Videos yUmS-F9IX90

Don't Build Slop (4 Levels of AI Agent Maturity) - Ara Khan, Cline

index_state ready data_status ok

AI Engineer· published 2026-05-19· 0:18:51· en-US· indexed 2026-08-10 23:24

Open on YouTube

65 shot(s).

keyframes kept every frame deduplicated

What was stored

cues: 243; whisperx 243
chunks: 33; from 243 cues
keyframes: 28; kept of 65 captured
frames with text: 28; 503 lines read
chapters: 12; from the source metadata
keyframe bytes: 6.6 MB; word timings on 243 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage	state	model	started	took
`fetch`	done	—	2026-08-10 23:19	1m 41s
`stt`	done	—	2026-08-10 23:21	25s
`chunk`	done	—	2026-08-10 23:21	0s
`text_embed`	done	—	2026-08-10 23:21	1s
`keyframe`	done	—	2026-08-10 23:21	1m 52s
`ocr`	done	—	2026-08-10 23:23	11s
`frame_embed`	done	—	2026-08-10 23:24	5s

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 829.1
1. PRESENTINGSPONSOR1.00
2. Google DeepMind1.00
0:13 #2 done3 line(s)

shot 2·sharpness 907.7
1. PLATINUM SPONSORS0.98
2. # Braintrust0.96
3. WorkOS OpenAI0.95
0:17 #3 done6 line(s)

shot 3·sharpness 556.1
1. Don't Build S1.00
2. (4 Levels of AI A0.97
3. Maturity)1.00
4. AlEngineer0.99
5. EUROPE1.00
6. 20261.00
0:40 #4 done9 line(s)

shot 4·sharpness 2078.3
1. Don't Build Slop0.98
2. (4 Levels of AI Agent0.99
3. ★1.00
4. AIE1.00
5. Maturity)1.00
6. Google DeepMind1.00
7. AlEngineer0.99
8. EUROPE0.96
9. 20260.99
0:46 #5 done8 line(s)

shot 5·sharpness 1844.1
1. The Mass Psychosis Problem1.00
2. AIE1.00
3. ★1.00
4. ★1.00
5. Google DeepMind0.98
6. AlEngineer1.00
7. EUROPE1.00
8. 20260.99
0:53 #6 done8 line(s)

shot 6·sharpness 2129.2
1. ***0.78
2. AIE1.00
3. ★1.00
4. ★1.00
5. ★★0.96
6. Engineering the future of Al1.00
7. AlEngineer1.00
8. EUROPE0.95
1:21 #7 done10 line(s)

shot 7·sharpness 1937.2
1. The Mass Psychosis Problem1.00
2. AIE1.00
3. ★1.00
4. ★1.00
5. ★1.00
6. # Braintrust0.96
7. WorkOS OpenAI0.94
8. AlEngineer1.00
9. EUROPE1.00
10. 20261.00
1:53 #8 done40 line(s)

shot 8·sharpness 2743.4
1. New Agent1.00
2. 3N0.92
3. build this locally0.99
4. Marketplace0.96
5. Environment ready1.00
6. cline/kanban1.00
7. Worked for 3m 22s0.95
8. arafatkatze/about1.00
9. The build completed successfully with no errors. There are no code changes to commit since the build only produced0.99
10. Local build setup1.00
11. 00.65
12. output artifacts that are already listed in .gitignore.1.00
13. Here's a summary of the local build:1.00
14. arafatkatze/arafatk.github.io0.99
15. Environment setup:1.00
16. arafatkatze/arafatkatze.gith...1.00
17. • Installed Node.js v18.9.0 (required by the project's engines field)0.98
18. AIE1.00
19. ★1.00
20. ★1.00
21. arafatkatze/cline1.00
22. cline1.00
23. Build steps completed (pnpm run build ):0.97
24. • Installed pnpm 8.1.1 (required by packageManager field)0.99
25. • Ran pnpm install to install all dependencies0.98
26. ★1.00
27. 1. build:cache - Generated blog/post cache files into public/data/0.98
28. ★1.00
29. ★1.00
30. ★1.00
31. Open Workspace0.97
32. 2. build: feed - Generated RSS/Atom/JSON feed files0.98
33. 3. next build - Compiled the Next.js application (linting + type checking passed with only warnings, no errors)0.99
34. 4. next export - Static HTML export to /workspace/out/ (74 pages, 2,372 total files)0.98
35. 5. next-sitemap - Generated sitemap at https://about.sourcegraph.com/sitemap.xml0.99
36. The build produces only warnings (array index keys, promise-returning functions in void attributes, etc.) but no errors.0.99
37. All output is in .gitignore so there are no files to commit.0.98
38. AlEngineer0.96
39. AlEngineer1.00
40. EUROPE1.00
1:55 #9 skipped

shot 9·duplicate of #8
1:56 #10 done32 line(s)

shot 10·sharpness 1961.2
1. New chat0.98
2. Search1.00
3. 88 Plugins0.98
4. ①Automations0.99
5. hreads1.00
6. cloud-platform1.00
7. claude-code1.00
8. sdk-wip1.00
9. kanban1.00
10. codex1.00
11. AIE1.00
12. core-platform1.00
13. harbor1.00
14. Let's build0.99
15. ★1.00
16. ★1.00
17. ★1.00
18. ★1.00
19. cline0.93
20. cloud-platform1.00
21. Build a classic Snake game in this1.00
22. Create a one-page $pdf that0.99
23. Create a plan to...0.99
24. repo.1.00
25. summarizes this app.0.99
26. Ask Codex anything. @ to add files, / for commands, $ for skilils0.94
27. GPT-5.40.99
28. Extra High1.00
29. AlEngineer0.97
30. AlEngineer1.00
31. EUROPE1.00
32. EUROPE0.97
2:02 #11 skipped

shot 11·duplicate of #8
2:07 #12 skipped

shot 12·duplicate of #8
2:09 #13 skipped

shot 13·duplicate of #10
2:12 #14 done24 line(s)

shot 14·sharpness 3148.7
1. Level1.00
2. What You Do1.00
3. What You Learn1.00
4. 11.00
5. Use a framework1.00
6. How agents work at a surface level1.00
7. AIE1.00
8. ★1.00
9. 21.00
10. Build it yourself as a state machine1.00
11. Architecture, modularity, model-independence0.99
12. ★1.00
13. ★1.00
14. ★1.00
15. 31.00
16. Kanban the flow1.00
17. Observability, debugging, systems thinking1.00
18. 41.00
19. Ship to the cloud0.97
20. Parallelization, scale, production operations1.00
21. AlEngineer0.96
22. AlEngineer1.00
23. EUROPE1.00
24. EUROPE0.97
2:14 #15 done24 line(s)

shot 15·sharpness 1450.4
1. 日0.97
2. Migrating Auth Service to gRPC with Protobuf0.97
3. New0.88
4. . Search0.92
5. Migrate the auth service from REST to gRPC. Keep the REST endpoints working as a falback.0.97
6. SESSIONS1.00
7. acme-productions1.00
8. Let me start by exploring the current auth service structure.0.97
9. Migrating Auth Service to gRPC wit0.97
10. Bulding Multi-Tenant Data Pipeline0.93
11. acme-web1.00
12. 2d0.91
13. AIE1.00
14. Adding Dark Mode Toggle with CSs..0.96
15. 3d0.94
16. ★1.00
17. ★1.00
18. ★1.00
19. ★1.00
20. backend1.00
21. local-demos1.00
22. AlEngineer0.97
23. AlEngineer0.99
24. EUROPE1.00
2:54 #16 done23 line(s)

shot 16·sharpness 3172.0
1. Level1.00
2. What You Do1.00
3. What You Learn1.00
4. Use a framework1.00
5. How agents work at a surface level1.00
6. AIE1.00
7. ★1.00
8. 21.00
9. Build it yourself as a state machine1.00
10. Architecture, modularity, model-independence1.00
11. ★1.00
12. ★1.00
13. ★1.00
14. 31.00
15. Kanban the flow1.00
16. Observability, debugging, systems thinking0.99
17. 41.00
18. Ship to the cloud0.99
19. Parallelization, scale, production operations1.00
20. Engineering the future of Al0.99
21. AlEngineer1.00
22. BUROPE0.90
23. 20260.98
3:01 #17 skipped

shot 17·duplicate of #16
3:25 #18 done17 line(s)

shot 18·sharpness 1997.0
1. Framework1.00
2. Al Agent0.97
3. crewai0.99
4. AIE1.00
5. ★1.00
6. LangChain1.00
7. LangGraph1.00
8. CrewAl1.00
9. AutoGen1.00
10. Llamalndex1.00
11. ★1.00
12. ★1.00
13. Top 5 Agentic Al Agent Frameworks for Agency Automation 20260.99
14. Engineering the future of Al0.99
15. AlEngineer1.00
16. BUROPE0.86
17. 20261.00
3:55 #19 skipped

shot 19·duplicate of #18
4:17 #20 skipped

shot 20·duplicate of #5
4:25 #21 skipped

shot 21·duplicate of #14
4:28 #22 done9 line(s)

shot 22·sharpness 1988.5
1. ★1.00
2. AIE1.00
3. ★1.00
4. ★1.00
5. ★1.00
6. AlEngineer0.96
7. AlEngineer0.99
8. EUROPE1.00
9. EUROPE0.99
4:39 #23 skipped

shot 23·duplicate of #16

Transcript

243 cues· 3,558 words· 18,151 chars

chunk 0 · 0:15–0:59 · 129 words · 679 chars
0:15 Hi there.
0:17 Very nice to see you guys.
0:20 It's so interesting to see.
0:22 It's been such a long day, and you guys still showed up.
0:24 And I'm always very flattered when I'm trying to present something, and people are there.
0:30 It makes you feel like what you do matters.
0:33 So today, the topic that I want to talk about is don't belt slop four levels of AI agent maturity.
0:39 And what I'm trying to do here is that I'm going to talk about something that a lot of people brought this up to me.
0:44 And there's a mass psychosis problem.
chunk 1 · 0:46–1:29 · 136 words · 691 chars
0:46 So basically, it's something like this.
0:48 Every person feels like there's this giant set of robots around all of you.
0:54 They're just breezing through, doing a lot of things.
0:59 And you're in the middle, and you're so confused.
1:01 You're so confused of what should I do?
1:04 Should I have 15 agents ripping through all the time, and I'm just vibing?
1:10 Or should I just be a peasant and just read through every line of code?
1:14 And it's very hard to make up your mind.
chunk 2 · 1:17–2:00 · 164 words · 842 chars
1:17 I feel like you come to a place like this, and you're always like, the FOMO just gets to you.
1:21 You get a panic attack or something.
1:22 But I kind of want to pull back out of this.
1:26 And I want to be like, guys, let's slow down.
1:28 Let's just slow down.
1:29 And let's think through what are the problems that we can actually solve that will actually help you build really useful agents.
1:36 And I want to take care of every end of the spectrum of your necessity to build things really fast, but at the same time, your necessity to, at a certain point, build things slow and take things to production.
chunk 3 · 1:47–2:31 · 159 words · 816 chars
1:47 So that's basically the goal.
1:49 To give an example of mass psychosis, there's a lot of things that are very similar.
1:53 So I'll give you an example of three UIs.
1:56 And these are three frontier labs.
1:58 And I guarantee you, not one of you can predict which one is which.
2:01 One of them is Factory.
2:02 One of them is Codex.
2:03 And one of them is Cursor.
2:05 I am positive none of you know which one is which.
2:08 Even I don't know which one is which.
2:10 But yeah, so that's basically the point.
2:11 Everything is the same.
2:12 You want to do your own thing.
2:13 So to build agents in particular, here's what I would present here, that I want to break down the problem of building agents into four different parts.
chunk 4 · 2:23–3:03 · 144 words · 736 chars
2:23 The first part is just trying to somehow figure out a way to see if this even makes sense, like this even works.
2:30 That's where you use a framework.
2:31 The second part is when you actually are serious, like, OK, I actually want to do something here.
2:36 Now you build things by yourself as a state machine, like building an actual agent.
2:40 The third part is the UX workflow, where you use Kanban, which I suggest is a great form factor to be able to work with agents.
2:48 And the fourth part would be shipping to cloud.
2:50 So this is mostly an outline read of how we want to be able to work with agents.
chunk 5 · 2:55–3:39 · 159 words · 825 chars
2:55 But I think it will give you a good heuristics of how you would solve this problem.
3:00 So level one of building agents is literally just use a framework.
3:04 And I think there's a couple of different frameworks, like link chain, link graph.
3:09 I don't use them.
3:11 I work at client.

Chapters

0:00 Introduction: Addressing the "mass psychosis" of agent-building
2:13 The Four Levels of Agent Maturity overview
3:00 Level 1: Using existing agent frameworks
4:13 Level 2: Building your own agents (with five rules)
4:36 Rule 1: View every agent as a state machine
5:55 Rule 2: Keep it simple (Avoid unnecessary system prompt bloat)
7:20 Rule 3: Integrate with a CLI for a pseudo-RL pipeline
8:43 Rule 4: Don't build "slop" (Focus on thoughtful architecture)
9:43 Rule 5: Master frontier model APIs (Reasoning traces)
11:03 The UX form factor: Why Kanban boards are best for agents
13:12 Shipping to the cloud for scalability
17:15 Q&A session: Planning and state transitions in Kanban