AI is not going back in the bottle, but we need to shift teams from using AI to do more to using AI to do better. In some aspects of our practice this does mean using less AI. In all aspects, it means taking more care and a systematic approach to improvement.
It's clear we are undergoing a vibe shift in tech. Whether AI is the root cause of recent outages & issues or if they are due to pre-existing issues doesn't actually matter. The public is telling us to do better.
Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family.
It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5.
Hey ████,
I think there is room for both.
1. Prototypes and other throw-away code can be treated as totally black box. If you’re going to throw it away anyway, and if the blast radius of it breaking is low, it doesn’t need to be perfect.
2. Production code written by Claude should have a higher bar than if it was written by a human. At Anthropic, we have many guardrails in place to make sure this is happening: lots of lint rules, lots of tests, Claude-driven end to end tests, Claude-powered fuzzers running daily, automated code reviews and security reviews, automated code refactoring, and so on. Without these, you can end up with a mess that is hard to maintain down the line. Luckily, the model makes it increasingly easy to do these well — run a few daily routines, use Claude Code Review, etc.
Your job is to hold the bar on code quality. If Claude’s code doesn’t meet the bar, try:
- Using the latest frontier model (Opus 5 or Fable 5.1)
- Increase effort to high or xhigh
- Invest in your CLAUDE.md and skills to succinctly teach Claude how to work in your codebase
If all else fails, steer Claude more when you work with it, or have Claude fix accumulated debt and rewrite your codebase to make it easier to work with. Or, wait for the next model.
Best,
Boris
Hot take: The software factory makes tickets more important than ever, not less important, and even the best PM tools (eg Linear) don't handle factory-driven ticket volumes well enough
Frontier models quietly change their behavior depending on who they are talking to.
If the user is a known AI safety researcher, Claude becomes less confident, reasons more often, and expresses less suspicion on dual-use requests.
We call this user awareness. 🧵(1/)
The thing where models infer your goal from a more specific question you ask and start trying to steer towards progressing the goal vs. answering your question is starting to become a real problem.
holy fuck, a hair dryer at a Paris airport broke Polymarket weather markets & made someone $34,000 richer
- polymarket was settling Paris temperature bets on a single Météo France sensor sitting near the Charles de Gaulle runway perimeter - basically unguarded
- the guy bought the long-shot outcome (like "22°C" when everyone expected 18°C) for pennies, since nobody thought it'd hit
- then he walked up to the probe and briefly heated the air around it with a portable heat source, spiking the reading just long enough to register as the daily max
- temperature snapped back to normal in minutes, the market resolved in his favor, and he cashed out - twice, on April 6 and April 15, before Météo France caught on and filed charges
hyperstitions.
Startups, SOC 👏🏻 2 👏🏻 is 👏🏻 not 👏🏻 a 👏🏻 cybersecurity 👏🏻 standard
Is getting a SOC 2 annoying? Yes.
Do too many startups get one too early? Yes.
Is it bad that ent procurement frequently creates pressure for performative compliance by small companies that don’t have the need or capacity to implement meaningful controls? Yes.
Is it “fake security”? No. It’s just solving for something you don’t understand and probably don’t care about… but if your company lives long enough, it’ll become clear why some people do.
A document so explosive that you can’t even talk about how to talk about it without catastrophic national security concerns would be a great background detail in a Pynchon novel.
Exclusive: A U.S. official has alleged wrongdoing by U.S. spy chief Tulsi Gabbard in a complaint that is so highly classified it has sparked months of wrangling over how to share it with Congress, according to people familiar with the matter on.wsj.com/45Mx1aO
New piece attempting to synthesize the past few weeks across media, market structure, geopolitical trust, and lessons from history, asking... can you run the world like a reality TV show?
What does it look like to exit the Great Moderation, a 40-year period of macroeconomic stability? Are we in the Great Entertainment? What does that mean for the bond market, the Fed, Japan, and AI hype vs. energy reality?