ashu garg

Are @AnthropicAI & @OpenAI overvalued ? Both cos are similar in size (2026E $26B net for Anthropic and $28B for OpenAI), but with very different business mixes, and trade at ~35X and ~30X revenues respectively.

This week has been one of the most consequential in AI this year. 4 launches. 4 different labs. 4 different winners. Voice → OpenAI. GPT-Live launched this week and you can interrupt it, talk over it, and it keeps up. Until now, voice assistants were walkie-talkies. This is a phone call. Expect every consumer voice product to be rebuilt around this within a year. Image → Bytedance. Seedream 5 Pro slots in just behind OpenAI's GPT Image 2 on quality but at a fraction of the price - and it's aimed at the commercial center of the market: editing existing images and producing infographics, not generating art. Price → xAI. Grok 4.5 is the first launch of the Cursor/SpaceX era, and the claim is aggressive: Claude Opus 4.7-level quality at $2/M input and $6/M output. For reference, Opus 4.7 lists at $5/M in and $25/M out. If the benchmarks hold, that's a model claiming parity with the frontier at roughly 1/4 of the price on output tokens… the line item that dominates every agent workload. Coding → still Anthropic. OpenAI's GPT 5.6 Sol and Sol Ultra launched yesterday to solid but not spectacular early reads - the consensus is they close the gap with Claude Fable 5 but aren’t quite there. A year ago the question was which lab wins AI. This week suggests what I’ve been saying for a while: none of them. There is no single frontier - there are many, and a different lab leads each one. More in my recent blog below.

@JayaGup10 @AnthropicAI @OpenAI Why not ?

@ashugarg @AnthropicAI @OpenAI No

@JayaGup10 @AnthropicAI @OpenAI I voted. Did u ?

@ashugarg @AnthropicAI @OpenAI What do you think Ashu

Meta’s Muse Spark 1.1 has great price-to-performance, appears to match GPT-5.5/Claude on most use cases except for coding and multimodal.​​​​​​​​​​​​​​​ Didn’t expect @finkd to validate me so quickly 😜

Interesting time to vote now that Muse Spark 1.1 is out!

@lindaxie @willwarren Huge congratulations

Personal news: my husband @willwarren and I are expecting a baby girl this winter and we couldn't be more excited! 😊

Microsoft just committed $2.5B to launching a company to help enterprises move AI from pilots to production. That tracks with what we're hearing from both sides of the table: pilots are everywhere right now. The problem is that getting started is easy, but getting into production is not. Pilots are tested in clean conditions with curated data, a narrow use case, a controlled environment. Production is a cross-section of real life, and real life is messy. Data is incomplete, workflows don’t match the diagram, and edge cases show up on day one. So a pilot that “works” often hasn’t actually been tested against the thing that determines whether it survives contact with production. For founders, the takeaway is to show outcomes. Enterprise buyers aren’t asking if your product works in a demo. They’re asking if it works in their business, on their data, with a measurable result they can point to.

@realAkashAnand RESEARCH

At YC we have an internal platform called Bookface, which has better startup advice than anything you'll find on Google. But (a) only YC founders have access, and (b) it doesn't cover every topic. Plenty of times I've searched and it come up empty. Turns out Claude Fable 5 is really good at multi-step research. So I built a skill that helps anyone learn about a topic or figure out how to implement something at work. It finds essays and articles with immediate action points, written only by top-tier operators, founders, and experts who've actually done the thing and can tell you exactly what they did in the situation you're in. Fluffy SEO blogs are strictly rejected. And it does a MUCH better job than Google AI mode, Perplexity, or even Claude deep research. I’ve been using this every day, with great results. Here are some things I’ve researched this week: - How to define top-level KPIs and metrics for my startup - How to run a company all-hands with my employees - How to prompt Claude to pick the right fonts for video - How to measure the success of features (small, medium, and large) It’s useful for everything from sales, to marketing, to engineering, to product, to design, to general advice. If you want this skill, comment "RESEARCH" and I'll send it over. (You'll have to follow me so I can DM you.)

Alex Karp rant on AI Labs bundling is right. But the answer is not replacing one bundle with another. So we are unbundling Palantir AI stack: - Data layer: Databricks & Snowflake - AI layer: Anthropic & OpenAI - Ontology: Unity, Atlan, Horizon - Security: Skyflow Here’s how starting with @PalantirTech’s own explanation of what they do:

Over the last 30 years, career ladders rewarded specialization. @Azeem thinks AI may finally flip the script in favor of generalists. In his recent conversation with my partner @joannezchen, he captured something important about where work is heading. AI can complete tasks, but it can’t decide which tasks are worth doing, how they fit together, or whether the output is any good. In a world where execution is cheap, the ability to work across functions and turn an ambiguous problem into a useful outcome becomes more important than ever before. The best builders are already becoming more fluid: part product thinker, part systems designer, part operator, part storyteller. The leverage now available to those people is massive. AI automates more of the routine work, but that creates more room for humans to do the work that still requires judgment: deciding what matters, setting direction, building trust, and taking responsibility for the outcome. The goalposts will keep moving. As AI gets better, human judgment will have to stay out front. youtube.com/watch?v=gYGyfH…

Indeed. We are going from “humans using software” to “software using humans” to get work done — a very fundamental shift.

If I could edit Infosys's Wikipedia -- I would change it to this:

Agreed. HOWEVER the difference is that Databricks is a data platform - the catalog, the query engine etc. So you trade one full stack service for another. Skyflow is the ONLY “middleman” that doesn’t hold your data hostage!

Every IT Services consultant is now an FDE😜

The labs are already coming for the application layer. Anthropic has been the most aggressive with Claude Code, Cowork and most recently Claude Tag. OpenAI is turning ChatGPT into a super app spanning coding tools and agents, while moving into devices and robotics. Google is leveraging its broad application footprint across Android and Google Workspace to embed Gemini. The startups that win will build products the labs cannot easily replicate: products wired into a specific customer’s environment and shaped by proprietary workflows. More in this month’s B2BaCEO: linkedin.com/pulse/ais-winn…

Anthropic shipped an agent that lives inside your org’s Slack, builds memory, and runs autonomously for extended periods. As long-horizon autonomy improves, we’ll see more workflows designed for agents first. Instead of a person leading the work and looping in an AI tool for help, the agent starts the task, and the human verifies, redirects, and supplies judgment where needed. Anthropic’s approach offers an early glimpse of what this might look like. Claude is provisioned with its own workspace identity, with access scoped by channel rather than borrowed from any one user. However, for these agents to be truly useful, enterprises will need infrastructure that’s still being constructed: agent identity, authentication, access controls, and accountability. This moment reminds me of where the PC was in 1991 and the internet in 2000: awareness is near-universal and adoption is exponential, but most of the category-defining products—along with the deeper rewiring of how work gets done—are still ahead.

Building exceptional products is still hard, even in the AI era. This is a moment to think through customer and market segmentation. For founders, there are two questions to ask yourself: 1️⃣ First, what do your customers care about most when it comes to AI? Early adopters will buy a product that’s still rough around the edges, and invest in co-development. Companies who value control want local models that can be run within their VPCs. Cost-conscious customers will run long trials with multiple vendors before making a commitment. 2️⃣ Second, what’s the nature of the problem your product solves? Horizontal tools are deployed company-wide, but you are truly competing with the giants. Departmental products, focused on functions like finance, HR, and GTM, need reliable off-the-shelf solutions with clear ROI. Finally, there are solutions that respond to strategic, often customer-facing, board-level priorities. These are heavily-customized and bespoke solutions, often for teams that want to be the first to market. In the latest edition of B2BaCEO, I break down how startups can build something durable in this era of AI: linkedin.com/pulse/ais-winn…