Lorin Hochstein

I have no information about how this incident came to be but I can confidently predict that people will blame it on greedy execs and sloppy devs, regardless of what the actual details are. And they will therefore learn nothing from the details.

Anthropic recently published a white paper on how LLMs have changed how work is done within the company. My own thoughts on that white paper are here: surfingcomplexity.blog/2026/10/18/a...

Show me… networking! www.githubstatus.com/incidents/3q...

Lord forgive us our trespasses and also if you’re not totally swamped you know that loitering thing from Philly back in ‘86 was bullshit

I wrote up a quick blog post about another GitHub incident: surfingcomplexity.blog/2026/10/03/q...

Tragically, I feel like we don’t use “lo and behold” as much as we used to.

New post on a recent GCP outage and the "human error" that "caused" it: surfingcomplexity.blog/2026/10/02/t...

Lorin Hochstein, Staff Software Engineer at Airbnb and previously at Netflix, says every software system has a limit, no matter how well it's designed.

Abstract Data

Bonus blog post for today: my thoughts on yet another GitHub incident. surfingcomplexity.blog/2026/09/19/s...

Today's blog post: a play in one act. surfingcomplexity.blog/2026/09/19/h...

The more powerful the abstraction, the harder it is to debug when it isn’t working as expected.

Reminder that technologists have no special ability to predict the impact of the technologies they build.

I respect my haters and share their concerns

When did we give up on the dream of building undersea colonies?

The bureaucraticization will continue until resilience improves

Why don’t we have falses or truthhoods?

If an entity isn’t the source of truth, does that mean it’s the source of falsehoods?

Tired: anthropomorphization Wired: the intentional stance

I picked up a copy of the collected letters of JRR Tolkien for reasons, and this is the most bullshit table of contents I've ever seen. I'm actively angry that it exists

They call them edge cases because they’ll cut you

There is nothing more predictable than the emergence of a surprising failure mode

If I worked in reliability at an AI company, I would simply instruct the AI not to go down

One property of AI agents that is all-too-human: set the incentives high enough, and cheating becomes inevitable.

AI SRE solutions require a high level of crosscheckability. Their diagnoses need to be easy for humans to cross-check.

Another day, another public incident write-up, another blog post with my thoughts on it. This time, it's @buildkite.bsky.social's excellent recent write-up: surfingcomplexity.blog/2026/09/01/t...

the chair isn't "holding" my ass up, chairs don't "hold" anything. they are merely rigid objects I, the active participant, sit my ass on

One more blog post for today, about the GitHub Actions incident that happened last week: surfingcomplexity.blog/2026/08/29/q...

Wrote a post about some availability risks that I think are fundamentally unavoidable: surfingcomplexity.blog/2026/08/29/o...

for extra safety, consider avoiding computers

Muppet version of The Usual Suspects

The three things I worry about most when it comes to major incidents are databases, networks, and migrations. You’re safest if you can just avoid all three completely.

Data shows that Mark Zuckerberg is the Anti-Dolly Parton -- instead of being adored across the political spectrum like Dolly, he is almost universally reviled by everyone, regardless of political beliefs futurism.com/artificial-i...

Stop, stop, you’re making Edward Teller cry!

Couldn’t ‘Agree to Disagree’ more

Two (unrelated) saturation-related public incident write-ups dropped today! The BuildKite one has a ton of great detail. www.buildkitestatus.com/incidents/tm... www.githubstatus.com/incidents/y1...

looking forward to doom engineering

Tired: loop engineering Wired: doom loop engineering https://finance.yahoo.com/technology/ai/articles/job-seekers-using-ai-apply-173021648.html

Geez, how many times do we Gen Xers have to watch Optimus Prime die?

I think California should rename one of its cities “Ontario” to express solidarity with Canada. Ideally, one with an airport.

"Risks are acceptable, until they aren't" Bureaucratic fantasy documents and a place to park unsolvable issues.

I expected the future to involve more holograms.

Every risk register should have an entry that reads “we have misjudged the risks” which is in the “likelihood: high” and “impact: high” region of the risk matrix.

At this point, I'd settle for plain old human general intelligence.

Sometimes I want to write a post, and then I remember that I already wrote it a couple of years ago. So I'll just re-up this one: surfingcomplexity.blog/2024/02/17/w...