Felipe O. Carvalho @_Felipe

@ApacheArrow / Databases / Compilers. (past @SDFLabs, VoDa, @Spotify). Rust/C++/TLA⁺🇧🇷 → 🇸🇪 → 🌎 https://t.co/vxbdByfADI felipe.rs Joined June 2008

Tweets

28K
Followers

5K
Following

2K
Likes

37K

Felipe O. Carvalho @_Felipe

9 hours ago

@floydophone Being able to actually ship very high-quality products?

2 0 1 75 0

View Details

Microservices forced your teams to modularize, and isolate, so they stop fighting about every impl. detail. The overhead was worth because it enabled new hires to add incremental value. Does this still make sense in the era of agents? Maybe it still does because of the isolation.

Chris Munns @chrismunns

a day ago

microservices were a mistake. im sorry. just monolith all your shit until forever. shard your db when you are ready.

77 75 2K 170K 173

15 1 38 7K 7

View Details

Felipe O. Carvalho @_Felipe

10 hours ago

@CorvusCrypto Smaller teams and generous token budgets might be the next scaling formula.

0 0 2 24 0

View Details

Felipe O. Carvalho @_Felipe

10 hours ago

@CorvusCrypto I have been in so many meetings trying to get upstream services to do a thing only they could do and would never do. Microservices are frustrating. In the modularized-monolith world anyone can get agents to mess with other team’s code and that is also stressful.

1 0 2 72 0

View Details

Felipe O. Carvalho @_Felipe

18 hours ago

@deterministic_d Yes, but at least you don’t have to deal with dependency version hell.

0 0 0 160 0

View Details

Felipe O. Carvalho @_Felipe

20 hours ago

@dillon_mulroy Rust is C++-flavored OCaml

1 1 11 1K 0

View Details

Felipe O. Carvalho @_Felipe

20 hours ago

@thdxr React was hated when it was launched. Which completely puzzled me. Then it succeeded as it deserved. Then a completely new generation of haters showed up, but it is too established at this point.

0 0 2 433 0

View Details

Felipe O. Carvalho @_Felipe

2 days ago

@karanvasudeva @readwithai Exactly! The bottlenecks move somewhere else. Including users.

0 0 1 19 0

View Details

Felipe O. Carvalho @_Felipe

2 days ago

LLMs are not an excuse to scope-creep, but it's so tempting. Too often, tooling improvements are completely offset buy exuberant greed.

5 1 20 1K 3

View Details

Felipe O. Carvalho @_Felipe

2 days ago

@VictorTaelin Relativization of tragedy is cringe, but in this case I feel very much compelled to say that much worse things have happened and are happening to humanity

1 0 15 2K 0

View Details

Felipe O. Carvalho @_Felipe

2 days ago

@PThorpe92 @AvgDatabaseCEO AvgDB is unique and plays on its own league

0 0 2 54 0

View Details

Felipe O. Carvalho @_Felipe

2 days ago

@Valen_UMR To disable the implicit default constructors that are generated. Rust doesn't have default constructors, so you opt-in by #[derive(Default)]. Which Claude happily adds all the time.

0 0 0 34 0

View Details

Felipe O. Carvalho @_Felipe

3 days ago

C++ = we just auto-generate a bunch of methods for you Rust = be explicit about the operators you want, no implicits Claude = Eq, PartialEq, Ord, PartialOrd, Debug, Clone, Default, …

Felipe O. Carvalho @_Felipe

3 days ago

@charliermarsh The problem with these being added automatically is not the compile time or binary size (linker removes unused symbols m), but the fact that they make things that shouldn’t be Hash/Eq/… implement them naively.

1 0 17 16K 1

6 2 56 14K 19

View Details

Felipe O. Carvalho @_Felipe

2 days ago

*by

0 0 0 161 0

View Details

Felipe O. Carvalho @_Felipe

2 days ago

This hasn't been an exploitable advantage for ages now. I mean, months.

Roy @usr_bin_roygbiv

2 days ago

You can look so much smarter than your coworkers purely by using a better model than them it's kind of funny

28 19 831 27K 81

0 0 10 1K 0

View Details

Felipe O. Carvalho @_Felipe

3 days ago

@badlogicgames And they were a thing for like 2 days

0 0 2 402 0

View Details

Felipe O. Carvalho @_Felipe

3 days ago

@adhiadhiadhiadh @geistgrail The linker you’re running will discard the unused symbols, but when I declared something pub in my crate and released it, I gave you the RIGHT to be using it. I can’t change the source code without risking breaking your build.

0 0 2 97 0

View Details

Felipe O. Carvalho @_Felipe

3 days ago

@kamikaz1_k @garybernhardt The caching means you start the agents by quickly copying the prefix of the conversation but you still need the destination memory, right? Because the working set of an agent is the entire context (?). If the sub-agents go for too long, don’t you keep using all that memory?

1 0 0 31 0

View Details

Felipe O. Carvalho @_Felipe

3 days ago

@LewisCTech If your struct is pub, game over. The generated code is now a public interface you have to provide to your users.

0 0 1 506 0

View Details

Felipe O. Carvalho @_Felipe

3 days ago

@kamikaz1_k @garybernhardt An LLM works by calculating over the entire context so far. Passing all the context is expensive. Passing small bits of relevant context (what is done for tool calling) is imperfect.