New blog post, about a great little way to add fairness to distributed systems.
Combining stochastic fairness queuing, best-of-two, and shuffle sharding. Read it here: brooker.co.za/blog/2026/02...
Now on YouTube, Monday's CMU Database Seminar on Aurora DSQL. Thanks again to the awesome CMU folks (especially Andy Pavlo) for inviting me to do this talk.
www.youtube.com/watch?v=mK2h...
Yes. More accurately, I think I mean "code for which a good (if informal) specification of correctness exists", where that specification can be written business logic, or even something like "number go up".
We aren't there 100% for complex specifications yet, but the rate of improvement is v high
I don't think, at least for business logic code, having humans in the loop of reviewing code adds a lot of value. We haven't quite built the tools that make it unnecessary, but I don't think human-in-the-code is where we end up long-term.
Today's Postgres vs. World Seminar Speaker: Marc Brooker (@marcbrooker.bsky.social) will present the architecture of Amazon's Aurora DSQL Postgres-compatible serverless OLTP DBMS. Zoom talk open to public at 4:30pm ET. YouTube video available after: db.cs.cmu.edu/events/pg-vs...
Hiking too. It took an economic activity (walking to get somewhere) and turned it into a leisure activity (walking for the joy of exploration or exercise). Again, nothing wrong with that.
But the economic activity of walking is much less relevant now (at least as a large % of daily time spend).
The example was picked for that reason. Knitting, in form, is hard to automate. But the truly economically interesting cases have been replaced or automated.
What remains is craft. People love to do it, people love to see it and buy it. Great! But it's no longer a large-scale economic activity.
The cost of turning written business logic into code has dropped to zero. At best, near-zero.
The cost of integrating services and libraries, the plumbing of the code world, has dropped to zero. At best, near-zero.
What does that mean for the future?
New blog post: brooker.co.za/blog/2026/02...
The world if MacOS supported SECCOMP_RET_USER_NOTIF
I'm still learning how easy it is to go from "software I wish existed" to "software that exists" now, especially for anything that's a script, automation, or small utility.
Everything I internalized in 20 years about the cost of building things has changed in less than two years.
I’m happily building DSQL persistence for @temporal.io.
@marcbrooker.bsky.social and team have gifted us a fab service.
I readily accept I may be messing up my integration 😂 Guidance greatly accepted 🙏
(Oh! @kiro.dev has proven invaluable in this ❤️)
New blog post "Pass@k is mostly bunk" brooker.co.za/blog/2026/01...
Pass@k, especially for k>3, can make basically anything look good.
In my first blog post of the year: The Box.
Constraining what AI agents can do is required for making them safe and reliable. Pushing that control out of the agent itself, into its environment (The Box), is the only way to get the outcome we need.
brooker.co.za/blog/2026/01...
Moving Towards Spec-Driven Development with Kiro and Agentic AI. Listen to @marcbrooker.bsky.social on The Real Python Podcast @realpython.com realpython.com/podcasts/rpp...
Finally, the session i have been most looking forward to at #reinvent2025! @marcbrooker.bsky.social discussing virtualization techniques used with services like #AgentCore - a 500-level Chalk Talk (with no slides).
We need more of these types of sessions @awscloud.bsky.social
On the success of ‘natural language programming’ via @marcbrooker.bsky.social brooker.co.za/blog/2025/12... #programming #genAI #aI
New short blog post, touching on how I'd redesign UUIDv7 to keep the database performance benefits while mitigating most of the downsides.
Short version: replace unix_ms with H(unix_ms, unix_ms >> 13 | salt).
Get yer extremely niche RFC commentary here: brooker.co.za/blog/2025/10...
New blog post, reflecting on nearly seven years since the Firecracker launch, and how we're using Firecracker to power serverless databases (in Aurora DSQL) and infrastructure for AI agents (in Bedrock AgentCore).
Here's the post: brooker.co.za/blog/2025/09...
NULL BITMAP: Using Huffman Coding for Sampling Discrete Distributions: buttondown.com/jaffray/arch...
In DynamoDB, writes and their preconditions are evaluated atomically, and are linearizable. You could say that makes them strict serializable transactions, but that's not the most useful mental model for the app developer.
TL;DR Despite the similar name, Dynamo and DynamoDB have fairly little in common architecturally, and make some very different decisions.
Seattle friends - in case you missed it: lu.ma/umtmqb38
The Seattle Systems group is hosting their very first meetup! @marcbrooker.bsky.social will be presenting Aurora DSQL.
Excited to see more systems groups popping up (thanks @eatonphil.bsky.social). My colleague @devhawk.net is co-organizing.
People often ask me about the differences in architecture between Amazon Dynamo (the 2007 SOSP paper), DynamoDB (the AWS serverless NoSQL database), and Aurora DSQL (the AWS serverless SQL databases).
I memoized the response on my blog. brooker.co.za/blog/2025/08...
If you're into databases and you're in Seattle, @marcbrooker.bsky.social is talking at Seattle Systems on Aug 21: lu.ma/umtmqb38
LLMs are more powerful, more dependable, more efficient, and more flexible when deployed as a component of a carefully designed system.
New blog post: brooker.co.za/blog/2025/08...
Check it out on YouTube (youtube.com/watch?v=UGdM...), Apple Podcasts (podcasts.apple.com/us/podcast/t...), or Spotify (open.spotify.com/show/5GtS7mZ...).
I had a great time talking to Ben and Will from Antithesis on the BugBash podcast about software testing and validation at AWS.
In this wide-ranging hour we talk deterministic simulation testing, fuzzing, formal methods, the history and future of testing, and much more.
Fun conversation with @jkxosound.com about where I think things are going in the AI agents space, Strands Agents, and rats 🐀
I hear TPC-C is the best, most representative, benchmark for modern analytics applications.
Prototyping agents is easy, fast, and fun (especially if you use Strands). But how to you get them into production with all the scalability, security, and resilience you need?
We've been building Bedrock AgentCore to answer exactly that question. Check it out: aws.amazon.com/blogs/aws/in...
