Engineer. Systems thinker. All things observability. Observability is my ikigai.

San Francisco, CA
Hits too close to home.
Interviewing a staff engineer in 2026
191
Suman Karumuri retweeted
There's been a ton of talk about the role of humans in code review, and when and how humans should be signing off on changes. I believe that, long-term, humans have no role in routinely reviewing code.
137
168
1,715
507,497
100% you don’t need mcp.
Rant time 😈 I implemented and shipped support for MCP at Clerk, I contributed to the MCP spec and SDKs, and I was involved in steering the direction of the protocol in the early days, even attending in-person meetings about it with a variety of industry players. I am now very strongly against MCP and advise against using it both as a consumer and as a provider. Unfortunately, I don't think there's any going back at this point, but I feel sadly confident that it will waste countless hours of everyone's time over the coming years, when a much easier and better answer is right under our noses. Let me tell the full story.
2
186
Suman Karumuri retweeted
Bad news: Fable nuked my entire dev machine Claude decided to test a sandbox it was building by running `rm -rf` on my home directory The sandbox didn't work. It's all gone
792
335
5,711
1,271,493
Ending reservations altogether is a very tough political sell. A better idea: reservation should be a ladder, not an inheritance. Within reserved categories, prioritize first-gen beneficiaries. If a family has already benefited, the next opportunity should go one who hasn’t .
5
306
Suman Karumuri retweeted
I asked Fable how hard these problems are, and its response is worth reading. “On the Fields Medal scale, any single one of these…would plausibly anchor a medal case” It’s crazy to see this happening
An internal version of Astra, @OpenAI’s next major model family, solved 10 major open problems in mathematics, quantum complexity, and theoretical computer science. We believe it will be a major step for scientific reasoning. openai.com/index/ten-advance…
95
286
4,015
860,652
Broken search is driving me insane! I tried clearing up the search index and reindexing. Tried several other fixes. Nothing works! At this point, leaning towards vibe coding Mac search bar.
Mac search has been broken for 2 years, and nobody at Apple seems to care. - I open spotlight search - Type "calc" - Calculator appears - I hit return But in the 0.2 seconds before I hit return, it swaps to a random file with the word "calc" in it somewhere. And I open a file I didn't want. This happens to me every single day. Very unbecoming of a $5T company whose customers are still obsessed with its products.
1
3
246
Suman Karumuri retweeted
quick update on how this is going: they have gone back to linear because maintaining their internal tool that they vibecoded was taking away from their actual work’s bandwidth.
an employee at this startup that my sister works at, made their own JIRA - they've all fully discarded JIRA/Linear/Trello, and are using their own tool. It's faaar better than JIRA and is just as detailed. and this person is not even a developer, he leads QA.
234
518
9,879
2,340,372
Suman Karumuri retweeted
Its funny how everyone's idea of heaven is a complete untouched paradise, crystal clear waters, lush green forests, enough food for everyone, classless, moneyless, stateless, a Utopia of peace, unity andhappiness. But here on earth it's considered a leftist ideology.
947
14,743
93,133
905,259
Suman Karumuri retweeted
anthropic including fable in the max plan
🅶🅾🆁🅺
84
323
6,537
617,231
There will only be so many that - Truly believe in you - Take a bet on you - Help you for no real reason - Are your champion - Promote you - Have your back - See you are great That doesn't mean they are perfect In fact, they are the ones that may also be hardest on you
2
24
79
Suman Karumuri retweeted
"mom, how did we get so poor?" "your father had Claude Max, ChatGPT Pro, Cursor Pro and shipped absolutely nothing"
288
907
13,526
717,797
Suman Karumuri retweeted
Turns out black mirror was a very good source for startup ideas.
88
113
2,125
93,455
So an internal wiki?
Yes. You need a centralized way to host and version skills.
79
Suman Karumuri retweeted
you can outsource your thinking but you cannot outsource your understanding
294
4,276
19,118
3,117,675
In KalDB.com, indexing and serving are separate. You can rebuild indexes offline without impacting production. Worst case: slightly stale data. Best case: the system stays available. We don’t need faster reindexing. We need architectures where it doesn’t matter.
80
When it falls behind, you don’t degrade. You break. In 2026, this shouldn’t be acceptable. Indexing should be offline, versioned, swappable, and completely decoupled from serving.
58
GitHub has been down for over a day… because it’s reindexing. This is a failure mode we’ve normalized. We built systems where data is written once, indexed elsewhere, and the product depends on that index being perfectly up to date.
80
Suman Karumuri retweeted
Honestly, this is the most accurate diagram I've seen. Waterfall: You plan for 18 months and deliver exactly what nobody needs anymore. Agile: You deliver something usable at every step, but the CEO keeps asking, "Where's the car?" AI: You get the car on day one. It has six wheels, the doors are on backwards, and it has a rocket launcher. You spend more time making it yours than actually "building"; it's shaping. owning. verifying. That's what the best AI developers do now. They don't build. They shape and own.
130
1,658
8,462
785,495
Truly insane hack.
URGENT PSA - New supply chain attack vector that I found WILD > AI LLMs hallucinate package names roughly 18-21% of the time. Hackers have started pre-registering those hallucinated names on PyPI and npm with malicious payloads; they call it "slopsquatting" You can only imagine what's next
Community note
The 'slopsquatting' attack vector was documented as early as April 2025 and not newly discovered. The cited 18-21% package hallucination rate applies to open-source LLMs; commercial models average 5.2% according to the referenced study using pre-2025 models. socket.dev/blog/slopsquat… arxiv.org/pdf/2406.10279
77