Lenny's PodcastWhat happens after coding is solved? | Fiona Fung (Claude Code & Cowork)
CHAPTERS
- 0:00 – 4:03
Why “coding is solved”: 8× code shipped and ambition becomes the bottleneck
Lenny opens with Anthropic’s claim that engineers are shipping ~8× more code than in 2025, setting up the core thesis: coding speed is no longer the primary constraint. Fiona frames the new frontier as higher ceilings for builders—what matters now is ambition, judgment, and verification.
- •Anthropic engineers shipping dramatically more code quarter-over-quarter
- •Coding no longer the bottleneck; capability ceiling has risen
- •Shift from “can we build it?” to “how ambitious should we be?”
- •High agency paired with high accountability as a team value
- 4:03 – 9:57
Fiona’s 25-year arc: from Vim and OS-level work to IDEs, dogfooding, and shipping online
Fiona traces pivotal shifts in engineering practice: early low-level IBM work, discovering IDEs at Microsoft, and the move from CD-based release cycles to online shipping. These transitions foreshadow today’s AI-driven step-change and her long-standing emphasis on dogfooding.
- •IBM DB2/OS services era: terminal tools, Vim, manual debugging
- •Joining Microsoft/Visual Studio: IDEs as a major productivity leap
- •Shipping on CDs drove heavy planning and hard deadlines
- •Online shipping changed cadence, iteration speed, and feedback loops
- •Dogfooding as a core mechanism for fast, authentic feedback
- 9:57 – 10:03
An AI-native team in 2026: roles blur and everyone becomes a builder
Asked what an “AI-pilled” software team looks like, Fiona describes role boundaries dissolving across engineering, design, and PM. The emphasis moves toward end-to-end builders who can conceive, ship, and iterate rapidly—while maintaining quality through better verification.
- •Roles blur: more disciplines commit code, not just engineers
- •Team identity shifts toward “builders” with end-to-end ownership
- •Throughput explosion forces new verification and quality approaches
- •Quality systems must scale as contribution sources diversify
- 10:03 – 12:42
Manager operating system: Claude sessions for visibility, coaching, and outcome reviews
Fiona explains a new management practice: running a persistent Claude Code session with access to repos, Slack, and metrics to review monthly progress with teams. Instead of relying on manual summaries, she uses AI to surface what shipped, how it performed, and where quality hotspots may be emerging.
- •Claude Code instance connected to repos, Slack channels, and metrics
- •Monthly “look back” sessions: what shipped, outcomes, and feedback loops
- •Using AI to spot incident patterns and propose investment areas
- •Management shifts from tracking activity to coaching on impact and learning
- 12:42 – 14:41
Routines: automating feedback triage into summaries and ready-to-review PRs
Fiona describes how “routines” turn her morning feedback ritual into an automated pipeline. Claude monitors multiple feedback sources, summarizes themes, and can even draft PRs so leaders and teams can review and ship faster with less manual overhead.
- •Feedback arrives from many sources (internal, email, social, partners) consolidated in Slack
- •Routines automate daily monitoring and thematic summarization
- •Agents can propose fixes and generate PRs for human review
- •Scaling throughput requires automating the “keeping up” work
- 14:41 – 16:56
Code review after AI: frameworks, specs-in-repo, and verification-first thinking
Code review becomes a major bottleneck when output accelerates. Fiona explains how Claude helps by validating changes against explicit frameworks—especially when specs and “what good looks like” are checked into the repo—while reserving deep human review for critical areas.
- •Human review remains vital for deep subject-matter areas
- •Claude review scales when given clear validation frameworks
- •Keeping specs in the repo enables automated checks against intent
- •Test-driven development becomes easier when tests can be generated automatically
- 16:56 – 19:38
Who to hire now: creative product builders + deep systems experts
Fiona outlines two hiring profiles that matter most in AI-native teams: product-sense builders who dream, ship, and iterate, and deep systems experts who can verify and handle hard infrastructure challenges. The conversation ties this to the broader “ambition shift,” where engineers can now tackle unfamiliar domains with AI support.
- •Need for distributed systems expertise alongside product generalists
- •“Trust but verify”: deep experts are crucial to validate high-stakes areas
- •Creative builders own delight, iteration, and end-to-end product quality
- •Ambition increases as AI enables cross-domain execution (e.g., mobile work by non-mobile engineers)
- 19:38 – 25:49
Thriving vs. resisting: growth mindset, fear, and “what’s within my control?”
Fiona contrasts engineers who flourish with AI tools against those who resist: the winners lean in with curiosity and a growth mindset. She reframes frustration as often rooted in fear and encourages focusing on controllable actions, illustrated through personal stories from her early life and career.
- •Growth mindset: what made you successful before may not work now
- •Resistance often masks fear and perceived loss of control
- •Practical reframing: identify one action within your control
- •Doing “scary” things periodically is a path to continued growth
- 25:49 – 31:43
Bridging the AI divide: small businesses, community impact, and teaching through use cases
Fiona shares her passion for helping small businesses adopt AI, rooted in childhood experiences and community-building through local shops. She explains how Cowork can eliminate painful admin work and argues that spreading adoption happens best by sharing concrete, relatable use cases with people who are hesitant.
- •Small business owners are time-constrained and margin-constrained—AI can be leverage
- •Cowork as “magic” for invoices, PDFs, and expense workflows
- •Unexpected use cases: document search, menu retrieval, market-style analysis
- •Advice: start conversations by sharing a life-changing use case; make tools more equitable
- 31:43 – 35:08
How Anthropic finds the next big thing: latent demand and smoothing “hoop-jumping” behavior
The discussion turns to why Anthropic seems early to major opportunities like coding and knowledge work. Fiona describes watching for latent demand—users stretching products in unintended ways—and then improving the experience where people are clearly “jumping through hoops.”
- •Latent demand: noticing emergent user behavior outside intended use
- •Rapid internal dogfooding provides fast iteration loops
- •Turning workaround behaviors into first-class product experiences
- •Hypothesis-driven product bets informed by real usage patterns
- 35:08 – 49:53
Next frontier: async fleets of agents, high agency + accountability, and ROI over token-maxing
Fiona predicts a move toward asynchronous work where routines spawn agents that run independently and return drafts, summaries, and PRs. They discuss the cultural and operational implications: granting autonomy, enforcing accountability, and shifting measurement from raw activity (tokens, lines of code) to outcomes and impact.
- •Routines act like cron jobs that kick off agents to do real work
- •As verification improves, agents can gain more autonomy to “go for it”
- •High agency must be paired with clear accountability and hypotheses
- •Productivity metrics are slippery; prioritize outcomes over motion
- •Frameworks like “bad vs. sad” help standardize quality focus across surfaces
- 49:53 – 1:08:32
Keeping humans strong: managers as ICs, preventing atrophy, and reducing loneliness
Fiona explains why Anthropic managers start as ICs and continue doing hands-on work: to stay in the flow, build rapport, and keep touch with product reality. They also explore what’s lost—flow and social coding—plus countermeasures like pairwise programming lunches and hackathons.
- •Managers start as ICs to learn the codebase/tools before people management
- •Dogfooding + small PRs keep leaders close to quality and user experience
- •Skill atrophy concerns: still valuable to “double-click” into dependencies
- •Loneliness increases with agent-heavy workflows; pair programming and hackathons rebuild connection
- •Teams learn rapidly by observing each other’s Claude workflows
- 1:08:32 – 1:38:44
What’s still unresolved: context switching, planning cadence, engineering education, and culture at scale
Fiona revisits open questions: how far to automate reviews, how org structures (like mobile) evolve, and how to manage context switching with many agents running. She also describes shifting from six-month roadmaps to just-in-time monthly planning, worries about educating the next generation, and highlights culture as her biggest “keeps me up at night” concern.
- •Open questions: mobile org structure, automated review limits, equal productivity amid role blur
- •New pain: context-switching load with many async agents
- •Planning shifts to lightweight, monthly JIT priorities with weekly checks
- •Uncertainty in training future engineers—possible move toward apprenticeship models
- •Culture as a living system: maintaining “one team” as the org grows fast
