Posts

Claude Fable 5 Shows a New Way to Release Powerful AI

Image
Anthropic has released its strongest public Claude yet , and that alone is big news. Claude Fable 5 is clearly a capability story. Anthropic says Fable 5 is its most capable model ever made widely available, and Google Cloud describes it as strong at multi-step reasoning, software work, and document analysis. That is a real jump. But what caught my attention is not only the model. It is the way Anthropic chose to release it, and what that says about where the most advanced AI models are going. The Interesting Part Is The Release Anthropic did not ship this model like a normal product release. It shipped with clear limits . Powerful models may increasingly reach the public with built-in control layers around them. In simple terms, Fable 5 suggests a new release pattern: better model, but not direct access to everything that model can do. The Public Version Has Limits The first detail is the easiest one to miss. Fable 5 is described as a Mythos-class model. In plain language, that...
Read more

Anthropic’s Series H and Draft S-1 Point to a Bigger Shift in Frontier AI

Image
 On May 28, 2026 , Anthropic announced a $65 billion Series H at a $965 billion post-money valuation . On June 1, 2026 , it said it had confidentially submitted a draft S-1 to the SEC, which means it now has the option to go public later if conditions are right. Those look like normal company-building signals. But the more interesting story is not the valuation, and not the IPO option by itself. It is the kind of company Anthropic has had to become first. Read next to Anthropic’s own posts on compute expansion , Responsible Scaling Policy , Constitutional Classifiers , and Project Glasswing , the signal is hard to miss: frontier AI companies are starting to look less like ordinary software companies and more like operators of critical systems. This is less an IPO story than an operating-model story. What Do I Mean By "Operators Of Critical Systems" Here? I do not mean Anthropic is literally becoming a power company, a telecom, or a public utility. I mean somethin...
Read more

Why Frontier AI Needs Critics It Can't Buy

Image
  The Vatican was the dramatic setting. The more important signal came from Chris Olah's remarks : every frontier AI lab is trapped inside incentives that can conflict with doing the right thing. That is the part builders should pay attention to. A lab at the frontier is saying, in public, that it cannot be its own final critic. This Is Not Really a Church Story You do not need to be Catholic to find the point here. The Vatican matters as a setting because it sits outside Silicon Valley's incentive stack. It is one example of an institution the labs do not control, cannot hire, and cannot fold into a product roadmap. Anthropic's own follow-up, Widening the conversation on frontier AI , makes the scope even clearer. They are not only talking to clergy. They are talking to philosophers, humanists, lawyers, writers, psychologists, and civic leaders across traditions. The real story is simple: the labs are looking for outside judgment because inside judgment is co...
Read more

Project Glasswing Update: The Bottleneck Is Moving From Discovery to Patching

Image
    Last month, I wrote that Project Glasswing mattered less as a model announcement and more as a deployment signal . Anthropic had a frontier model with unusually strong cyber capability, and instead of shipping it broadly, it wrapped it in a controlled defensive program. The new initial update for Project Glasswing makes that decision easier to understand. It does not prove vulnerability research is a solved problem. But it does show something important: AI-assisted vulnerability discovery is scaling faster than the human systems that verify, disclose, patch, and deploy fixes. What Changed In The First Month? Anthropic says its approximately 50 partners have already found more than 10,000 high- or critical-severity vulnerabilities across systemically important software. The public examples are notable: Cloudflare says Mythos Preview helped it find 2,000 bugs , including 400 high- or critical-severity issues, with a false positive rate its team consid...
Read more

The Hype Meets the Data: State of AI 2026 Results

Image
Last month, I shared the open survey for the State of AI 2026, asking you to help replace the endless industry hype with actual, ground-level field data. The results are finally out, collected from 7,258 web developers . After reviewing the deep breakdown across every section—from models and agents to demographics and real-world pain points—the verdict is clear. AI-assisted coding is no longer a bleeding-edge experiment. It has crossed the threshold to become standard engineering practice. Here is what the data actually tells us about how software development is changing in 2026. ⚡ The 50% Threshold is Shattered In 2025, developers reported that AI generated roughly 28% of their code on average. This year, that figure has doubled to 56% . The heaviest adoption growth is occurring among developers who rely on AI for over 75% of their output. Interestingly, the data shows that higher reliance on AI correlates with fewer years of professional experience. But even for seasoned vet...
Read more

Adding Session Memory Without Building a Preference Engine

Image
  I just added another feature to Kino , my educational, cinema-themed distributed systems project. The LangGraph-based agent service is one part of that platform. This time the feature was memory . But not the vague, hyped kind. I did not want Kino to pretend it knows a user's taste forever. I wanted something much narrower and much more useful: Short-term conversational memory for follow-up turns that helps the agent continue the current search without pretending to know more than it actually does That decision mattered a lot. It kept the feature small enough to trust, but still visible enough to feel like real agent behavior. What changed The new capability is simple to describe. A user can start with: Discover exactly 3 comedy movies from 2010 onward from Kino's catalog. Then follow up with: I didn't like them, please discover different ones. And Kino can continue from the current thread instead of tr...
Read more

Not Classic RAG: Building a Structured-Retrieval Discovery Agent with LangGraph

Image
  I just added a new feature to Kino , my educational movie-discovery project built with LangGraph: a prompt-driven discovery flow that finds grounded titles from a local catalog. The easy label would be RAG . More precisely, it is not classic RAG . What I built is closer to a structured-retrieval agent : the model interprets the request, a narrow service returns structured facts, and deterministic code enforces the final result. That distinction sounds academic at first. In practice, it changed almost every implementation choice. What I actually built Kino is an educational project, but this feature forced a very real architecture decision. I wanted a user to be able to type something like Discover comedy movies from 2010 onward from Kino's catalog. and get back grounded titles from the project's own data. The final flow is intentionally small: an LLM interprets the user's request a single search_titles tool turns that into a structured catalog query t...
Read more