think(x) retweeted
We’re extremely excited to announce that Xata is joining ClickHouse!
Blog post: xata.io/blog/xata-is-j…
And the most important details are here in thread.
Blog post: xata.io/blog/xata-is-j…
And the most important details are here in thread.
5
12
76
10,366
think(x) retweeted
One of the fastest ways to size up an engineer is to listen to the first question they ask about a new service.
Junior engineers ask “which database?”
Senior engineers ask “how does the data move?”
Architects ask “what happens when it breaks?”
Anyway… A few months back, an s advising dropped this in the design review channel:
“Should we use Postgres or Mongo for the new service?”
He’s one of the sharpest coders I know, and that made it worse.
I’m a patient guy. I’ve answered this question more times than I can count. But this was the third service in a row where the team picked a database before anyone wrote down a single read or write.
It was also the third time they’d end up bolting on a cache, a read replica, and a migration plan six months later.
This isn’t 2006. The year is 2026. Managed Postgres exists. DynamoDB exists. Spanner exists. Cassandra, ClickHouse, Redis, all one API call away.
Choosing an engine is the easy part. Anyone who starts there skips the hard part.
OK, I’ll stop ranting and get to the question he really wanted to ask:
“How do I pick a database I won’t regret in a year?”
At a high level, every database choice comes down to two things.
First, the workload. How does your service read and write data? What’s the read-to-write ratio at peak? Key lookups, range scans, or joins? How much staleness can you tolerate? One region or many?
Second, the team. What can your on-call engineer actually run at 3 a.m.? When nobody has operated the “better” database, it becomes the riskier one.
Everything after that is picking the engine that fits.
Here’s how each workload answer changes the outcome:
Write-heavy → log-structured storage beats B-tree indexes
Range scans → an ordered key store beats a hash-partitioned one
Strong consistency → single-leader with synchronous replication, or consensus
Multi-region writes → conflict resolution becomes your design problem
High staleness tolerance → a cache absorbs the load, and the primary choice matters less
Next time someone asks “Postgres or Mongo?”…
Don’t answer.
Ask them how the data moves.
The simple rule I follow always is: Workload first. Engine second.
Junior engineers ask “which database?”
Senior engineers ask “how does the data move?”
Architects ask “what happens when it breaks?”
Anyway… A few months back, an s advising dropped this in the design review channel:
“Should we use Postgres or Mongo for the new service?”
He’s one of the sharpest coders I know, and that made it worse.
I’m a patient guy. I’ve answered this question more times than I can count. But this was the third service in a row where the team picked a database before anyone wrote down a single read or write.
It was also the third time they’d end up bolting on a cache, a read replica, and a migration plan six months later.
This isn’t 2006. The year is 2026. Managed Postgres exists. DynamoDB exists. Spanner exists. Cassandra, ClickHouse, Redis, all one API call away.
Choosing an engine is the easy part. Anyone who starts there skips the hard part.
OK, I’ll stop ranting and get to the question he really wanted to ask:
“How do I pick a database I won’t regret in a year?”
At a high level, every database choice comes down to two things.
First, the workload. How does your service read and write data? What’s the read-to-write ratio at peak? Key lookups, range scans, or joins? How much staleness can you tolerate? One region or many?
Second, the team. What can your on-call engineer actually run at 3 a.m.? When nobody has operated the “better” database, it becomes the riskier one.
Everything after that is picking the engine that fits.
Here’s how each workload answer changes the outcome:
Write-heavy → log-structured storage beats B-tree indexes
Range scans → an ordered key store beats a hash-partitioned one
Strong consistency → single-leader with synchronous replication, or consensus
Multi-region writes → conflict resolution becomes your design problem
High staleness tolerance → a cache absorbs the load, and the primary choice matters less
Next time someone asks “Postgres or Mongo?”…
Don’t answer.
Ask them how the data moves.
The simple rule I follow always is: Workload first. Engine second.
18
8
134
7,259
think(x) retweeted
every AI lab press release:
"our most capable model yet"
yes, that's how version numbers work
"our most capable model yet"
yes, that's how version numbers work
1
1
5
441
think(x) retweeted
Strange for me to say but if you have an SDK and it doesn't start onboarding with this kind of "copy a prompt for your agent to set it up" it's unlikely your SDK will see rapid adoption
This is from Jev
The "best practices" for software changed very fast
This is from Jev
The "best practices" for software changed very fast
66
26
432
23,729
brutal .. “Tesla wanted to unseat the Ford F-150, instead it made the Edsel look successful.”
arstechnica.com/cars/2026/11/t…
arstechnica.com/cars/2026/11/t…
0
1
1
106
think(x) retweeted
Do some cool research, get $25k free Modal credits! Really looking forward to see what comes out of this.
Over the past few months, our team visited universities to meet students doing AI research. We left convinced that some of the most original thinking in the field is coming from undergrads.
We met students studying how models reason, building new evaluation methods, and applying ine. Their ideas were ambitious and their research was rigorous. And nearly all of them were held back by the same thing: compute.
The GC Research Fellowship, in partnership with @modal, gives 20 exceptional undergraduates up to $25,000 each in compute to pursue their biggest questions.
Applications are open: generalcatalyst.com/research-fello…
3
8
99
11,257
such an epic talk.. wish the lighting was a bit better..
enjoyed being in the room and experiencing @bcantrill’s “reality distortion field”.. thanks for hosting @datadoghq and @rust_nyc
enjoyed being in the room and experiencing @bcantrill’s “reality distortion field”.. thanks for hosting @datadoghq and @rust_nyc
Video for my @rust_nyc talk, "Code Matters: Reflections on Rust in the LLM Age", is now up:
youtube.com/watch?v=5D7QPO…
2
1
5
1,136
think(x) retweeted
#Roma metterà a disposizione 10mila biglietti gratuiti per gli spettacoli del Teatro dell’Opera! Perché la cultura è un diritto di tutti!
51
204
1,870
54,224
“India’s claims of a dramatic decline in poverty rest on questionable comparisons, methodological gaps and an economic record that tells a rather different story.
Poverty cannot be measured away when hardship remains visible on every street.”
360info.org/poverty-in-ind…
Poverty cannot be measured away when hardship remains visible on every street.”
360info.org/poverty-in-ind…
0
0
0
39