Another good article explaining agents and harness.

I've been writing a whole guide! simonwillison.net/guides/age…

很好的综述

Agents doing boring work is good for solid engineering.

If you don't know what a harness is, read this.

I've got an agent in a loop optimizing a renderer with the goal to minimize frame times (and tests to measure). It got times down from 88ms to 2ms and allocations down from ~150K to 500. Sounds good, right? Wrong. This is exactly why agent psychosis is a big fucking problem. As an experiment, I rewrote the Ghostty core render state in Go, with access to identically laid out data structures as Ghostty and the exact same validation tests. I made a purposely naive renderer (simple, correct, but slow). 88ms per frame with 150,000 allocations (horrendous, lol)! I then kickstarted a Ralph loop to bring the frame times down. I told it it can't modify input data structures or the public API or tests (they're correct), but it can do anything else it wants. It got to work. It has worked for about 4 hours. I've spent around $350 on this experiment so far. The results? 88ms => 1.5ms 150K allocs => ~500 allocs Incredible right? Nope. My hand-written renderer I ported has frame times (same benchmark) of ~20us (0.020ms) and 0 allocations in the update path. This is the problem with psychosis and lacking systems understanding. If you don't understand the system, you're going to accept that this is an incredible result. If you understand the system, you'll see better solutions immediately and can do roughly 75x better on throughput. The people who blindly trust agent output are in the former camp. They're sheeple, overdrinking from a fountain of mediocrity. Standard disclaimer: I use AI all the time. I like AI. The point I'm making is to not blindly accept results. Think. Analyze. Learn.

Mock foodpharmer all you want but a part of this change has to be credited to him for making people aware of what they are consuming. Else nobody would have given a rats ass about consumer benefits

I believe that spec-driven development, whether formal specs or conversations, is the future. But writing great specs is hard, and always has been. That's why I'm super excited about this new blog post from the @kirodotdev team, and our automated reasoning teams at AWS.

We built a database system on S3 Express, and I wrote a book about latency, so let me comment on some of the technical details. When we talk about write latency in a database system, the limiting factor is typically the time to commit to durable storage. If you are using object storage as your backing store, you have two options for the write path, depending on your durability model. The first option is to write to the object store asynchronously, which means object storage is irrelevant to your write latency, and therefore, uninteresting when comparing S3 Express and NVMe latency. The second option, which is what we actually do, is that you write to S3 Express synchronously, and now your commit latency is bound by it. In other words, when a commit needs to become durable, we perform a PUT request on S3 Express as a replacement for a `fsync()` call on NVMe. In other words, if the PUT request succeeds, your commit is now durable on S3 Express, just like with `fsync()`, it's durable on disk. (Of course, you often want replication to multiple disks, but you can parallelize that, and it does not change the scenario very much.) In this scenario, we can meaningfully compare NVMe and S3 Express latency: NVMe has up to 100x lower latency because enterprise-grade NVMe latency is in the tens of microseconds, whereas S3 Express latency is in the single-digit millisecond range. Of course, depending on your database architecture, you may also see additional latency from networking, but there is no scenario where object storage has lower latency than NVMe. Of course, there are many other advantages to building on top of object storage. For example, high availability, elastic storage scaling, cost-efficiency, and, in my opinion, competitive latency for many use cases. But we're talking about just pure write latency; S3 Express is clearly not faster than NVMe.

Barbara Liskov is a Turing award winner famous for her contributions to programming languages and distributed systems. I interviewed her recently about: • Being rejected from college based on gender • The software crisis of the 1970s • Paxos vs Viewstamped replication (her invention) and why one is more well known • Stories of Dijkstra and how his work influenced hers • Why her Turing award was questioned Where to watch: • Youtube - youtube.com/watch?v=T9CGjbPZ… • Spotify - open.spotify.com/episode/7yM… • Apple Podcasts - podcasts.apple.com/au/podcas… • Transcript - developing.dev/p/turing-awar…

The DDIA book (Designing Data Intensive Applications) is a classic no-doubt and has much in-depth stuff about distributed systems (Nowhere close to completing it though). However it is also too much to digest for any beginner (me). Personally I found this one a much better starter book. Gives a birds-eye-view of the systems and components along with a hands-on exercise after each pattern. Also just a bit over ~160 pages - readable in a few days. A week or so if doing hands-on›

Marc Brooker (AWS Distinguished Eng): "The downside of caches, especially in distributed systems, is they have this mode, where the cache is empty or contains the wrong data. The system is slow, often down, because now the backend isn't scaled to deal with all of this uncached traffic. Customers are very disappointed and often it is down in a stable way. Like it's still it's down, but it's not going to come back up under its own energy. Because, for example, all of this traffic is causing a huge amount of contention in my database or is saturating the network and so I can't even refill the cache. It's not even getting the right kind of data in In general, I prefer to see the teams around me avoiding caching where possible." @MarcJBrooker

I love Geoff. But he understands even less than Dario about the effects of technological revolutions on the labor market. Again, don't listen to AI scientists, as brilliant as they might be, and even less to AI CEOs, as successful as they might be, for questions of labor economics. Listen to reputable economists who have studied these things like @Ph_Aghion , @DAcemogluMIT , @erikbryn , @amcafee , @davidautor , etc.

In India, the unemployment rate for illiterates is 3%. For graduates aged 15 to 24, it's 40%. The more educated you are in this country, the more likely you are to be jobless. Not because education is worthless, but because it creates aspirations the economy struggles to absorb. 🧵👇

Introducing Claude Design by Anthropic Labs: make prototypes, slides, and one-pagers by talking to Claude. Powered by Claude Opus 4.7, our most capable vision model. Available in research preview on the Pro, Max, Team, and Enterprise plans, rolling out throughout the day.

Honestly I really *enjoy* engineering work. I also doubt you can develop the structured thinking required to be a leveraged engineer without actually doing it every day. If you want to keep doing leveraged engineering work, come to Convex.

Opus 4.7 criticizing work done with GPT 5.4 and getting everything wrong. GPT 5.4 critique to the original claims are all spot on matching the errors that I outlined here.

Introducing Claude Opus 4.7, our most capable Opus model yet. It handles long-running tasks with more rigor, follows instructions more precisely, and verifies its own outputs before reporting back. You can hand off your hardest work with less supervision.

Time for a big systems advice thread! In distributed systems there's no magic "push everything to prod at once" button. Every service gets pushed independently and nodes within a service get updated incrementally. If you mess up forwards/backwards compatibility you can fail irrecoverably. So how to avoid this? 1/5: Decouple data and code changes. Never push out a release that changes how data is stored at the same time as the code that uses this new data. If there's a bug and you need to roll back to the old version of your code it won't be able to handle the new data in the new format. Instead push out a release that first changes the data in a way that’s compatible with both the old and new code (e.g., optional fields etc), when that’s stable push out the new code that uses it, then when that’s stable you can change the data to remove backwards compatibility. This is known as a “migration” in the database world and yes it’s annoying, but yes you need to do it.

have you read the red team blog post? i believe it has oodles of technical detail you'd appreciate red.anthropic.com/2026/mytho…