Gary Ma

We just dropped a 15-part comprehensive report on AI spend in the enterprise. Here are my top 3 takeaways: - Spend grew 5.4x last year to $177B, with most spend coming from existing buyers - Most dollars go to the frontier with 69% going to top 3 labs (Anthropic, OpenAI, Google) vs 50% last year - 86% of LLM API spend goes to Anthropic (55%) and OpenAI (31%) Crazy to think AI spend in our first report 3yrs ago was only $2.5B. Read the entire report here:

Why would Elon cap usage of Grok Bot when its traction is going so well? @dylan522p explains: "You have to remember what Elon's trying to do right now. He's taking SpaceX and turning it into a neocloud, in some sense. He's building ridiculous amounts of compute." "If you take his balance sheet, he's got like $80B of cash. $20B is going to go to rockets. The other $60B he can spend on compute." "He's raising another $40B right now. And at the same time he claims he wants to build 13 gigawatts. If you build all of that out, it's $400B+." "So where does he get the cash? He has to monetize his compute as effectively as possible. To be able to pay for all the capex that he wants to do." "And given what's going on in the spot compute market — being able to sell GPUs for $15 an hour, when the TCO of them is only $4 an hour, you have this massive delta. So he's exploiting that." "He needs to focus on, 'How do I build the most compute, period?' That's step 1. Step 2 is, 'How do I use that compute most effectively to dominate the universe?'" "The only way to build 13 GW financially is if he makes as much money as possible from the compute he has. So if that means he ripped some compute out of xAI's team to sell it to Anthropic and Google, he had to do that. Because ultimately a year from now, his xAI team will have a lot more compute to deliver by doing that." "So he needs to be able to generate all this cash flow, and bridge the financial gap, to be able to spend $300B, $400B, $500B on capex over the next year, year and a half."

AI agents running on employee endpoints are becoming a systemic security risk for enterprises & governments. Some are responding by restricting tools like Claude Code, as Goldman Sachs reportedly has. In this security advisory, we outline the risks and recommend a path forward.

Its very interesting to note the pace of application features from the foundational model companies. As the cost of the marginal token approaches zero, the value is not in selling tokens. It is very clearly in moving up the stack and owning the application layer. The first victim will not be established SaaS but AI Natives that spend heavily on the labs now - legal, sales, customer service AI Natives tbc. This is the lowest hanging fruit and the roadmap is clear from the traces and spans the model companies have from these companies already. The next are the "connective tissue" applications that make using the models a foundational part of a workflow. Take the google doc editing flows in OpenAI. Take the google drive/doc integration into chats. Both are very good.

human-in-the-loop └ human-on-the-loop └ human-near-the-loop └ human-cc'd-on-the-loop └ human-legally-liable-for-the-loop

Increasingly this will be the new job description for a lot of roles, not just coding. We’re clearly moving from a world where agents help us do our work incrementally better or faster, to us helping agents do their work. This turns everyone in one of these positions to that of a manager of a team of agents, instead of an individual contributor with just some added leverage from AI. The biggest areas of impact will be information-heavy jobs, like coding, cybersecurity, legal work, biotech, engineering, and many other fields. And as a result, in these jobs your leverage will begin to be decoupled from the normal ratio of hours spent working to output generated. For essentially the history of knowledge work the main variable you had to play with was hours of work done (assuming fixed level talent/coordination/strategy). Now we’ve separated these variables because agents can parallelize work for us, to in theory execute as much as we need to pull off whatever goal we’re trying to accomplish. The people and organizations that discover this sooner, and orient their operations around it, will see compounding gains in their workflows.

GPT-4 could do 4 minutes of independent work with a 50% chance of success. Current models can do multiple weeks of work, and January 2026 is a reasonable cut-off for the moment agentic work started. Large productivity impacts before that would mostly be confined to coding.

5 hour limit gone for $200 “ultra” Claude tier. Anthropic is on a run. Super cool.

We've evaluated Gemini 4 Carbon and Gemini 4 Argon on FrontierCode 2.0, our coding benchmark that measure's a model's ability to go beyond the spec. Gemini 4 Carbon scores near SOTA with a score of 40% and a cost of $4.04 per task, behind Claude Opus 5.5, GPT-6 Astra, and Claude Sonnet 5.5. Gemini 4 Argon scores 26.3% with a cost of $6.41 per task, ahead of Muse Wonder 1.0, but behind Claude Haiku 5.5 and Grok 4.7. See the leaderboard with full scores: cognition.com/frontiercode

North Korea GDP just went up $5.57B

Binance's initial investigation suggests that the security incident may involve a compromised DevOps engineer’s laptop, which allowed an attacker to gain access to a service container and submit unauthorized withdrawal transactions. We have found no evidence of any wallet private key leak. We will share a detailed technical post-mortem once the investigation is complete and the facts are fully verified. The estimated total impact sits at around $6.3B. We have worked with blockchain partners to freeze affected assets, with ~$3 billion frozen on-chain so far. Binance will cover all losses from the SAFU fund and the company’s funds. User funds are safe. We will also publish an updated Proof of Reserves once the remediation is complete. As an immediate precaution, Binance has: - temporarily suspended deposits and withdrawals while security checks are completed; - revoked the affected address authorisation; - engaged law enforcement and blockchain intelligence partners to trace and recover funds. Our teams are working around the clock to resolve this issue safely and thoroughly. We are working closely with our law enforcement partners to confirm the identity of these criminals and bring them to justice. Please rely only on official Binance channels for updates. Please feel free to contact Binance Customer Support if you have any further questions. binance.com/en/support/annou…

Amazon locked out my Grok Bot. So it just bought the same thing from a different site (at a lower price).

AI is already running across the enterprise, and securing it is now one of CISOs' biggest concerns Already in production at their companies: - Employee generative AI tools (92%) - AI coding assistants or agents (83%) - MCPs or agent-to-app integrations (55%) - Autonomous agents with tool access or write permissions (34%) Their biggest concerns: - AI attacks outpacing their team (55%) - Employees leaking data into AI tools (47%) - AI agents exceeding permissions or getting compromised (46%) From a16z's @OwenColby and @steph_zhang, breaking down a survey of 100 CISOs at companies with 1,000+ employees: a16z.news/p/the-new-security…

This is correct. The vast majority of enterprise workflows have not actually had agents applied to them in any meaningful way. The economy probably has this deployed < 1% right now. At best you have a few automated systems in a company that do a single pass of an LLM to make a decision or provide feedback, but very little real “automation”. Some of this is a statement about the adoption rates of technology in enterprises, but the vast majority of it is really just due to the capability of AI itself. That’s all changing. Now with computer use, long running agents, harness improvements and surrounding infrastructure, we’re finally at a point where we can start to apply agents at more complex workflows. We will see self-improving systems that can take any particular workflow goal and improve the execution rates over a volume of instances. And with the emerging form factor of agent swarms, you can just choose to deploy any amount of compute at a problem and agents will off and do the work. This won’t be about and users prompting agents to complete a specific task for them, but instead the organization reengineering the underlying workflows to take advantage of agents in their processes. This will have a massive impact in finance, life sciences, healthcare, and other information-heavy domains. Lots of opportunity to build and deploy these systems right now.

Databricks CEO Ali Ghodsi says if frontier models stopped advancing, it wouldn't matter for most orgs because enterprise AI is bottlenecked on context, not intelligence: "There's no agentic work that's automated the whole enterprise. That has just not happened. Why is that? The real reason is the models are smart enough, but they just don't have the context that exists inside of any organization." "They have not been in every meeting. They don't know what's in everybody's heads. They don't know all the processes. There's always a couple of employees who know everything in every organization. You go tap on their shoulder, and everybody's like, 'Oh my God, what would happen if he or she quits?'" "If you just fused that and gave that context into the AI models, just the frontier today, I think there's so much productivity gains you could get for any organization on the planet." "For that, we actually don't need smarter models... If the frontier doesn't advance, it doesn't actually matter for the vast majority of organizations on the planet. They're just so far behind in the adoption curve of actually automating things and getting value out of this stuff." @alighodsi Read the latest Profiles of the Future by @bhorowitz and @mikemcg0 on Databricks & the Future of the Firm: a16z.news/p/databricks-and-t…

Summing up the challenge of commercializing robots in one photo The bar for both reliability and output quality is much higher for physical labor than it is for knowledge work. A chatbot that produces work that is just “good enough” or low quality can still be useful. A mediocre rough draft could still save time because a person can keep what’s good and fix the rest before its final. Copywriting, slides, and contracts can all be edited. What is considered “good enough” for physical work has much higher standards. A bad weld can ruin metal. A wrong dose given to a patient is unacceptable. Outputs from work done by humans or robots in the physical often serve as direct inputs into the workflow that are hard to recover from if an error arises. Outputs from work done by humans or Chatbots for knowledge work are much more tolerant of errors and poor quality because supervision can be baked in before the output is used further down the workflow. The reliability threshold is 2-4 orders of magnitude greater for physical work than knowledge work and it’s what we’ve started to achieve this year. The upside is that by optimizing for reliability, robots will be able to work at a high level of utilization and operate at higher levels of autonomy compared to general purpose AI models in 2023. Machines that can deliver up to 24 hours of work a day as opposed to minutes or hours of work.

Agents can now shop for domain names.

An abliterated LLM just found me ~$3,300 in live ElevenLabs keys on the internet. All the OpenAI / Anthropic keys it found were already drained which makes me think they've already been actively monitoring this. Pretty insane.