Dmitry Ishkov ๐Ÿค๐Ÿ’™๐Ÿค

3K posts

Dmitry Ishkov ๐Ÿค๐Ÿ’™๐Ÿค banner
Dmitry Ishkov ๐Ÿค๐Ÿ’™๐Ÿค

Dmitry Ishkov ๐Ÿค๐Ÿ’™๐Ÿค

@swe_dima

Software Engineer on the go ๐Ÿ’ป๐Ÿƒ๐ŸŽ’ #NoToWar

Remote Se uniรณ Temmuz 2010
625 Siguiendo127 Seguidores
Dmitry Ishkov ๐Ÿค๐Ÿ’™๐Ÿค
Hard to grasp how 100 years ago most people didnt know how to read/write Then you visit another country & after a week you can say 10+ words. But if that language has another alphabet, good luck trying to read it I cant read anythin in Georgian/Chinese if my life depended on it
English
0
0
0
10
Dmitry Ishkov ๐Ÿค๐Ÿ’™๐Ÿค
Sitting next to a table with other Russians and overheard from a grandma that she turned a VPN on her phone to check messages. Wild side effects of online censorship in Russia.
English
0
0
0
17
Dmitry Ishkov ๐Ÿค๐Ÿ’™๐Ÿค
Travelling through China - and I can't agree more on the "context is king". The new "ask maps" within Google Maps is absolutely useless since it just doesn't have the data You can't expect LLMs to learn everything as a part of training - so whoever provides more context - wins
English
0
0
0
22
Dmitry Ishkov ๐Ÿค๐Ÿ’™๐Ÿค
I think that's one of secrets to productivity. In the world of LLM it's as relevant as ever. Instead of prompting to make yet-another-hotfix feature it's important to take a step back and see how to make customer experience wholesome in the long run.
Franรงois Chollet@fchollet

Good design is the art of packing 1,000 "hows" into a single "what". Good design is compression: making the numerator trend towards infinity while the denominator stays at 1.

English
1
0
0
33
Dmitry Ishkov ๐Ÿค๐Ÿ’™๐Ÿค
Apple: you can't make significant changes to apps after a review. Wechat: ... (it's pretty much its own internet where each shop can make its own widget)
English
0
0
0
27
Dmitry Ishkov ๐Ÿค๐Ÿ’™๐Ÿค
2. you spend your brain energy going through if-else's and data flow, check how it looks on mobile, and other quirks. Because if something goes wrong I will be hearing from customer support or stumble on it in error reporting. 2nd guy might be smart, but I will never hire him.
English
0
0
0
39
Dmitry Ishkov ๐Ÿค๐Ÿ’™๐Ÿค
in reviewing other's code I generally stumbled into 2 types of people: 1. you look thorough a PR for serious issues and fundamental misconceptions and let that person ship. If something gone wrong you see a PR with a fix later.
English
1
0
0
32
Dmitry Ishkov ๐Ÿค๐Ÿ’™๐Ÿค
Been replacing external services with in-house tools long before LLMs (for 12+ years) - different customer support channels, issue trackers, analytics etc It's almost always superior to external ones by deeper integration into your system, instead of trying to fit the unfittable
Guillermo Rauch@rauchg

Almost every SaaS app inside Vercel has now been replaced with a generated app or agent interface, deployed on Vercel. Support, sales, marketing, PM, HR, dataviz, even design and video workflows. Itโ€™s shocking. The SaaSpocalypse is both understated and overstated. Over because the key systems of record and storage are still there (Salesforce, Snowflake, etc.) Understated because the software we are generating is more beautiful, personalized, and crucially, fits our business problems better. We struggled for years to represent the health of a Vercel customer properly inside Salesforce. Too much data (trillions of consumption data points), the ontology of Vercel was a mismatch to the built-in assumptions, and the resulting UI was bizarre. We generated what we needed instead. When you donโ€™t need a UI, you just ask an agent with natural language. Weโ€™ve also been moving off legacy systems with poor, slow, outdated, and inconsistent APIs, as well as just dropping abstraction down to more traditional databases. UI is a function ๐‘“ of data (always has been), and that ๐‘“ is increasingly becoming the LLM.

English
0
0
0
35
Dmitry Ishkov ๐Ÿค๐Ÿ’™๐Ÿค
gave another try to Claude Code Web - unfortunately, still of limited use for bigger repos, since it doesn't support running inside Docker. Without running the code you can't run tests, and without the feedback loop it's just not very good. cc @bcherny
English
0
0
0
41
Dmitry Ishkov ๐Ÿค๐Ÿ’™๐Ÿค
@rails Kamal was a nice experiment - I ran it in production for over a year without issues, but Kubernetes simply has better tooling and observability. And it's easy to add more services to it. So deleting the Kamal VM today.
English
0
0
0
24
Dmitry Ishkov ๐Ÿค๐Ÿ’™๐Ÿค
I know that @rails Kamal is supposed to make things simple, but I just set up another Kubernetes cluster in a matter of hours with full CI / CD with autoscaling and different automation scripts. Slightly more effort for bigger return.
English
1
0
0
42
Dmitry Ishkov ๐Ÿค๐Ÿ’™๐Ÿค
I need to think about physical limitations. Making another DB query is a matter of 10ms. Making another LLM call, however small, is 300+ms. Chain a few of those and users would stare at loading screens all day.
English
0
0
0
24
Dmitry Ishkov ๐Ÿค๐Ÿ’™๐Ÿค
For an app I've been building that heavily uses LLM, I can't help but think how closely it resembles a manufacturing business. Like if building a car, I think how much each part costs, I meticulously count prompt tokens and try to choose properly priced models for each subtask.
English
1
0
1
42