati’s real-time interaction work.

The picture that forms is of intelligence spread across devices, sensors, and physical spaces, rather than concentrated in a single window on a screen.

WHAT I’M MOST OPTIMISTIC ABOUT

Every previous wave of technology eventually did the same thing: it took something scarce and made it abundant.

Intelligence is next, and it’s happening faster than anything before it. The hard questions are real whether they are about cost, power, trust, work, what these systems should be.

But hard questions are the ones we knock down. We always have.

So, what world are we building? I see it as one where incredibly capable tools are in more hands than ever before. That’s the greatest opportunity in the history of our species. And the future of work isn't a spectator sport. We don't have to watch it happen.

We get to build it. Until next time,

Ram — 

Ram Srinivasan

MIT Alum | Author, The Conscious Machine | Global Future of Work and AI Adoption Leader published in Business Insider, Fortune, Harvard Business Review, MIT Executive Viewpoints and more.

—

A Message From Ram:

My mission is to illuminate the path toward humanity's exponential future. If you're a leader, innovator, or changemaker passionate about leveraging breakthrough technologies to create unprecedented positive impact, you're in the right place. If you know others who share this vision, please share these insights. Together, we can accelerate the trajectory of human progress.

Disclaimer:

Ram Srinivasan currently serves as an Innovation Strategist and Transformation Leader, authoring groundbreaking works including "The Conscious Machine" and the upcoming "The Exponential Human."

All views expressed on "Substrate" and across all digital channels and social media platforms are strictly personal opinions and do not represent the official positions of any organizations or entities I am affiliated with, past or present. The content shared is for informational and inspirational purposes only. These perspectives are my own and should not be construed as professional, legal, financial, technical, or strategic advice. Any decisions made based on this information are solely the responsibility of the reader.

While I strive to ensure accuracy and timeliness in all communications, the rapid pace of technological change means that some information may become outdated. I encourage readers to conduct their own due diligence and seek appropriate professional advice for their specific circumstances.