vidtheque.

read-only demo

Videos 8qWIPUia2O8

Every Harness Will Become A Claw — Sam Bhagwat, Mastra

index_state ready data_status ok

AI Engineer· published 2026-07-21· 0:15:35· en-US· indexed 2026-08-10 19:43

Open on YouTube

Scene timeline

  1. Shot 0, 0:00 to 0:03, 1 of 1 keyframes kept
  2. Shot 1, 0:03 to 0:05, 1 of 1 keyframes kept
  3. Shot 2, 0:05 to 0:12, 1 of 1 keyframes kept
  4. Shot 3, 0:12 to 0:31, 1 of 1 keyframes kept
  5. Shot 4, 0:31 to 0:34, 1 of 1 keyframes kept
  6. Shot 5, 0:34 to 0:51, 1 of 1 keyframes kept
  7. Shot 6, 0:51 to 1:03, 1 of 1 keyframes kept
  8. Shot 7, 1:03 to 1:29, 1 of 1 keyframes kept
  9. Shot 8, 1:29 to 1:45, 1 of 1 keyframes kept
  10. Shot 9, 1:45 to 2:34, 1 of 1 keyframes kept
  11. Shot 10, 2:34 to 3:02, 1 of 1 keyframes kept
  12. Shot 11, 3:02 to 3:31, 1 of 1 keyframes kept
  13. Shot 12, 3:31 to 3:59, 0 of 1 keyframes kept
  14. Shot 13, 3:59 to 4:28, 0 of 1 keyframes kept
  15. Shot 14, 4:28 to 4:57, 0 of 1 keyframes kept
  16. Shot 15, 4:57 to 4:59, 1 of 1 keyframes kept
  17. Shot 16, 4:59 to 5:27, 0 of 1 keyframes kept
  18. Shot 17, 5:27 to 5:55, 0 of 1 keyframes kept
  19. Shot 18, 5:55 to 6:24, 0 of 1 keyframes kept
  20. Shot 19, 6:24 to 6:52, 0 of 1 keyframes kept
  21. Shot 20, 6:52 to 7:09, 0 of 1 keyframes kept
  22. Shot 21, 7:09 to 7:35, 1 of 1 keyframes kept
  23. Shot 22, 7:35 to 8:02, 0 of 1 keyframes kept
  24. Shot 23, 8:02 to 8:28, 0 of 1 keyframes kept
  25. Shot 24, 8:28 to 8:54, 1 of 1 keyframes kept
  26. Shot 25, 8:54 to 9:20, 0 of 1 keyframes kept
  27. Shot 26, 9:20 to 9:38, 0 of 1 keyframes kept
  28. Shot 27, 9:38 to 9:39, 0 of 1 keyframes kept
  29. Shot 28, 9:39 to 10:05, 1 of 1 keyframes kept
  30. Shot 29, 10:05 to 10:44, 1 of 1 keyframes kept
  31. Shot 30, 10:44 to 11:11, 1 of 1 keyframes kept
  32. Shot 31, 11:11 to 11:40, 1 of 1 keyframes kept
  33. Shot 32, 11:40 to 12:09, 0 of 1 keyframes kept
  34. Shot 33, 12:09 to 12:38, 0 of 1 keyframes kept
  35. Shot 34, 12:38 to 13:07, 0 of 1 keyframes kept
  36. Shot 35, 13:07 to 13:36, 0 of 1 keyframes kept
  37. Shot 36, 13:36 to 14:05, 0 of 1 keyframes kept
  38. Shot 37, 14:05 to 14:07, 0 of 1 keyframes kept
  39. Shot 38, 14:07 to 14:33, 0 of 1 keyframes kept
  40. Shot 39, 14:33 to 14:59, 0 of 1 keyframes kept
  41. Shot 40, 14:59 to 15:01, 0 of 1 keyframes kept
  42. Shot 41, 15:01 to 15:19, 1 of 1 keyframes kept
  43. Shot 42, 15:19 to 15:35, 0 of 1 keyframes kept

43 shot(s).

keyframes kept every frame deduplicated

What was stored

cues
195
whisperx 195
chunks
27
from 195 cues
keyframes
20
kept of 43 captured
frames with text
20
328 lines read
chapters
0
from the source metadata
keyframe bytes
5.0 MB
word timings on 195 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage state model started took
fetch done — 2026-08-09 23:48 0s
stt done — 2026-08-09 20:39 17s
chunk done — 2026-08-09 20:40 0s
text_embed done — 2026-08-10 19:43 0s
keyframe done — 2026-08-09 20:40 2m 04s
ocr done — 2026-08-09 20:42 7s
frame_embed done — 2026-08-10 19:43 4s

Frames, and what the machine read

  • 0:02 #0 done2 line(s)

    shot 0·sharpness 456.7

    1. AlEngineer0.96
    2. World's Fair0.97

  • 0:03 #1 done2 line(s)

    shot 1·sharpness 663.3

    1. AlEngineer0.95
    2. World's Fair0.99

  • 0:10 #2 done24 line(s)

    shot 2·sharpness 2742.3

    1. LAB & PLATINUM SPONSORS0.99
    2. Amazon AGI Lab0.98
    3. ANTHROP\C1.00
    4. Google DeepMind1.00
    5. MINIMAX0.93
    6. OpenAI0.92
    7. Akamai1.00
    8. arize0.92
    9. aws1.00
    10. Braintrust bright data0.98
    11. B1.00
    12. Browserbase1.00
    13. docker1.00
    14. :neo4j0.90
    15. ORACLE1.00
    16. PayPal1.00
    17. qodo1.00
    18. reducto1.00
    19. Sonar1.00
    20. Makers of1.00
    21. togetherai1.00
    22. Unblocked1.00
    23. WorkOS1.00
    24. SonarQube1.00

  • 0:18 #3 done3 line(s)

    shot 3·sharpness 158.9

    1. mistra0.80
    2. AlEngineer1.00
    3. World's Fair0.98

  • 0:32 #4 done5 line(s)

    shot 4·sharpness 653.7

    1. EVERY HARNESS0.99
    2. WILL BECOME A CLAW0.99
    3. A talk by Sam Bhagwat0.98
    4. @calcsam1.00
    5. June 30, 20261.00

  • 0:39 #5 done16 line(s)

    shot 5·sharpness 1786.3

    1. About me0.99
    2. Principles1.00
    3. Principles1.00
    4. of Building0.99
    5. Sam Bhagwat (@calcsam)1.00
    6. of Building Al Agents1.00
    7. Al Agents0.96
    8. Co-founder, CEO Mastra1.00
    9. 240K+ copies distributed0.99
    10. Prev: co-founder Gatsby0.98
    11. (also author)1.00
    12. About Mastra1.00
    13. Leading TypeScript agent framework1.00
    14. $35M raised from YC, Spark, Gradient, etc0.99
    15. Running in prod across F500, scaled startups, and Al-native cos1.00
    16. @calcsam1.00

  • 0:52 #6 done46 line(s)

    shot 6·sharpness 3261.6

    1. Mastra is running in prod across F500,1.00
    2. scaled startups, and Al-native cos1.00
    3. AI SDR + SALES PRODUCTIVITY1.00
    4. PROSUMER B2C VERTICAL AGENTS0.99
    5. Range1.00
    6. Counsel1.00
    7. indeed1.00
    8. CarGurus0.98
    9. PFF0.92
    10. WorkOS1.00
    11. ▲ΛRTISAN0.85
    12. INVESTING1.00
    13. MEDICAL1.00
    14. CAREER1.00
    15. E-COMMERCE1.00
    16. SPORTS ANALYTICS1.00
    17. ENTERPRISE SEARCH1.00
    18. AGENT-IFYING BUSINESS PROCESSES0.99
    19. AD CREATION1.00
    20. SoftBank1.00
    21. MarshMcLennan1.00
    22. iress1.00
    23. PERCEPTA1.00
    24. Brex1.00
    25. PayPal1.00
    26. Adobe1.00
    27. INVESTING1.00
    28. MEDICAL1.00
    29. FIN. SERVICES0.97
    30. FRAUD DETECTION1.00
    31. B2B IN-APP CHATBOT0.99
    32. AI SRE + DEV PRODUCTIVITY0.99
    33. AGENT VIBE CODING1.00
    34. factorial1.00
    35. ITERABLE1.00
    36. 丝Sanity0.91
    37. workdày0.93
    38. PayPal1.00
    39. salesforce1.00
    40. replit1.00
    41. Contentstack0.98
    42. Abnormal0.97
    43. Tena0.95
    44. MongoDB1.00
    45. cvent1.00
    46. RIVERSIDE1.00

  • 1:16 #7 done3 line(s)

    shot 7·sharpness 154.7

    1. mesta0.71
    2. AlEngineer0.99
    3. World's Fair1.00

  • 1:34 #8 done33 line(s)

    shot 8·sharpness 1867.8

    1. AlEngineer0.99
    2. World's Fair0.96
    3. Welcome to the harness era1.00
    4. CLOUD1.00
    5. SLACK1.00
    6. LOCAL1.00
    7. OSS0.57
    8. FRAMEWORKS1.00
    9. PRESENTED BY0.97
    10. PRODUCTS1.00
    11. Devin1.00
    12. Cursor1.00
    13. LangChain DeepAgents0.98
    14. Microsoft1.00
    15. Factory Droid0.97
    16. Claude Code1.00
    17. Cloudflare Flue1.00
    18. Claude Tag0.99
    19. Codex1.00
    20. Vercel Eve1.00
    21. INTERNAL TOOLS1.00
    22. Ramp Inspect1.00
    23. 口0.92
    24. OpenCode1.00
    25. Mastra1.00
    26. Stripe Minions1.00
    27. F10.68
    28. Pi1.00
    29. Both harnesses and kits to build your own.0.99
    30. @calcsam0.97
    31. 03 / 120.95
    32. Engineering the future of Al0.99
    33. World's Fair0.97

  • 2:28 #9 done22 line(s)

    shot 9·sharpness 1865.0

    1. AlEngineer0.96
    2. World's Fair0.99
    3. We're ascending on the agentic spectrum0.99
    4. Initiative &1.00
    5. Learning1.00
    6. PRESENTEDBY1.00
    7. + Always-On1.00
    8. + Durable &0.99
    9. Microsoft1.00
    10. Dogged1.00
    11. Actions1.00
    12. Local -> Cloud0.99
    13. LLM1.00
    14. Agent1.00
    15. Harness1.00
    16. Claw1.00
    17. more autonomy1.00
    18. @calcsam0.99
    19. 04 / 120.95
    20. TRACK 1· JUNE 30, 20260.95
    21. World's Fair0.97
    22. Claws & Personal Agents0.99

  • 2:59 #10 done15 line(s)

    shot 10·sharpness 1619.4

    1. AlEngineer0.99
    2. World'sFair1.00
    3. Actions / LLM ⇒ Agent0.95
    4. PRESENTED BY0.94
    5. Microsoft1.00
    6. Loop1.00
    7. Tools1.00
    8. Memory1.00
    9. Retry failed tasks0.99
    10. Context engineering1.00
    11. State1.00
    12. @calcsam1.00
    13. TRACK 1· JUNE 30, 20260.97
    14. Vorld'sFair1.00
    15. Claws & Personal Agents0.99

  • 3:17 #11 done21 line(s)

    shot 11·sharpness 2283.7

    1. AlEngineer0.96
    2. World's Fair0.99
    3. Durable & Dogged / Agent ⇒ Harness0.97
    4. PRESENTED BY0.95
    5. Microsoft1.00
    6. Planning mode1.00
    7. Parallel subagents1.00
    8. TUI + slash cmds0.98
    9. Skills1.00
    10. Dynamic subagents1.00
    11. Background bash1.00
    12. Auto-compaction1.00
    13. Thread persistence1.00
    14. Queue / steer /0.97
    15. Session-long1.00
    16. interrupt0.99
    17. tool approval1.00
    18. @calcsam1.00
    19. TRACK 1· JUNE 30, 20260.96
    20. World's Fair0.98
    21. Claws & Personal Agents0.99

  • 3:34 #12 skipped

    shot 12·duplicate of #11

  • 4:03 #13 skipped

    shot 13·duplicate of #11

  • 4:45 #14 skipped

    shot 14·duplicate of #11

  • 4:58 #15 done16 line(s)

    shot 15·sharpness 1798.8

    1. AlEngineer0.99
    2. World'sFair1.00
    3. Always-on / Local ⇒ Cloud Harness0.96
    4. PRESENTED BY0.95
    5. Microsoft1.00
    6. Multiplayer1.00
    7. Slack + web UI1.00
    8. Mobile + tunneling0.98
    9. Cloud sandboxes1.00
    10. More parallelism1.00
    11. PR-native workflows1.00
    12. @calcsam0.97
    13. 08 / 120.85
    14. TRACK 1· JUNE 30, 20260.97
    15. World's Fair0.96
    16. Claws & Personal Agents0.99

  • 5:18 #16 skipped

    shot 16·duplicate of #9

  • 5:36 #17 skipped

    shot 17·duplicate of #15

  • 6:12 #18 skipped

    shot 18·duplicate of #15

  • 6:30 #19 skipped

    shot 19·duplicate of #15

  • 7:04 #20 skipped

    shot 20·duplicate of #9

  • 7:17 #21 done20 line(s)

    shot 21·sharpness 2004.3

    1. AlEngineer0.99
    2. World's Fair0.98
    3. Initiative + Learning/1.00
    4. Harness ⇒ Claw0.98
    5. PRESENTED BY0.99
    6. Microsoft1.00
    7. Heartbeat1.00
    8. Channels1.00
    9. Always-on daemon1.00
    10. Subscriptions1.00
    11. Persistent memory1.00
    12. Continual learning0.99
    13. real1.00
    14. claws:1.00
    15. OpenClaw1.00
    16. Hermes Agent1.00
    17. @calcsam1.00
    18. TRACK 1· JUNE 30, 20260.97
    19. World's Fair0.99
    20. Claws & Personal Agents0.99

  • 7:38 #22 skipped

    shot 22·duplicate of #21

  • 8:17 #23 skipped

    shot 23·duplicate of #21

Transcript

195 cues· 2,477 words· 13,304 chars

  1. 0:12 I'll get us started.
  2. 0:14 So long day of talks and how are y'all feeling?
  3. 0:20 Cool.
  4. 0:21 Yeah, good to see we still have some energy.
  5. 0:24 I know there's a lot of evening events.
  6. 0:27 We've heard a lot about the present, and I'm gonna talk about the future.
  7. 0:31 So my talk is called Every Harness Will Become a Claw.
  8. 0:35 Here's a little bit about me.
  9. 0:36 I am the co-founder, CEO of Mastra.
  10. 0:40 We are a TypeScript agent framework.
  11. 0:42 I am also the author of a book that you may have gotten a copy of either outside or at a previous event.
  12. 0:48 We have seen a lot of agents running in production over the last 18 months.
  13. 0:54 And I'm saying that as kind of context and stage setting for the thoughts and ideas that I'm about to share right now.
  14. 1:05 And the thing that I'm going to say is welcome to the harness era.
  15. 1:10 What do I mean by the harness era?
  16. 1:12 Well, let's just talk about the types of harnesses that we see right now.
  17. 1:16 We see local harnesses.
  18. 1:18 We use them every day, daily driving our coding, right?
  19. 1:22 We see cloud harnesses.
  20. 1:24 These are both products that we can purchase as well as if we work in some of these companies that have built their own internal coding agents that live on Slack.
  21. 1:33 And then, of course, we have your friendly local open source frameworks that have some of these primitives and give you the tools that you need to build your own.
  22. 1:43 And that's where we fit in.
  23. 1:46 Now let's talk about where we are sort of collectively as an industry and how things have evolved over the last, we'll say, year or two.
  24. 1:54 to 18 months.
  25. 1:56 Last year at AI Engineer, we were talking a lot about agents.
  26. 1:59 We were talking about the agent loop.
  27. 2:00 We were talking about agents versus workflows.
  28. 2:06 As we're thinking about the agentic spectrum, I often compare it to self-driving as a spectrum.
  29. 2:12 There are different levels of self-driving autonomy, whether that's lane assist, whether that's Tesla FSD, whether that's I'm sitting in the back of my Waymo and there's nobody behind the steering wheel.
  30. 2:25 There are various aspects to the agentic spectrum between LLMs, agents, harnesses, and claws.
  31. 2:30 And I'm gonna talk about what we've seen and where we're going.
  32. 2:35 What makes an agent different than an LLM?
  33. 2:37 Hopefully we mostly know this, but just as a quick refresher, right?
  34. 2:40 It's the agent loop, it's tool calls, it's memory, it's the ability to retry failed tasks, it's context engineering.
  35. 2:48 Dex is a close friend and an inspiration for this, one of the inspirations for this talk.
  36. 2:52 And it's agent state, right?
  37. 2:54 These are some things that, like, hey, I'm running an agent in a loop, and I can't just do this with a one-shot call to an LLM.
  38. 3:02 I've tried to make these qualities, I'm not sure what the quality of taking actions is, active or something, so I just put action.
  39. 3:10 But qualities here are starting to emerge when we move from an agent to a harness.
  40. 3:17 Durability and doggedness.
  41. 3:19 A friend of mine was referring to an agent that he was using, and he called it dogged, which I really like.
  42. 3:23 And I'm taking that for this talk.
  43. 3:25 So durability, just the sheer quality of being able to run not for minutes, but for hours or days.
  44. 3:34 What encompasses this?
  45. 3:35 Well, sometimes it's like, hey, I lost a connection in the middle of the turn, but I persisted the stream, and so now I can resume from the place where I started.
  46. 3:45 There's planning mode.
  47. 3:46 We all see this in Cloud Code.
  48. 3:49 Parallel subagents being able to fan out multiple tasks at the same time.
  49. 3:53 We have more affordances with a 2E and slash commands.
  50. 3:57 We have skills.

Open at this second