What Does Done Even Mean? Agents and Paperclip's Liveness Model - Dotta, Paperclip

AI Engineer · 7 min · 82 sentences · from YouTube's caption track

Each timecode opens YouTube at the start of that sentence. Line anchors (#s42) are the cue ids in the WebVTT, and every line carries its start and end seconds. All transcripts has every talk, and the whole corpus as one file.

1. 00:00An agent opens a pull request.
2. 00:01It passes the tests.
3. 00:03It updates the documentation.
4. 00:04It closes the issue and comments, "Looks done to me."
5. 00:07But is it actually done?
6. 00:09Is it done enough to merge?
7. 00:10Is it done enough to deploy?
8. 00:12Is it done enough to announce to your customers?
9. 00:15These are fundamentally different operational claims, and most agent systems just flatten it to a single green check mark.
10. 00:21I'm Dota.
11. 00:22I'm the creator of Paperclip, and I'm going to give you some hard-earned lessons that we've learned in creating Paperclip's liveness model.
12. 00:28What does done even mean?
13. 00:30Here's the thing.
14. 00:31Programming is solved, and agents can now produce more code and documentation faster than any human can ever verify.
15. 00:38And this actually gives us a new failure mode, is that agents can actually create more work than humans have time to verify.
16. 00:45So we need a way to verify that our agents are done more than just letting them check a checkbox.
17. 00:50Done doesn't mean that an agent just changed the status of a task being done.
18. 00:54Saying that something is done is actually a bundle of claims.
19. 00:57You're saying that an artifact was produced, that you have evidence that the task is actually complete, and you have a rubric in which you can verify against.
20. 01:07You know exactly who the owner is for the next step, and you know exactly what the next step is.
21. 01:14There's different levels to how done something is.
22. 01:16The producer might claim something is complete, but you need to have a reviewer, another party that looks at it and finds no obvious issues.
23. 01:23You want to verify and make sure that the evidence actually meets a specified standard.
24. 01:28You want to make sure that a person who is authorized to approve it actually approves that the work is done.
25. 01:34And you want to make sure that there's somebody who actually stands behind the decision.
26. 01:37And ideally, what you want is that the outcome has actually survived real-world conditions.
27. 01:44Because exhaustive human verification fails at high volume.
28. 01:49You might be able to verify a few tasks per day, but essentially, if you have humans verifying all the tasks and they have to sign off on it, you eventually what you just get is a form of verification theater.
29. 02:03What you need is a protocol for defining how tasks actually progress through a system.
30. 02:09You want to make sure that tasks are always kept moving, but they don't get stuck into invalid states.
31. 02:16You need a control plane that actually has the execution of the tasks being tied to specific contracts and constraints about what the system will do and what agents it will hand off your next task to.
32. 02:30Because really what you're trying to play against is this idea around keeping work moving, but also having it verified.
33. 02:37When a task has been reviewed by a human, you get the assurance that it's correct.
34. 02:41But having a human verify it means that the task is dead in its tracks.
35. 02:46You also want to keep liveliness.
36. 02:48Liveliness means that the work is continuing with no blockers.
37. 02:52And you're always trying to keep these two things in balance.
38. 02:54If you have tasks that are completely alive with no approvals, then what you get is this classic AI slop because you're producing a lot of things with kind of no quality control
39. 03:04and it's worse than creating nothing after a long period of time.
40. 03:08But if you have peer review, um then you have this enormous review queue where humans can't actually review it by hand anyway.
41. 03:17These agents will be creating far more than you can ever actually review, and so we have to find a way to tease apart the bundle of claims that are involved in saying a task is done.
42. 03:27With Paperclip, we have a number of mechanisms to keep this going.
43. 03:31You might think that you can easily just write a for loop over your task manager and have your agents work, but quickly you'll find that falls apart.
44. 03:39As soon as you start integrating task dependency trees, blockers, multiple agents, item potent checkouts, like locks on checkouts, you find that this tension between liveliness and verification actually gets quite complicated.
45. 03:55There's really three invariants that are extremely important when you're thinking about what you want out of a control plane for your agentic work.
46. 04:03You want to ensure that productive work continues.
47. 04:05You want to make sure that only real blockers stop work.
48. 04:09And you want to make sure that infinite loops are bounded.
49. 04:11In Paperclip, we have built a number of mechanisms to deal with this problem.
50. 04:16So, for example, every time you have a task, there's clear transitions to what the next state could be.
51. 04:23We have first-class blockers between tasks, and the control plane enforces those blockers.
52. 04:28We have moments of interactive human approval, where human choices leave an audit trail.
53. 04:33You can set reviewers and approvers on tasks explicitly, meaning when this task completes, another agent can review it.
54. 04:41We also have the idea of watchdogs, which is this maximizer mode, which says, um, try as hard as you can to make sure that this happens.
55. 04:50When you have a watchdog, it's another agent, um, who is given a goal, and it enforces that all of your agents continue to work until that goal has been achieved.
56. 04:58The important thing here is that the watchdog within Paperclip is harness agnostic.
57. 05:04You can use it with Pi, OpenGL, Hermes, Claude Code, Codex, whatever you're using, you have one consistent interface for ensuring that goal is complete.
58. 05:15So, one of the best pieces of advice we have is that you stop treating done as a Boolean and treat it more like an object.
59. 05:21This isn't specific to Paperclip.
60. 05:23It's just advice on how you think about what is done.
61. 05:27Humans automatically paper over these details, but when we're building agentic systems, it's important that your agents can distinguish between the different pieces of what they're claiming when they say something is done.
62. 05:36The artifact that they're saying is complete, the scope, the rubric or the standard, the evidence that it's done, who verified the work, who has the authority to sign off on the work,
63. 05:47And what risk might be left?
64. 05:48And really, what's the next action going to be?
65. 05:51So, you want to make sure that when you define done, it's not just a checkbox.
66. 05:56So, if you want to get 100 times more work done, you should steal this checklist.
67. 06:00You need to define exactly what does done mean for this task.
68. 06:05You definitely want to separate the verifier from the author.
69. 06:08Often, this means you're using a different model.
70. 06:10So, if you're coding using Claude, have Codex verify.
71. 06:14You want to ask your agents to provide evidence.
72. 06:16Don't just ask them to say, "Is this done?"
73. 06:19But, give them the tools they need to verify that the work is done.
74. 06:22Write the code to have the custom browser harness.
75. 06:25Write the code to take the screenshots.
76. 06:27Make sure they have access to a browser.
77. 06:29Make sure that they have custom agent hooks or custom agent tooling to actually run through and click the buttons and try it out themselves and verify that the work is truly done.
78. 06:39Make sure you have a clear chain of custody, that every agent knows that as soon as they're done, who they're supposed to give the work to next.
79. 06:48It can be easy to just fire off a single-line instruction and vibe with whatever comes back.
80. 06:54But, if you have serious work that you're accountable for, it's very important that you define what done really means in as much detail as possible, and that you have a structure
81. 07:04for your agents, so that way they can verify that all of the claims that are involved with something being done are actually met.
82. 07:11Thank you.