Context Is the New Code — Patrick Debois, Tessl

What was stored

cues

276

whisperx 276

chunks

48

from 276 cues

keyframes

56

kept of 70 captured

frames with text

56

1,451 lines read

chapters

9

from the source metadata

keyframe bytes

7.3 MB

word timings on 276 cues

Provenance

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
• 

  0:34 #3 done8 line(s)

  shot 3·sharpness 511.1

  1. AlEngineer0.98
  2. Context is the0.97
  3. EUROPE1.00
  4. Context Developm1.00
  5. Al Engineer London 2020.99
  6. AlEngineer0.98
  7. EUROPE1.00
  8. CHUBE1.00
• 

  0:43 #4 done7 line(s)

  shot 4·sharpness 512.7

  1. AlEngineer0.99
  2. Context is the0.97
  3. EUROPE1.00
  4. Context Developm1.00
  5. Al Engineer London 2020.99
  6. AlEngineer0.99
  7. EUROPE1.00
• 

  1:10 #5 done12 line(s)

  shot 5·sharpness 1615.2

  1. Context is the new Code0.98
  2. AIE1.00
  3. ★1.00
  4. ★0.99
  5. Context Development Lifecycle1.00
  6. 40.97
  7. Al Engineer London 2026 - Patrick Debois0.99
  8. AlEngineer1.00
  9. EUROPE1.00
  10. Google DeepMind0.99
  11. AlEngineer1.00
  12. 20260.95
• 

  1:37 #6 done50 line(s)

  shot 6·sharpness 1911.2

  1. Context is the new Code1.00
  2. Prompt → Code0.99
  3. Code → Skill0.92
  4. • ••0.68
  5. Prompt0.99
  6. Code1.00
  7. Write a function to fetch user1.00
  8. def fetch_user(uid, retries=3):1.00
  9. data from an API endpoint.0.99
  10. for i in range(retries):1.00
  11. Add error handling and retry0.99
  12. try:1.00
  13. AIE1.00
  14. ★1.00
  15. logic with a timeout.1.00
  16. r.raise_for_status()1.00
  17. f"/users/{uid}", timeout=5)0.99
  18. r = requests.get(0.99
  19. ★1.00
  20. generates1.00
  21. return r.json()0.99
  22. ★1.00
  23. ★1.00
  24. ★1.00
  25. packaged as1.00
  26. Code1.00
  27. † Skill0.89
  28. def fetch_user(uid, retries=3):1.00
  29. for i in range(retries):1.00
  30. ## API Fetch Best Practices0.99
  31. try:1.00
  32. · Always set a request timeout0.98
  33. r = requests.get(0.99
  34. - Retry on transient errors0.97
  35. f"/users/{uid}", timeout=5)0.99
  36. · Re-raise on last attempt0.96
  37. r.raise_for_status()1.00
  38. return r.json()1.00
  39. ## Evals0.99
  40. except Exception as e:0.99
  41. ✓ timeout param is set0.98
  42. if i == retries - 1: raise0.98
  43. ✓ retries ≥ 30.95
  44. Prompt → Code (left) · Code → Skill (right)0.99
  45. AlEngineer0.99
  46. EUROPE1.00
  47. Braintrust1.00
  48. WorkOS OpenAI0.94
  49. AlEngineer0.99
  50. 20260.99
• 

  1:47 #7 skipped

  shot 7·duplicate of #6

• 

  2:17 #8 done50 line(s)

  shot 8·sharpness 1830.0

  1. Context is the new Code1.00
  2. Prompt → Code0.98
  3. Code → Skill0.91
  4. • • •0.79
  5. Prompt0.99
  6. Code1.00
  7. Write a function to fetch user1.00
  8. def fetch_user(uid, retries=3):1.00
  9. data from an API endpoint.0.99
  10. for i in range(retries):1.00
  11. Add error handling and retry0.97
  12. try:1.00
  13. AIE1.00
  14. ★1.00
  15. logic with a timeout.1.00
  16. r.raise_for_status()1.00
  17. f"/users/{uid}", timeout=5)0.98
  18. r = requests.get(1.00
  19. ★1.00
  20. ★1.00
  21. ★1.00
  22. ★1.00
  23. generates1.00
  24. return r.json()0.98
  25. packaged as0.99
  26. Code1.00
  27. † Skill0.80
  28. def fetch_user(uid, retries=3):1.00
  29. for i in range(retries):1.00
  30. ## API Fetch Best Practices1.00
  31. try:1.00
  32. r = requests.get(0.98
  33. · Always set a request timeout0.98
  34. · Retry on transient errors0.96
  35. f"/users/{uid}", timeout=5)0.99
  36. · Re-raise on last attempt0.97
  37. r.raise_for_status()1.00
  38. return r.json()1.00
  39. ## Evals0.99
  40. except Exception as e:1.00
  41. ✓ timeout param is set0.98
  42. if i == retries - 1: raise0.97
  43. ✓ retries ≥ 30.94
  44. Prompt → Code (left) · Code → Skill (right)0.99
  45. AlEngineer0.98
  46. EUROPE1.00
  47. AlEngineer0.96
  48. AlEngineer0.99
  49. 20261.00
  50. EUROPE1.00
• 

  2:41 #9 done13 line(s)

  shot 9·sharpness 1479.2

  1. to think in parallels0.99
  2. AIE1.00
  3. ★1.00
  4. 2009 - What if Ops was like Dev0.99
  5. 40.95
  6. 2025 - What if Context was like Code0.98
  7. AlEngineer0.99
  8. Patrick Debois - Product Devrel Tessl0.97
  9. EUROPE1.00
  10. AlEngineer0.97
  11. AlEngineer1.00
  12. 20260.97
  13. EUROPE1.00
• 

  3:34 #10 done19 line(s)

  shot 10·sharpness 2079.6

  1. Context Development Lifecycle1.00
  2. Generate1.00
  3. Distribute1.00
  4. Making implicit knowledge explicit1.00
  5. Context as a package0.99
  6. AIE1.00
  7. ★1.00
  8. ★1.00
  9. ★1.00
  10. Evaluate1.00
  11. Observe0.99
  12. TDD for context1.00
  13. Learn from use in the wild0.96
  14. https://tessl.io/blog/context-development-lifecycle-better-context-for-ai-coding-agents/1.00
  15. AlEngineer1.00
  16. EUROPE1.00
  17. Engineering the future of Al1.00
  18. AlEngineer0.99
  19. 20260.98
• 

  3:47 #11 done21 line(s)

  shot 11·sharpness 2571.9

  1. Page 7 of 441.00
  2. Today's Agenda1.00
  3. *★*0.54
  4. 011.00
  5. Generate — create & curate context0.96
  6. AIE1.00
  7. ★1.00
  8. ★1.00
  9. ★1.00
  10. ★★0.87
  11. 021.00
  12. Evaluate — test & measure context quality0.99
  13. 031.00
  14. Distribute — package & share context0.99
  15. 04 Observe — monitor & improve in production0.98
  16. AlEngineer0.98
  17. EUROPE1.00
  18. 041.00
  19. Engineering the future of Al1.00
  20. AlEngineer1.00
  21. 20261.00
• 

  3:50 #12 done10 line(s)

  shot 12·sharpness 1116.0

  1. Page 8 of 440.99
  2. AIE1.00
  3. Generate1.00
  4. ★1.00
  5. ★1.00
  6. AlEngineer0.99
  7. EUROPE1.00
  8. Engineering the future of Al1.00
  9. AlEngineer1.00
  10. 20260.98
• 

  4:03 #13 done31 line(s)

  shot 13·sharpness 2466.8

  1. Prompts -Humans as context engine0.99
  2. Claude Code v2.1.961.00
  3. Tips for getting1.00
  4. Welcome back Patrick!1.00
  5. started1.00
  6. Ask Claude to create...0.99
  7. *★★0.61
  8. AIE1.00
  9. ★1.00
  10. ★1.00
  11. Opus 4.6 (1M context)· Claude Team· Tessl AI0.97
  12. No recent activity1.00
  13. Recent activity1.00
  14. ★1.00
  15. Limited1.00
  16. ★1.00
  17. ★1.00
  18. ★1.00
  19. ~/dev/aie1.00
  20. Tell me when my talk is at AI engineer - https://www.ai.engineer/europe0.99
  21. Fetch(https://www.ai.engineer/europe)1.00
  22. Received 447KB (200 OK)0.98
  23. Your talk "Context Is the New Code" is on April 10 (Day 3) at 11:15am in the0.99
  24. Westminster track.1.00
  25. https://www.ai.engineer/europe1.00
  26. AlEngineer0.98
  27. EUROPE1.00
  28. Tell ae0.74
  29. Engineering the future of Al0.98
  30. AlEngineer1.00
  31. 20260.99
• 

  4:38 #14 done39 line(s)

  shot 14·sharpness 1662.8

  1. Rules, Instructions - Agent.md0.98
  2. AGENTS.md1.00
  3. # AGENTS.md0.94
  4. used by over 60k open-source projects.0.95
  5. A simple, open format for guiding coding agents,1.00
  6. - Install deps: pepe install0.77
  7. as Setup commands0.87
  8. - Start dev server: pope de0.89
  9. Think of AGENTS.md as a README for agents: a dedicated,1.00
  10. - Run tests: prpn test0.85
  11. predictable place to provide the context and instructions to help1.00
  12. Al coding agents work on your project.0.98
  13. ss Code style0.89
  14. - Typešcriat strict mde0.92
  15. AIE1.00
  16. ★1.00
  17. View on GitHub1.00
  18. - Single quotes, no senicolans0.95
  19. - Use functional patterns where possible0.97
  20. ★1.00
  21. ★1.00
  22. ★1.00
  23. ★1.00
  24. Why AGENTS.md?0.97
  25. 40.92
  26. build steps, tests, and conventions that might clutter a README or aren't relevant to human contributors.0.99
  27. stimes detailed context coding agents need:0.98
  28. We intentionally kept it separate to:0.98
  29. Give agents a clear,predictable place for instructions.0.92
  30. Keep READMEs concise and focused on human contributors.1.00
  31. Provide precise, agent-focused guidance that complements existing README and docs.0.98
  32. Rather than introducing another proprietary fle, we chose a name and format that could work for anyone.1.00
  33. Agent.md1.00
  34. AlEngineer0.98
  35. EUROPE1.00
  36. AlEngineer0.97
  37. AlEngineer1.00
  38. 20261.00
  39. EUROPE1.00
• 

  5:04 #15 done43 line(s)

  shot 15·sharpness 2154.5

  1. Context from code and docs0.99
  2. 图Context70.92
  3. Ref.1.00
  4. plan1.00
  5. docs1.00
  6. mcp1.00
  7. blog1.00
  8. dashboard1.00
  9. Up-to-date Docs1.00
  10. for LLMs and Al code editors1.00
  11. Plan - orchestrate coding agents across large projects. Iry it fres0.97
  12. Get the latest documentation and code into Cursor, Claude, or other LLMs0.99
  13. *★★0.56
  14. ★1.00
  15. Context for your agent.1.00
  16. AIE1.00
  17. ★0.90
  18. Search a library (e.g. Next, React)0.98
  19. ★1.00
  20. or1.00
  21. Stop hallucinations by searching public and private docs. No context0.99
  22. bloat.1.00
  23. ★1.00
  24. ★1.00
  25. ★1.00
  26. https://context7.com/0.99
  27. Chat with Docs1.00
  28. https://ref.tools/0.97
  29. Context Hub1.00
  30. 40.91
  31. Coding agents hallucinate APis and forget what they learn in a session. Context Hub gives0.99
  32. them curated, versioned docs, plus the ability to get smarter with every task. All content1.00
  33. is open and maintained as markdown in this repo — you can inspect exactly what your0.99
  34. agent reads, and contribute back.0.99
  35. license MIT npm v0.1.30.96
  36. node >=180.98
  37. Quick Start1.00
  38. AlEngineer0.98
  39. https://github.com/andrewyng/context-hub0.99
  40. EUROPE1.00
  41. Engineering the future of Al0.99
  42. AlEngineer1.00
  43. 20260.99
• 

  5:09 #16 done12 line(s)

  shot 16·sharpness 541.9

  1. Context from co0.99
  2. R1.00
  3. AlEngineer0.99
  4. EUROPE1.00
  5. Context Hub1.00
  6. is open and mptned as maldon n ths repo-0.58
  7. Coding agents halucinate As and forget what they0.89
  8. them curated, versoned docs pus the ablity to get0.82
  9. agent reads, and contribute back.0.99
  10. Quick Start1.00
  11. AlEngineer0.98
  12. EUROPE1.00
• 

  5:41 #17 done38 line(s)

  shot 17·sharpness 1968.3

  1. Context Connectors1.00
  2. MCP Registry0.97
  3. Unblocked1.00
  4. Connect Unblocked to your knowledge sources1.00
  5. *★★0.56
  6. ★1.00
  7. AIE1.00
  8. ★1.00
  9. ★1.00
  10. Connect models to the real world1.00
  11. ★1.00
  12. ★1.00
  13. ★1.00
  14. Servers and tools from the community that connect0.99
  15. API1.00
  16. models to files, APis, databases, and more.0.99
  17. Search MCPs1.00
  18. N0.94
  19. All MCP servers1.00
  20. 881.00
  21. Markitdown1.00
  22. Install1.00
  23. Netdata1.00
  24. Install0.96
  25. Convert various file formats (PDF, Word, Excel,0.99
  26. Real-time infrastructure monitoring with metrics,0.99
  27. images, audio) to Markdown.1.00
  28. logs, alerts, and ML-based anomaly detection.0.99
  29. By microsoft 90,8610.93
  30. By netdata 78,0950.96
  31. Install the Unblocked MCP plugin1.00
  32. Github MCP - https://github.com/mcp0.99
  33. https://getunblocked.com/unblocked-mcp/0.99
  34. AlEngineer0.99
  35. EUROPE1.00
  36. Engineering the future of Al1.00
  37. AlEngineer1.00
  38. 20260.99
• 

  5:51 #18 done33 line(s)

  shot 18·sharpness 2238.5

  1. Spec Driven Development0.99
  2. KIRO1.00
  3. README1.00
  4. Code of conduct1.00
  5. Contributing1.00
  6. MIT license1.00
  7. Security1.00
  8. Specs1.00
  9. AIE1.00
  10. What are specs?1.00
  11. ★1.00
  12. ★1.00
  13. ★1.00
  14. tracking and aocountablity.0.96
  15. With Kirols specs, you oan:0.94
  16. Break down requirements into user stories with acceptance criteria0.97
  17. Buld design dos with gequence diograrms and architectue plans0.93
  18. Track implementation progress acros discrote tasks0.96
  19. Spec Kit1.00
  20. Collaborate effectively botween product and engincoring teams0.98
  21. Core Structure1.00
  22. Build high-quality software faster.1.00
  23. An open source toolkit that allows you to focus on product scenarios and predictable1.00
  24. outcomes instead of vibe coding every piece from scratch.1.00
  25. Create Release failingStar78k0.89
  26. docs GitHub Pages0.98
  27. Kiro - http://kiro.dev/docs/specs/0.99
  28. Gitub Speckit - https://github.com/github/spec-kit1.00
  29. AlEngineer0.98
  30. EUROPE1.00
  31. Engineering the future of Al0.99
  32. AlEngineer0.95
  33. 20260.98
• 

  6:13 #19 done21 line(s)

  shot 19·sharpness 2553.2

  1. Today's Agenda1.00
  2. *★★0.52
  3. 011.00
  4. Generate — create & curate context0.98
  5. AIE1.00
  6. ★1.00
  7. ★1.00
  8. ★1.00
  9. ★1.00
  10. 021.00
  11. Evaluate — test & measure context quality1.00
  12. 031.00
  13. Distribute — package & share context0.99
  14. 04 Observe — monitor & improve in production0.98
  15. AlEngineer0.98
  16. EUROPE1.00
  17. 041.00
  18. AlEngineer0.95
  19. AlEngineer1.00
  20. 20260.98
  21. EUROPE1.00
• 

  6:54 #20 done15 line(s)

  shot 20·sharpness 1087.6

  1. 24,571.00
  2. 935.61.00
  3. AIE1.00
  4. Evaluate1.00
  5. ★1.00
  6. ★1.00
  7. ★1.00
  8. 730,91.00
  9. 40.97
  10. AlEngineer0.98
  11. EUROPE1.00
  12. AlEngineer0.97
  13. AlEngineer1.00
  14. 20260.99
  15. EUROPE1.00
• 

  7:13 #21 done33 line(s)

  shot 21·sharpness 1372.1

  1. Skill review ~ Linting0.95
  2. Validation1.00
  3. 100%1.00
  4. Validation - 11 / 11 Passed0.97
  5. AIE1.00
  6. CRITERIA0.99
  7. DESCRIPTION0.99
  8. ★1.00
  9. skill_md_line_count1.00
  10. SKILL.md line count is 354 (≤ 500)0.96
  11. ★1.00
  12. ★1.00
  13. frontmatter_valid1.00
  14. YAML frontmatter is valid0.99
  15. name_field0.95
  16. 'name' fleld is valid: terrafom-style-guide'0.90
  17. 40.95
  18. description_field1.00
  19. description' field is valid (165 chars)0.98
  20. compatibility_field1.00
  21. 'compatibility' field not present (optional)0.95
  22. allowed_tools_field1.00
  23. allowed-tools' field not present (optional)0.98
  24. metadata_version1.00
  25. 'metadata' version not present (optional)0.98
  26. metadata_field1.00
  27. 'metadata' fild not present (optional)0.94
  28. https://docs.tessl.io/evaluate/evaluating-skills1.00
  29. AlEngineer0.99
  30. EUROPE1.00
  31. Engineering the future of Al0.99
  32. AlEngineer1.00
  33. 20260.99
• 

  7:53 #22 done36 line(s)

  shot 22·sharpness 1550.1

  1. Skill syntax check~Grammarly1.00
  2. Discovery1.00
  3. 75%1.00
  4. Based on the skills description, can an agent find and select0.99
  5. This is a solid description with clear what' and when' clauses and good distinctiveness for Terraform-specific work. The main weaknesses are moderate0.99
  6. specificity (lacks concrete action examples) and limited trigger term coverage (missing common user phrases like TaC or 'tf files).0.98
  7. *★★0.53
  8. Suggestions1.00
  9. AIE1.00
  10. ★1.00
  11. Add specific concrete actions like 'create modules, define resources, configure providers, set up backends to improve specificity0.99
  12. ★1.00
  13. ★1.00
  14. ★1.00
  15. ★1.00
  16. Expand trigger terms to include common variations: infrastructure as code', 'IlaC'", 'tf files', 'terraform modules', 'cloud infrastructure'0.96
  17. 40.90
  18. Dimension0.97
  19. Reasoning1.00
  20. Score0.98
  21. Specificity1.00
  22. Names the domain (Terraform HCL) and mentions style conventions and best practices, but doesn't list0.99
  23. 2/30.98
  24. specific concrete actions like 'create modules', 'define resources', configure providers', or 'set up state0.97
  25. backends'.0.99
  26. Completeness1.00
  27. Clearly answers both what (Generate Terraform HCL code following HashiCorp's official style conventions0.99
  28. 3/31.00
  29. and best practices) and when (Use when writing, reviewing. or generating Terraform configurations') with0.99
  30. explicit trigger guidance.0.98
  31. https://docs.tessl.io/evaluate/evaluating-skills1.00
  32. AlEngineer0.98
  33. EUROPE1.00
  34. Engineering the future of Al0.99
  35. AlEngineer1.00
  36. 20260.99
• 

  8:36 #23 done34 line(s)

  shot 23·sharpness 1627.4

  1. Skill syntax check~Grammarly0.99
  2. Discovery1.00
  3. This is a solid description with clear what' and when' clauses and good distinctiveness for Terraform-specific work. The main weaknesses are moderate0.98
  4. specificity (lacks concrete action examples) and limited trigger term coverage (missing common user phrases like 'TaC or 'tf files).0.98
  5. *★★0.63
  6. Suggestions1.00
  7. AIE1.00
  8. ★1.00
  9. Add specific concrete actions like 'create modules, define resources, configure providers, set up backends' to improve specificity1.00
  10. ★1.00
  11. ★1.00
  12. ★1.00
  13. ★1.00
  14. Expand trigger terms to include common variations: 'infrastructure as code', 'IlaC'", 'tf files', 'terraform modules', 'cloud infrastructure'0.96
  15. Dimension1.00
  16. Reasoning1.00
  17. Score1.00
  18. Specificity0.96
  19. Names the domain (Terraform HCL) and mentions style conventions and best practices, but doesn't list0.99
  20. 2/30.98
  21. specific concrete actions like 'create modules', 'define resources', configure providers', or 'set up state0.97
  22. backends'.1.00
  23. Completeness1.00
  24. Clearly answers both what (Generate Terraform HCL code following HashiCorp's official style conventions0.99
  25. 3/31.00
  26. and best practices) and when (Use when writing, reviewing, or generating Terraform configurations') with0.99
  27. explicit trigger guidance.0.99
  28. https://docs.tessl.io/evaluate/evaluating-skills1.00
  29. AlEngineer0.94
  30. EUROPE1.00
  31. Braintrust1.00
  32. WorkOS OpenAI0.90
  33. AlEngineer1.00
  34. 20260.92

Transcript

276 cues· 4,062 words· 21,490 chars

Chapters

1. 0:00 Introduction to the talk
2. 1:14 Why context is the new code
3. 2:37 Introducing the Context Development Lifecycle
4. 3:50 Generate: Creating context for agents
5. 6:26 Evaluate: Testing your context
6. 13:59 Distribute: Sharing and packaging context
7. 17:49 Observe: Monitoring and feedback loops
8. 22:33 Conclusion and the context flywheel
9. 24:49 Q&A session