Videos I3znWC3MEXM

It's 10pm. Do You Know Where Your Agents Are? — Kim Maida, Keycard

index_state ready data_status ok

AI Engineer· published 2026-07-20· 0:23:02· en-US· indexed 2026-08-10 19:43

Open on YouTube

59 shot(s).

keyframes kept every frame deduplicated

What was stored

cues: 205; whisperx 205
chunks: 40; from 205 cues
keyframes: 36; kept of 59 captured
frames with text: 36; 529 lines read
chapters: 13; from the source metadata
keyframe bytes: 7.7 MB; word timings on 205 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage	state	model	started	took
`fetch`	done	—	2026-08-10 00:10	2m 07s
`stt`	done	—	2026-08-10 00:12	22s
`chunk`	done	—	2026-08-10 00:13	0s
`text_embed`	done	—	2026-08-10 19:43	0s
`keyframe`	done	—	2026-08-10 00:13	2m 22s
`ocr`	done	—	2026-08-10 00:15	12s
`frame_embed`	done	—	2026-08-10 19:43	6s

Frames, and what the machine read

0:02 #0 done2 line(s)

shot 0·sharpness 453.4
1. AlEngineer0.96
2. World's Fair0.97
0:03 #1 done2 line(s)

shot 1·sharpness 665.7
1. AlEngineer0.95
2. World's Fair0.99
0:10 #2 done24 line(s)

shot 2·sharpness 2743.2
1. LAB & PLATINUM SPONSORS0.98
2. Amazon AGI Lab0.98
3. ANTHROP\C1.00
4. Google DeepMind1.00
5. MINIMAX0.94
6. OpenAI0.92
7. Akamai1.00
8. arize1.00
9. aws1.00
10. Braintrust bright data0.99
11. B1.00
12. Browserbase1.00
13. docker1.00
14. :neo4j0.93
15. ORACLE1.00
16. PayPal1.00
17. qodo1.00
18. reducto1.00
19. Sonar1.00
20. Makers of0.99
21. togetherai1.00
22. Unblocked1.00
23. WorkOS1.00
24. SonarQube1.00
0:12 #3 done2 line(s)

shot 3·sharpness 207.8
1. AlEngineer0.99
2. World's Fair0.99
0:37 #4 done6 line(s)

shot 4·sharpness 892.9
1. AlEngineer0.99
2. World's Fair0.98
3. PRESENTED BY1.00
4. Microsoft1.00
5. World's Fair0.97
6. Engineering the future of Al1.00
1:06 #5 done14 line(s)

shot 5·sharpness 1059.2
1. Agents without access aren't useful1.00
2. AlEngineer0.99
3. World'sFair1.00
4. 0o0.57
5. PRESENTED BY1.00
6. oo0.64
7. 000.91
8. Microsoft1.00
9. 售0.55
10. Keycard1.00
11. XKmMoidekimmaida0.80
12. World'sFair1.00
13. TRACK 5· JUNE 30, 20260.96
14. Security1.00
1:35 #6 done7 line(s)

shot 6·sharpness 786.9
1. Window1.00
2. Help1.00
3. AlEngineer0.99
4. World'sFair1.00
5. World'sFair0.99
6. TRACK 5• JUNE 30, 20260.94
7. Security1.00
1:40 #7 done49 line(s)

shot 7·sharpness 811.8
1. Edit1.00
2. View1.00
3. History1.00
4. Bookmarks1.00
5. Profiles1.00
6. Window1.00
7. Help1.00
8. © Kim's Talk @..-18m left0.90
9. Tue Jun 30 2:51 PM1.00
10. AlEngineer0.99
11. localhost:30011.00
12. 1 Work0.82
13. World'sFair1.00
14. ON-CALL TRIAGE AGENT0.99
15. OPEN 51.00
16. WORKING 60.95
17. TRIAGED S0.99
18. RESOLVED S0.93
19. ESCALATED0.98
20. API KEYS1.00
21. 11:37 PM1.00
22. INCIDENT QUEVE0.99
23. AGENT1.00
24. INC-2289 tod0.91
25. MONITORING QUEUE1.00
26. Ille. The agent picks up new tickets as they arrive.0.97
27. Backup power supply failed in the serv.0.98
28. SHIFT1.00
29. 22:00 - 06:000.97
30. Website security certificate expiring so..0.96
31. INC-2299 today0.95
32. QUEUE0.97
33. AGENT1.00
34. On-Call Triage Agent0.99
35. 5 open tickets1.00
36. INC-2291 today0.98
37. Billing database broken, payments faili..0.95
38. INC-2292 today0.97
39. Orders stuck, main server frozen1.00
40. INC-2294 today0.95
41. Site failing for 1 in 3 visitors1.00
42. REQUEST1.00
43. AUTHORIZATION1.00
44. STATUS1.00
45. TIME1.00
46. Waiting for requests...0.97
47. World's Fair0.98
48. TRACK 5· JUNE 30, 20260.96
49. Security1.00
1:47 #8 done39 line(s)

shot 8·sharpness 759.0
1. ON-CALL TRIAGE AGENT1.00
2. OPEN 51.00
3. WORKING0.99
4. TRIAGED0.99
5. RESOLVED0.95
6. ESCALATED1.00
7. API KEYS0.99
8. 11:37 PM0.99
9. To exit ful creen, press and hold0.97
10. AlEngineer0.98
11. INCIDENT QUEUE1.00
12. AGENT1.00
13. World's Fair0.99
14. INC-2289 today0.97
15. MONITORING QUEUE1.00
16. Idle. The agent picks up new tickets as they arrive.1.00
17. Backup power supply failed in the serv...0.97
18. SHIFT1.00
19. 22:00 - 06:000.98
20. INC-2298 today0.96
21. QUEUE1.00
22. 5 open tickets1.00
23. Website security certificate expiring so...0.98
24. AGENT1.00
25. On-Call Triage Agent0.99
26. INC-2291 today0.97
27. Billing database broken, payments faili..0.99
28. INC-2292 today0.98
29. Orders stuck, main server frozen1.00
30. INC-2294 today0.98
31. Site failing for 1 in 3 visitors0.98
32. REQUEST1.00
33. AUTHORIZATION1.00
34. STATUS1.00
35. TIME1.00
36. Waiting for requests...0.96
37. World's Fair0.94
38. TRACK 5· JUNE 30, 20260.96
39. Security1.00
2:25 #9 skipped

shot 9·duplicate of #8
2:59 #10 skipped

shot 10·duplicate of #8
3:25 #11 skipped

shot 11·duplicate of #8
3:39 #12 skipped

shot 12·duplicate of #8
4:13 #13 skipped

shot 13·duplicate of #8
4:43 #14 skipped

shot 14·duplicate of #8
5:06 #15 done9 line(s)

shot 15·sharpness 784.9
1. Agents with API keys are overprivileged1.00
2. AlEngineer0.97
3. World's Fair0.96
4. change_role1.00
5. Keycard0.94
6. World's Fair0.95
7. X KmMoide/kimmaido0.84
8. TRACK 5· JUNE 30, 20260.95
9. Security1.00
5:37 #16 done14 line(s)

shot 16·sharpness 966.9
1. Human supervision isn't enough1.00
2. AlEngineer0.98
3. World's Fair0.96
4. Allow once1.00
5. PRESENTED BY0.99
6. Allow for this sessiun0.99
7. Microsoft1.00
8. Always allow1.00
9. --dangerously-skip-permissions1.00
10. Keycard0.96
11. World'sFair1.00
12. X KmMoidekimmaida0.84
13. TRACK 5 · JUNE 30, 20260.94
14. Security1.00
5:48 #17 skipped

shot 17·duplicate of #16
6:13 #18 done11 line(s)

shot 18·sharpness 1093.9
1. The agentic execution path1.00
2. AlEngineer0.99
3. World'sFair1.00
4. PRESENTED BY1.00
5. Microsoft1.00
6. oo0.92
7. Keycard0.95
8. World'sFair1.00
9. XKGmMoidekimmaido0.81
10. TRACK 5· JUNE 30, 20260.97
11. Security1.00
6:51 #19 done8 line(s)

shot 19·sharpness 1061.8
1. The agentic execution path0.98
2. AlEngineer0.99
3. World'sFair1.00
4. Keycard1.00
5. World's Fair0.97
6. XKmMoide/kimmaida0.84
7. TRACK 5 · JUNE 30, 20260.94
8. Security1.00
7:12 #20 skipped

shot 20·duplicate of #19
7:32 #21 skipped

shot 21·duplicate of #18
8:10 #22 done8 line(s)

shot 22·sharpness 818.1
1. Agent access problems1.00
2. AlEngineer0.98
3. World'sFair1.00
4. Keycard0.95
5. World's Fair0.96
6. XKmMoidekimmaida0.84
7. TRACK 5· JUNE 30, 20260.96
8. Security1.00
8:36 #23 done11 line(s)

shot 23·sharpness 1074.7
1. User authenticates with authorization server1.00
2. AlEngineer0.98
3. World'sFair1.00
4. oo0.96
5. oo0.87
6. oo0.96
7. Keycard1.00
8. World'sFair1.00
9. XKGmMoidekimmaide0.84
10. TRACK 5· JUNE 30, 20260.96
11. Security1.00

Transcript

205 cues· 3,345 words· 17,681 chars

chunk 0 · 0:13–0:50 · 124 words · 670 chars
0:13 So when I was a kid growing up in the 90s, we'd be out late all summer riding our bikes off in the neighborhood, playing with friends.
0:20 And from the 60s to the 80s, there were these public service announcements on TV where celebrities would come on and they would say, it's 10 p.m. Do you know where your children are?
0:30 Because apparently our parents at that time needed to be reminded that they had offspring they were responsible for.
0:36 And I feel like in this phase of AI where we are right now, we're entrusting agents with more and more responsibility, but we still kind of need that public service announcement that says it's 10 p.m. Do you know where your agents are?
chunk 1 · 0:54–1:23 · 83 words · 410 chars
0:54 So say you as a user want an agent to use an MCP server or API to accomplish tasks.
1:02 Now we know agents without access aren't useful.
1:05 So we give them an end file and we give them some API keys and we let them run off and go do their thing.
1:12 And this is fine until it's not.
1:14 So you've heard the horror stories, right?
1:16 Or maybe even experienced some of them yourself.
1:19 So let's see what this looks like in practice.
chunk 2 · 1:46–2:30 · 107 words · 563 chars
1:46 So I have an agent running here that is basically an incident management agent.
1:51 It is late night and there's a human user, but they're probably half asleep.
1:57 The agent is responsible for triaging issues that are coming in.
2:02 So if we look at the first one here.
2:10 So the first one is the backup power supply failed in the server room.
2:16 Now the agent is going to use an API key to read the system that is bringing in the reports.
chunk 3 · 2:22–3:03 · 101 words · 544 chars
2:22 It's going to evaluate what is written there, and it's going to decide that it can't do anything about this because it's a physical failure.
2:31 So it's going to escalate that ticket for the morning, and there's really sort of no problems yet.
2:38 So the second one is the certificate is expiring soon for TLS.
2:46 Agent is going to use that same API key again to read the ticket contents.
chunk 4 · 2:52–3:36 · 103 words · 559 chars
2:52 And then it is going to decide that it should renew the certificate.
2:56 And it's going to use another API key to call the cloud hosting service to renew the certificate.
3:03 And then it's going to use the API key from before to make the report for the morning team.
3:14 So here's where it kind of gets fun, right?
3:16 So this one is the billing database is broken and payments are failing.
chunk 5 · 3:23–4:07 · 121 words · 653 chars
3:23 Agent's going to read that.
3:27 it sees that the solution is pretty clear.
3:29 It says that the documented recovery is to delete the database and then restore from backup happen automatically.
3:36 So it has the Postgres connection string.
3:39 So it goes ahead and it drops the database and then it doesn't have a way to check to see if it was backed up.
3:44 So it just escalates that for the morning.
3:48 And this has really happened, right?
3:50 Like it's happened to high profile companies.
chunk 6 · 3:57–4:39 · 134 words · 666 chars
3:57 Now if we go to the next one, this one is the main server processes are frozen and they're not recovering.
4:04 Doing something about it is going to take prod offline for a brief amount of time.
4:08 And you can see the agent decided that it should do that.
4:11 And it's using the same API key now that it did to renew the certificate, right?
4:16 Because that API key is a kitchen sink.
4:18 It can do all of these things with it.
4:23 And then finally, we have sites failing for one in three users.
chunk 7 · 4:28–5:05 · 84 words · 428 chars
4:28 The recommended solution is to scale up.
4:31 And this is going to incur some amount of spend.
4:33 And then it goes ahead and it does that because, again, it can use that same API key to do this as well.
4:52 Right, so agents with API keys are indeed out past 10.
4:56 They're overprivileged, so this means they are able to act freely on decisions that they make that you may or may not agree with.
chunk 8 · 5:03–5:41 · 129 words · 692 chars
5:03 And they can do this even with your supervision.
5:05 So you might be familiar with the panic of trying to stop an agent mid-task because you told it to maybe read a project and it read the project and it found something it thought it should fix and then it starts writing.
5:17 Agents do that even while they're supervised and this is becoming even more of a problem because we have more and more agents that are running unsupervised and that only makes it worse because agents want to be helpful.
5:28 They're going to use all the permissions that they have access to in order to get the job done.

Chapters

0:00 It's 10pm, do you know where your agents are?
1:48 Demo: an incident agent on the night shift
3:18 When the agent drops the production database
4:52 Why agents are dangerously overprivileged
5:56 The agentic execution path
7:27 The fix: OAuth token exchange
8:32 Delegation: narrowing the user's access
9:23 Minting a fresh token per tool call
11:44 The demo again, now with token exchange
13:33 Policy blocks the database drop before it exists
14:27 Human approval backed by real policy
15:52 Works across CLIs, MCP servers, and any provider
17:34 Q&A