Anthropic engineering leader reveals how AI drove an eightfold increase in code output

Building the most AI-pilled engineering team in the world | Fiona Fung (Manager of the Claude Code and Cowork Teams)

As AI tools automate manual coding, engineering leadership must evolve to manage massive code volumes, asynchronous agent workflows, and shifting developer roles.

3 key takeaways
  1. 1AI tools have driven an eightfold increase in code output, forcing engineering teams to adopt asynchronous workflows and automated agent routines.
  2. 2Hiring priorities are shifting toward creative builders with strong product sense and deep systems experts who can orchestrate complex architectures.
  3. 3Traditional engineering metrics like lines of code are obsolete, replaced by proactive quality monitoring and developer frustration tracking.

Don't miss

Fiona Fung contrasts her early days shipping software on physical CDs at IBM with today's real-time, agent-driven engineering workflows.

The brief

Anthropic engineering leader Fiona Fung manages the teams behind Claude Code and Claude Cowork, experiencing firsthand a world where software engineering output has increased eightfold because coding is no longer the bottleneck.

This massive code spike shifts the developer role from manual syntax writing to high-level system orchestration, requiring engineers to deploy automated routines that coordinate fleets of autonomous AI agents on their behalf.

To manage this rapid output without losing control, Fung relies on proactive quality monitoring, tracking critical errors via specific thresholds and keeping developer frustration low through pairwise programming lunches.

Fung reflects on her 25-year career, comparing the slow feedback loops of shipping database software on physical CDs at IBM to the real-time, agent-driven workflows she manages today.

As traditional metrics like lines of code become obsolete, the most successful engineering teams will prioritize creative builders with strong product sense over developers who merely write code.

What was said on this episode

66 statements · 47 positive · 5 negative · 5 mixed · 9 neutral

  1. Anthropic engineers now produce eight times more code per quarter than in 2025.

    “Anthropic engineers on average have eight times as much code per quarter as they did compared to 2025.”

    Listen at 0:00

  2. Fiona Fungon AI-assisted codingPositive0:05

    AI-assisted coding removes coding as the bottleneck and expands individual capability.

    “Coding is no longer the bottleneck. It's lifted the ceiling of what anyone is able to do.”

    Listen at 0:05

  3. Fiona Fungon AI-assisted codingPositive0:05

    AI-assisted coding is no longer the main engineering bottleneck.

    “Coding is no longer the bottleneck.”

    Listen at 0:05

  4. Fiona Fungon AI-assisted engineering workNegative0:36

    Agent-heavy engineering work can become lonely.

    “It could start being a lonely experience because we all started just working with our agents so much”

    Listen at 0:36

  5. Fiona Fungon AI-related fearPositive0:58

    People facing AI-related fear should engage with it and focus on controllable actions.

    “My advice is lean in and ask what can I do about it? What is within my control?”

    Listen at 0:58

  6. Fiona Fungon codingPositive9:02

    Coding is no longer the primary engineering bottleneck.

    “coding is no longer the bottleneck”

    Listen at 9:02

  7. Higher AI-enabled software throughput shifts engineering attention toward verification.

    “How do we think about verification? Like that's kind of this other shift that I'm seeing.”

    Listen at 9:23

  8. Fiona Fungon Claude CodePositive10:13

    A Claude Code session can provide a manager visibility across team repositories.

    “I actually have a cloud code remote session that I enlist in all of our repos. And so this way I have full visibility into the work that everybody's doing.”

    Listen at 10:13

  9. Fiona Fungon ClaudePositive11:42

    Claude can analyze incidents to identify themes and suggest future quality investments.

    “by having Claude, like, so then it can also look at, like, some, like, for example, actually, yeah, we were. I was just looking at. Oh, you know, given some of the, you know, incidents that we have, like, let's look across all of this. Can we generate a theme? Like, what's a good area investment for us next?”

    Listen at 11:42

  10. Fiona Fungon ClaudePositive12:07

    Claude enables engineering managers to derive insights that were previously impractical.

    “I don't think I would have been able to, you know, have some of these insights with Claude.”

    Listen at 12:07

  11. Fiona Fungon Claude routinesPositive13:11

    Claude routines can automate managers’ feedback-monitoring workflows.

    “now I just have a routine that automates all this for me.”

    Listen at 13:11

  12. Fiona Fungon Claude routinesPositive13:11

    Claude routines can automate recurring feedback review for managers.

    “now I just have a routine that automates all this for me”

    Listen at 13:11

  13. Fiona Fungon Human code reviewPositive15:05

    Human experts should continue reviewing code in areas requiring deep subject expertise.

    “for the important like, like areas that need deep subject matter expertise, we definitely want to make sure we have the proper, you know, like humans. Still reviewing.”

    Listen at 15:05

  14. Fiona Fungon ClaudePositive15:20

    Claude performs well when validating work against explicit quality frameworks.

    “Claude is very good when you give it a framework to validate against those frameworks.”

    Listen at 15:20

  15. Fiona Fungon Repository specificationsPositive15:33

    Teams should store and regularly update product specifications in repositories.

    “if you have specs or check those into the repo and then make sure the spec also keeps up to date with the code frequently”

    Listen at 15:33

  16. Fiona Fungon Automated code reviewPositive15:44

    Repositories should contain explicit quality criteria for automated code review.

    “any time you have a statement of what good looks like, get that into the repo and then code review can make sure it's still matching what you set up to do.”

    Listen at 15:44

  17. Fiona Fungon AI-assisted testingPositive16:39

    AI automation can make established testing practices more efficient.

    “the fact that that's now automated and you can even revisit all these principles that have been around for a while, but now they actually might be even more efficient just because you have the models that can do more of the work for you.”

    Listen at 16:39

  18. Fiona Fungon Deep subject-matter expertisePositive17:47

    Teams should continue investing in deep subject-matter expertise where verification is needed.

    “wherever you need the deep subject matter expertise, I would say that's know an area to definitely still invest in.”

    Listen at 17:47

  19. Fiona Fungon ClaudePositive19:34

    Claude enables people to work beyond their traditional technical specialties.

    “It's lifted the ceiling of what anyone is able to do.”

    Listen at 19:34

  20. Fiona Fungon Growth mindsetPositive20:05

    A growth mindset helps people adapt to changing engineering work.

    “A growth mindset really, really helps.”

    Listen at 20:05

  21. Curiosity and continuous learning help people adapt to AI-driven change.

    “always leaning in with curiosity and always being able to learn”

    Listen at 20:58

  22. Fiona Fungon Claude CoworkPositive29:22

    Cowork can locate specific documents within a disorganized directory.

    “we gave Cowork access to the directory, found the menus.”

    Listen at 29:22

  23. Fiona Fungon Claude CoworkPositive29:36

    Cowork can provide local market analysis for a restaurant’s pricing decisions.

    “it came back with really cool, almost like market analysis.”

    Listen at 29:36

  24. Fiona Fungon AI adoption divideNegative31:20

    Sharing AI knowledge can help prevent an expanding adoption divide.

    “I just want to make sure we keep sharing the knowledge and making the tool equitable. Because if not, I'm concerned that the device grows larger and larger.”

    Listen at 31:20

  25. Fiona Fungon Product developmentPositive34:08

    Products should continuously incorporate feedback to improve reliability and quality.

    “always listening to feedback and always iterating, trying to make a delightful, reliable, high quality experience”

    Listen at 34:08

  26. Fiona Fungon Customer product usageMixed34:28

    Customers use products in unintended ways, beneficially or harmfully.

    “customers will use your product in ways that you did not intend for good or for bad.”

    Listen at 34:28

  27. Fiona Fungon Customer workaroundsPositive35:00

    Workarounds can reveal opportunities to simplify product experiences.

    “when you see people jumping through hoops to make something work, can you actually make that an even like a smoother and better experience?”

    Listen at 35:00

  28. Fiona Fungon Engineering workPositive35:36

    Engineering work is shifting toward asynchronous collaboration.

    “I would say we're shifting more towards async, like asynchronous.”

    Listen at 35:36

  29. Fiona Fungon AI-assisted engineering workPositive35:36

    AI-assisted engineering teams are shifting toward asynchronous work.

    “we're shifting more towards async, like asynchronous.”

    Listen at 35:36

  30. Fiona Fungon Claude routinesPositive36:39

    Claude routines can launch agents on a schedule for users.

    “it's able to actually go and kick off agents on your behalf”

    Listen at 36:39

  31. Fiona Fungon AI agent autonomyMixed37:59

    Greater agent autonomy becomes appropriate when verification is sufficiently reliable.

    “if verification is really good”

    Listen at 37:59

  32. Fiona Fungon High-agency teamsPositive39:05

    High-agency teams should pair freedom with accountability.

    “with high agency, it's also high accountability.”

    Listen at 39:05

  33. Fiona Fungon AI productivity measurementPositive41:10

    Productivity should be evaluated by whether output advances intended outcomes.

    “output. Is the output really going towards the outcome?”

    Listen at 41:10

  34. Fiona Fungon Engineering productivity metricsNeutral41:11

    Engineering productivity should be evaluated by outcome contribution rather than output volume.

    “Is the output really going towards the outcome?”

    Listen at 41:11

  35. Fiona Fungon Productivity measurementPositive41:29

    Teams should distinguish activity from genuine progress toward outcomes.

    “don't forsake motion for progress.”

    Listen at 41:29

  36. Fiona Fungon AI tooling adoptionPositive42:04

    Leaders should consult senior engineers when evaluating AI tooling.

    “definitely go on a listening tour for the, for the leaders listening to a podcast especially. You know, I love to focus on the senior engineers as well”

    Listen at 42:04

  37. Fiona Fungon Engineering metricsNeutral42:43

    Teams should regularly reassess whether their metrics still reflect intended outcomes.

    “Always keep in mind, is that metric really still serving the outcome that you were aiming for?”

    Listen at 42:43

  38. Fiona Fungon Organizational metricsPositive43:45

    Teams should regularly reassess whether metrics still serve their intended outcomes.

    “whatever metric, whether for productivity or even for product, always keep an eye and make sure that you're not just having blinders on.”

    Listen at 43:45

  39. Fiona Fungon Software quality managementPositive45:12

    Earlier quality detection improves proactive software quality management.

    “the more proactive we can be of like making sure we can get an earlier detection into quality”

    Listen at 45:12

  40. Fiona Fungon Software quality problemsNegative45:37

    Accumulated recoverable quality problems can become severe failures.

    “when you stack up SADs, it could generally go to bad.”

    Listen at 45:37

  41. Fiona Fungon Software quality thresholdsPositive46:57

    Teams should define their own severe and recoverable quality thresholds.

    “we give high agency to each team of what we think constitutes a bad and what's a sad.”

    Listen at 46:57

  42. Fiona Fungon Engineering managersPositive51:07

    New managers should spend maker time learning the codebase or product before managing.

    “give yourself that maker time to actually dive deep into the code and learn the code base or the product”

    Listen at 51:07

  43. Fiona Fungon AI-assisted engineeringPositive56:00

    Engineers should understand architecture and changes despite using AI tools.

    “make sure you're like taking the time too to all the work that you do still get that understanding of the architecture or the change”

    Listen at 56:00

  44. Fiona Fungon Pairwise programmingPositive56:57

    Pairwise programming helps engineers learn different AI-assisted workflows from one another.

    “when we do pairwise programming, we actually learn so much from each other.”

    Listen at 56:57

  45. Fiona Fungon AI-assisted product managementPositive1:00:31

    AI reduces product managers’ dependence on engineering bandwidth for ideas.

    “PM are no longer bottlenecked of if they have an idea waiting on engineering bandwidth.”

    Listen at 1:00:31

  46. Cross-disciplinary AI-generated contributions increase the need for verification.

    “the verification is important because when you have more different disciplines checking in, how do we make sure everybody has higher confidence?”

    Listen at 1:00:58

  47. Fiona Fungon AnthropicPositive1:02:23

    Most Anthropic commits are assisted by Claude.

    “most commits are cloud assisted”

    Listen at 1:02:23

  48. Fiona Fungon Engineering product sensePositive1:02:39

    Engineers should develop stronger product judgment as AI expands their responsibilities.

    “having engineers build and keep building that stronger product sense muscle”

    Listen at 1:02:39

  49. Fiona Fungon EngineersPositive1:02:39

    Engineers should develop stronger product judgment alongside technical skills.

    “engineers build and keep building that stronger product sense muscle”

    Listen at 1:02:39

  50. Fiona Fungon Technical and product rolesNeutral1:03:24

    Boundaries between engineering and other roles are becoming less distinct.

    “It's all blurring.”

    Listen at 1:03:24

  51. Fiona Fungon Product dogfoodingPositive1:04:23

    Leaders who use their products directly maintain closer awareness of product quality.

    “being able that helps me keep really close to the pulse”

    Listen at 1:04:23

  52. Fiona Fungon Product usageMixed1:05:00

    Real-world product use reveals unexpected behaviors and issues.

    “people will use your products in ways that, that you may not expect.”

    Listen at 1:05:00

  53. Facebook Marketplace performed poorly under Chilean low-LTE conditions.

    “the marketplace feed didn't even load very well on these low LTE situations.”

    Listen at 1:08:10

  54. Fiona Fungon Mobile engineering organizationsMixed1:09:21

    AI reduces the required size of specialized mobile teams but not expertise.

    “we definitely feel that it's really important like it's really important to still bring on folks with those expertise but we probably don't need like as large”

    Listen at 1:09:21

  55. Fiona Fungon AI experience verificationNegative1:10:50

    Verifying that AI-generated experiences match intended quality remains difficult.

    “that one is still a hard one to crack”

    Listen at 1:10:50

  56. Fiona Fungon Software engineering trainingNeutral1:13:55

    Software engineering training may shift toward fellowships or apprenticeships.

    “I wonder if it's for software engineering. It's almost like you go more towards a fellowship or apprenticeship program.”

    Listen at 1:13:55

  57. Fiona Fungon AI modelsMixed1:14:43

    Future models may reduce the importance of deep foundational engineering knowledge.

    “Maybe the models will get good enough that it doesn't matter about it.”

    Listen at 1:14:43

  58. Fiona Fungon Software systems understandingPositive1:14:47

    Understanding underlying systems may reveal opportunities to improve products despite AI abstraction.

    “I do think there's something about that double click because I think that's where there might be an opportunity to improve the product or the system.”

    Listen at 1:14:47

  59. Fiona Fungon AI model capabilitiesPositive1:17:44

    Teams should revisit previously failed AI automations as models improve.

    “it might be worth the time to revisit, because, you know, that now might be a new capability.”

    Listen at 1:17:44

  60. Fiona Fungon AI model improvementsPositive1:17:46

    Previously unsuccessful AI automations may become viable after model improvements.

    “because, you know, that now might be a new capability.”

    Listen at 1:17:46

  61. Organizational culture evolves and requires active maintenance.

    “culture is like a living, breathing thing. It's not just a poster you slap on a wall. And it changes over time”

    Listen at 1:18:23

  62. Fiona Fungon Team processesPositive1:23:15

    Teams should explicitly eliminate processes that no longer serve their purpose.

    “explicit permission to kill processes that no longer serve us.”

    Listen at 1:23:15

  63. Fiona Fungon Team planningPositive1:24:46

    Rapidly changing teams should use lightweight monthly planning.

    “we try to do like a month planning. Like, really lightweight.”

    Listen at 1:24:46

  64. Fiona Fungon Anthropic team planningPositive1:25:14

    Anthropic’s team uses lightweight just-in-time monthly planning.

    “we've shrunk it to jit monthly planning.”

    Listen at 1:25:14

  65. Fiona Fungon AI engineering prioritiesNeutral1:26:44

    Longer-term engineering priorities can change rapidly as the AI landscape evolves.

    “even those themes change, you know, so fast when the landscape changes.”

    Listen at 1:26:44

  66. Fiona Fungon AI development strategyNeutral1:26:44

    Longer-term strategic themes can change rapidly in AI development.

    “those themes change, you know, so fast when the landscape changes.”

    Listen at 1:26:44

Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.

Books & mentions

Lenny's Podcast: Product | Career | Growth

To hear more deep-dive interviews with world-class product and engineering leaders on scaling teams.

Weekly email

State of the Pod, every Monday

The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.

Listen to the full episode and explore every guest, topic, and moment on PodLume.