In late September 2026 I posted a multipart X thread that also covered a range of industry moves: a possible Firebase‑related iOS crash, my personal side‑project history, Meta’s poaching of MongoDB’s CEO CJ Desai, the slow Android launch of Clubhouse, and a guest lecture encouraging CS students to build side projects.
MongoDB’s valuation at the time of CJ Desai’s departure was $35 B.
Desai received a $32 M equity grant when he joined MongoDB.
By leaving after 11 months, he forfeited roughly $25 M of that grant.
A portion of his prior grant—$17.5 M—may now be worth very little.
Clubhouse spent 14 months developing its Android app, finally launching in May 2021.
The launch missed the platform’s hype peak of February 2021.
In my CS sophomore lecture, 20 % of students raised their hands when asked who builds side projects.
I raised a concern that an exclusivity model could lock out >60 % of global smartphone users.
OpenAI opened Codex to external models, allowing enterprises to run GLM‑5.3 Flash and Kimi K3 natively in Codex and count spend against their OpenAI commit.
Anthropic’s Claude Code is locked: “closed everything”, meaning no external model usage is permitted.
The Firebase SDK outage caused a 2 to 6 hour global outage for iOS apps.
Clients that cached the malformed server payload experienced 6 hours for clients that cached the malformatted server payload.
Perf. On iOS, every separately packaged KMP module has its own GC. Can+ does cause slowdowns/deadlocks.
Debugging: debugging into Kotlin, in XCode, is whacky.
The last time building software sped up ~10x (early 2000s, the agile movement), a massive interest in automated testing followed almost immediately (unit tests, TDD, XP etc).
Today, building software sped up 10x, easily: and a similar massive interest in automated testing + verification is following.
OpenAI: spend $$$/month/dev with us. Choose to spend it on our models or 3rd parties, our tools or a long list of standout AI vendors. Your choice!
Anthropic: spend $$$/month/dev with us; use our models and our tools. Forget using them w third parties.
I’m able to use my Codex Subscription in my own harness, and I love this! On the other hand, Claude subscriptions cannot be used outside of CC.
Grok pulled some shady stuff in the past, uploading unencrypted .env files and git history to the cloud, without telling users.
In late September 2026 I posted a multipart X thread that highlighted a 2026‑09‑30 podcast with the co‑creator of GIMP, former Google engineer behind Gmail’s backend, the Colossus storage system and now Cockroach Labs CTO, detailing his career from building GIMP through Google work on google3, Bazel, Spanner and CockroachDB, his B‑tree replacement for std::map (reducing node pointers from 2 to 1, cutting memory by ≈30 % and latency ≈15 %), a Swiss‑Table hash for Go (≈2× faster), Raft‑based strong consistency (99.999 % availability) and AI‑assisted coding guidance.
- Career timeline: built GIMP (04:00), joined Google in 2002 after a 2001 interview invitation, worked on Gmail backend, google3, Bazel‑style build files and the Colossus distributed file system.
- Technical contributions: B‑tree replacement for std::map reduced pointer count per node from 2 to 1; memory footprint ~30 % smaller; lookup latency improved ~15 %. Swiss Table for Go delivered ~2× higher inserts‑per‑second than Go’s native map.
- Distributed system insights: automatic sharding adopted in CockroachDB; Raft consensus provides strong consistency with 99.999 % availability; latency dominates performance, B‑tree improves locality.
- AI‑assisted development: AI can automate routine coding tasks but quality still depends on rigorous engineering practices; caution advised for AI‑generated code.
The very first version of the famous Google logo was made in GIMP, the free, open-source raster graphics and image editing software created by and a college roommate.
In 2001, a person reached out to , invited them to an interview and then made an offer.
While said no back then (thanks to the commute), they joined a year later, in 2002.
At Google, one of their colleagues noticed that std::map showed up in memory profiles.
also did something similar with Go’s map: it was performant, but they built a Swiss Table implementation that was faster.
That implementation later made it into the Go library, with the Go team helping to finish it!
If you squint hard, everything in distributed databases and storage systems starts looking like a B-tree.
There’s even a paper on this phenomenon, The Ubiquitous B-Tree