Videos CS5Cmz5FssI

How AI is changing Software Engineering: A Conversation with Gergely Orosz, @pragmaticengineer

index_state ready data_status ok

AI Engineer· published 2026-04-21· 0:26:42· en-US· indexed 2026-08-10 19:55

Open on YouTube

72 shot(s).

keyframes kept every frame deduplicated

What was stored

cues: 355; whisperx 355
chunks: 48; from 355 cues
keyframes: 66; kept of 72 captured
frames with text: 65; 714 lines read
chapters: 7; from the source metadata
keyframe bytes: 7.4 MB; word timings on 355 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage	state	model	started	took
`fetch`	done	—	2026-08-10 18:16	2m 30s
`stt`	done	—	2026-08-10 18:19	33s
`chunk`	done	—	2026-08-10 18:19	0s
`text_embed`	done	—	2026-08-10 19:54	0s
`keyframe`	done	—	2026-08-10 18:20	2m 35s
`ocr`	done	—	2026-08-10 18:22	14s
`frame_embed`	done	—	2026-08-10 19:54	12s

Frames, and what the machine read

0:03 #0 done2 line(s)

shot 0·sharpness 658.4
1. AlEngineer0.98
2. EUROPE1.00
0:08 #1 done2 line(s)

shot 1·sharpness 827.9
1. PRESENTINGSPONSOR1.00
2. Google DeepMind1.00
0:13 #2 done3 line(s)

shot 2·sharpness 907.6
1. PLATINUM SPONSORS0.98
2. # Braintrust0.96
3. WorkOS OpenAI0.95
0:17 #3 done7 line(s)

shot 3·sharpness 477.9
1. AUTHOR1.00
2. THE PRAGMATIC0.98
3. mat1.00
4. Engineer1.00
5. EUROPE1.00
6. ESENTED BY1.00
7. le DeepMind0.99
0:22 #4 done6 line(s)

shot 4·sharpness 537.2
1. FUUNDER0.91
2. AUTHOK0.64
3. IEngineer0.97
4. gmatic En1.00
5. ENDINEER0.97
6. SASNATIC0.74
0:45 #5 done23 line(s)

shot 5·sharpness 755.9
1. st1.00
2. AlEngineer0.99
3. Google DeepMind1.00
4. AlEngineer0.97
5. Oper1.00
6. EUROPE1.00
7. EUROPE1.00
8. M1.00
9. oft1.00
10. AIEngineer0.97
11. neo0.93
12. AIEngin0.97
13. EUROPE1.00
14. EUROPE1.00
15. lev1.00
16. WorkOS1.00
17. AIE1.00
18. ENSIMATIC0.77
19. Te1.00
20. PE1.00
21. AlEngi0.93
22. AIEngin0.95
23. EUROPE1.00
1:10 #6 done6 line(s)

shot 6·sharpness 356.1
1. AlEngineer0.99
2. EUROPE1.00
3. Microsoft0.99
4. THE PRAGMATIO0.95
5. ENGINEER1.00
6. Engineer1.00
1:36 #7 done6 line(s)

shot 7·sharpness 360.9
1. AlEngineer0.99
2. EUROPE1.00
3. Microsoft1.00
4. THE PRAGMATIL0.99
5. ENGINEER1.00
6. Engineer1.00
1:51 #8 skipped

shot 8·duplicate of #6
2:17 #9 done6 line(s)

shot 9·sharpness 365.0
1. AlEngineer0.96
2. EUROPE1.00
3. Microsoft1.00
4. THE PRAGMATIC1.00
5. ENGINEER1.00
6. Engineer1.00
3:05 #10 done6 line(s)

shot 10·sharpness 357.2
1. AlEngineer0.98
2. EUROPE0.99
3. Tessl1.00
4. THE PRAGMATIC1.00
5. ENGINEER1.00
6. Engineer1.00
3:36 #11 done18 line(s)

shot 11·sharpness 736.8
1. ai1.00
2. AlEngineer0.98
3. WorkOS1.00
4. AIEngineer0.96
5. Open1.00
6. EUROPE1.00
7. EUROPE1.00
8. AlEngineer0.96
9. Micro0.90
10. AIEngine0.91
11. EUROPE1.00
12. EUROPE1.00
13. Braintrust1.00
14. AIEn0.89
15. SUSEN0.75
16. EURUM0.92
17. AIEngine0.95
18. EUROPE1.00
4:08 #12 done6 line(s)

shot 12·sharpness 365.5
1. AlEngineer0.98
2. EUROPE1.00
3. Tessl1.00
4. THE PRAGMATIC0.97
5. ENGINEER1.00
6. Engineer1.00
4:22 #13 done6 line(s)

shot 13·sharpness 358.2
1. AlEngineer0.98
2. EUROPE1.00
3. Tessl1.00
4. THE PRAGMATIC1.00
5. ENGINEER1.00
6. Engineer1.00
4:45 #14 done20 line(s)

shot 14·sharpness 739.8
1. AlEngineer0.97
2. Braintrust1.00
3. AlEngineer0.97
4. Wor1.00
5. EUROPE1.00
6. EUROPE1.00
7. OAI0.95
8. AlEngineer0.93
9. Google DeMind0.96
10. AIEngine0.92
11. EUROPE1.00
12. EUROPE1.00
13. ari1.00
14. Microsoft1.00
15. AIF1.00
16. .OPL0.99
17. ALEngi1.00
18. AIEngine0.91
19. WBEUR0.71
20. EUROPE1.00
4:56 #15 done22 line(s)

shot 15·sharpness 749.6
1. AlEngineer0.97
2. Braintrust1.00
3. AlEngineer0.95
4. Wor1.00
5. EUROPE1.00
6. EUROPE1.00
7. Op0.93
8. AlEngineer0.98
9. Google De1.00
10. nd1.00
11. AlEngine0.93
12. EUROPE1.00
13. EUROPE1.00
14. ari1.00
15. Microsoft1.00
16. THE PRAGNATIC0.95
17. ENOINEER0.88
18. EUROPE1.00
19. ALEngin0.93
20. AIEngine0.94
21. NBEUR0.73
22. EUROPE1.00
5:07 #16 done12 line(s)

shot 16·sharpness 217.4
1. Tessl1.00
2. A0.96
3. AlEngineer0.98
4. World's Fair0.98
5. Microsoft1.00
6. aws0.87
7. SAN FRANCISCO1.00
8. Cloud1.00
9. Λarize0.91
10. Mongol1.00
11. A0.97
12. Aps://AI.ENGINEER0.92
5:17 #17 done17 line(s)

shot 17·sharpness 218.7
1. Tessl1.00
2. A0.99
3. AlEngineer0.99
4. M0.91
5. OpenAl0.88
6. World's Fair0.95
7. ADERT0.92
8. Microsoft1.00
9. ogle Cloud0.95
10. SAN FRANCISCO1.00
11. Λarize0.96
12. OctoAI0.80
13. MongoDB0.96
14. 2024P0.89
15. A0.96
16. neo4j0.90
17. hilps://AI.ENOINEER0.85
5:30 #18 done17 line(s)

shot 18·sharpness 733.4
1. AlEngineer0.96
2. Braintrust1.00
3. AIEngineer0.95
4. Wor1.00
5. EUROPE1.00
6. EUROPE1.00
7. AlEngineer0.97
8. Google Deepd1.00
9. AIEngine0.92
10. EUROPE1.00
11. EUROPE1.00
12. ari1.00
13. Microsoft1.00
14. AIEr0.90
15. E1.00
16. AIEngine0.93
17. EUROPE1.00
5:52 #19 done7 line(s)

shot 19·sharpness 356.8
1. AlEngineer0.98
2. ev0.97
3. EUROPE1.00
4. :neo4j0.88
5. THE PRAGMATIC1.00
6. ENGINEER1.00
7. Engineer0.96
6:25 #20 done7 line(s)

shot 20·sharpness 356.6
1. AlEngineer0.97
2. ev0.97
3. EUROPE1.00
4. :neo4j0.94
5. THE PRAGMATIC1.00
6. ENGINEER1.00
7. Engineer1.00
6:36 #21 done4 line(s)

shot 21·sharpness 352.5
1. AlEngineer0.97
2. EUROPE1.00
3. comet1.00
4. Engineer1.00
7:06 #22 done6 line(s)

shot 22·sharpness 353.7
1. AlEngineer0.97
2. EUROPE1.00
3. comet1.00
4. THE PRAGMATIC1.00
5. ENGINEER1.00
6. Engineer1.00
7:24 #23 done4 line(s)

shot 23·sharpness 344.4
1. AlEngineer0.99
2. EUROPE1.00
3. comet1.00
4. Engineer1.00

Transcript

355 cues· 4,997 words· 26,936 chars

chunk 0 · 0:21–1:00 · 114 words · 596 chars
0:21 All right, I'm gonna assume most of you, show of hands who subscribes to Pragmatic Engineer?
0:26 Oh my god.
0:29 He is, he needs no introduction then.
0:32 Let's get right into it.
0:35 What is token maxing and should everyone here be doing it?
0:41 So I heard about token maxing a week ago, or like week and a half ago first, and you know, some people have been doing it for longer, and I tweeted about it, I think three days ago, saying, oh, there's this token maxing, and again, you see it on social media, and my DMs were blowing up from people at large companies, I don't wanna name names, but like Meta, Microsoft,
chunk 1 · 1:01–1:46 · 127 words · 684 chars
1:01 Some other ones as well, like the likes of IGN and so many more.
1:07 And the story is a little bit different at every company on why people are doing it and whether they like it or whether they think it's good, but there's a few common things.
1:17 One is token output at these larger companies is measured in some way.
1:23 There's like either a leaderboard or there's a way to look up your peers.
1:27 Salesforce, for example, you can check the spend
1:31 the money spent that every person at the company did.
chunk 2 · 1:34–2:05 · 91 words · 505 chars
1:34 You can search in a tool that someone built and it shows how many dollars they spent on AI-related tokens.
1:41 And first there's this number, then there's this uncertainty in the tech industry, right?
1:46 We're kind of hearing layoffs, like massive cuts of the likes of Block,
1:50 And I mean, they're like, no matter how much tokens people spend, they were let go independent of this, but people start to think like, does, is it part of performance evaluations or promotions or all that?
2:01 And the answer is kind of so.
chunk 3 · 2:05–2:47 · 149 words · 795 chars
2:05 inside of meta I talk with managers and in the performance evaluation they have this data point which is one of many data points right the same way as as like diffs or impact or or code reviews of how helpful this person is but they do just like with any data points they sometimes pull it and use it so typically
2:25 And just like any data point, it can be weaponized.
2:27 So a low performer with low impact and a low token count, clearly not even trying.
2:33 And a high performer with high impact and high token count, clearly that's innovating, and this must be doing good.
chunk 4 · 2:38–3:18 · 144 words · 799 chars
2:38 So inside of these companies, specifically, I talk with a lot of people at Meta.
2:41 And again, this is not representative of 100% of Meta.
2:44 But they had this leaderboard where people showed up, and they have massive amounts of tokens.
2:48 And a lot of engineers got just scared, worried.
2:51 So they started TokenMax to try to generate tokens.
2:54 Stories that I've heard first or second hand from these people who told me firsthand is, for example, instead of reading the documentation, I will ask the agent to summarize it for me and ask questions, even though it doesn't do a good job answering it, but my token count goes up.
chunk 5 · 3:09–3:50 · 149 words · 844 chars
3:09 People just want to not be on the bottom 25% or bottom 50% for token count where these things are measured.
3:15 Inside of Microsoft, again, there's a leaderboard.
3:17 And I'm talking to people.
3:18 They're like, it's ridiculous how some people are just running autonomous agents to build junk, honestly, for the sake of having that number go up.
3:27 And sometimes it gets ridiculous, because inside of Meta, they had this leaderboard.
3:31 They got rid of it after an article came out, and it looked to me stupid, honestly.
3:35 So whoever built it just closed it down.
3:38 People are still token maxing, by the way, because there's this thinking that it might have gone, but we're engineers.
chunk 6 · 3:43–4:15 · 115 words · 629 chars
3:43 And don't forget, these are high-paying jobs, right?
3:45 You don't really want to lose a job over something stupid as you didn't have a high enough token count.
3:49 And that's how it feels.
3:50 But inside Salesforce, there's a target of minimum spend per month.
3:55 I think it's like $175 between things.
3:58 So people are like, again, beginning of the month, just token max to get there.
4:03 So it's weird.
4:04 And it started as a joke earlier.
4:05 A few months ago, token maxing was really just people going crazy and enjoying this thing and building cool stuff.
4:10 But it's kind of turned into, in a lot of companies, I think it's just a culturally weird thing.
chunk 7 · 4:16–5:00 · 130 words · 753 chars
4:16 It's a weird time to be in, because I remember lines of code used to be, when early developer productivity tools came out, like Velocity and Pluralsight Flow, they kind of measured lines of code and number of QPRs.
4:28 And we know that was stupid, and people kind of optimized for that at companies that did it.
4:33 But it's almost like now it's the top-running companies, like Meta and Microsoft, who are incentivizing people to do just stupid stuff, honestly.
4:42 Yeah, those are wild stories.
4:44 And one of the things, you're clapping for that.

Chapters

0:00 What is token maxing?
5:27 Is AI-driven productivity worth the hype?
12:42 How the role of the software engineer is changing
14:45 Are engineers now engineering managers for AI?
17:31 Large tech infrastructure and internal AI tooling
20:41 Why companies like Shopify invest heavily in AI churn
22:56 Growing The Pragmatic Engineer and finding product-market fit