Overview of What happens after coding is solved? with Fiona Fung
Fiona Fung, manager of Claude Code and Co-work at Anthropic, argues that AI has fundamentally changed software engineering: coding is no longer the bottleneck, verification and product judgment are. In this conversation with Lenny Rachitsky, she shares how her teams now work in an AI-native way, how she thinks about hiring, quality, management, culture, and why the biggest advantage in this new era is high agency paired with high accountability.
Key takeaways
- Coding has shifted from scarce labor to abundant output. At Anthropic, engineers are reportedly producing far more code per quarter than before, but the real challenge is now deciding what matters and ensuring quality.
- The role of the engineer is broadening. Engineers, PMs, designers, and other functions are all becoming more “builder-like,” with more people checking in code and contributing directly.
- Verification is the new bottleneck. Fiona emphasizes that the key challenge is no longer generating code, but making sure the code matches the intended product experience.
- High agency is rewarded. The people thriving most are proactive, ambitious, and willing to take ownership. But that freedom must come with clear accountability and a hypothesis for what they’re trying to solve.
- The future is more asynchronous. Routine-based, agent-driven workflows are pushing teams away from synchronous prompting and toward scheduled, automated task execution.
How AI is changing software engineering
Coding is no longer the limiting step
Fiona describes a major shift in the engineering process: engineers can now generate code much faster, and the job has expanded into deciding what to build, how to validate it, and how to measure whether it actually works.
The work is becoming more product-oriented
Because AI can handle more of the implementation, engineers are increasingly expected to:
- think like product builders
- understand user impact
- monitor feedback loops
- own outcomes, not just outputs
More disciplines are shipping code
On the Claude Code team, even non-engineering roles like PMs and designers are contributing directly to code. That increases velocity, but also increases the need for shared quality standards.
How Fiona manages AI-native teams
Routines replace a lot of manual manager work
Fiona uses Claude routines to automate parts of her daily management workflow:
- scanning feedback channels
- surfacing themes and bugs
- generating PRs and proposed fixes
- preparing summaries for review
This allows her to spend more time in higher-value conversations with her team instead of manually triaging everything herself.
She uses Claude as a management assistant
Rather than only using AI to ship code, Fiona uses it to:
- stay on top of feedback
- spot quality issues
- identify trends across incidents
- prepare for coaching and review conversations
Planning has become “just in time”
She says the team has moved from long planning cycles to lightweight monthly planning, with weekly check-ins because the environment changes too quickly for longer-term plans to stay reliable.
Quality, reviews, and the new bottlenecks
Human review still matters, but must be scaled differently
For deep subject-matter areas, human expertise remains important. But AI can now help validate against clear frameworks and standards.
“Trust but verify” is the operating principle
Fiona recommends:
- codifying what “good” looks like in the repo
- keeping specs up to date
- using AI to validate against those specs
- investing heavily in tests, evals, and monitoring
Better metrics are about outcomes, not output
She warns against vanity metrics like:
- lines of code
- token usage
- raw throughput
Instead, teams should focus on whether the work is actually moving the desired product or business outcome.
What Fiona looks for when hiring
1. Creative builders with product sense
These are people who:
- see a problem
- build something
- look at feedback
- iterate until it feels polished and delightful
2. Deep systems experts
As AI increases the number of people who can contribute, teams still need experts for the hardest, most technical parts:
- distributed systems
- mobile platforms
- architecture
- reliability and verification
Her hiring philosophy is essentially: broad builder energy plus deep expertise where it matters most.
What helps people thrive in the AI era
Growth mindset
Fiona repeatedly returns to one idea: what made you successful before may not work now. The people adapting well are:
- curious
- willing to learn
- open to changing habits
- willing to revisit old assumptions
Lean into fear
Her advice to people who feel overwhelmed or resistant:
- identify what is in your control
- take one concrete action
- don’t let fear turn into passivity
Keep doing hard/scary things
She believes growth comes from stepping into uncomfortable territory, whether that’s learning a new platform, shipping outside your comfort zone, or changing how you work.
AI adoption and avoiding a widening gap
Fiona is especially concerned about the divide between people who are leaning into AI and those who are resisting it. She believes the best way to help others is by:
- sharing concrete use cases
- starting conversations with friends, family, and small businesses
- showing how AI can save time on annoying, low-value tasks
Her broader goal is to make the tools more equitable and prevent the gap from widening.
Culture, loneliness, and team health
Culture is a living thing
Fiona says team culture must be actively maintained as the organization grows. It’s not a poster on the wall; it shows up in how people treat each other.
Her biggest fear: hidden problems
She wants people to be honest when things are not going well. Her nightmare scenario is a manager saying “everything is fine” when it isn’t.
AI can make engineering feel lonelier
Because so much work now happens with agents, some engineers miss the old “flow” of collaborative coding. To counter this, her team started:
- pair programming lunches
- hackathons
- more in-person or synchronous collaboration time
Product philosophy: dogfooding, anecdote, and latent demand
Dogfood everything
Fiona strongly believes leaders should use the product they build. It helps them:
- feel the product
- spot issues faster
- maintain quality standards
- stay connected to real user pain
Listen to anomalies and latent demand
She also stresses that users often use products in unexpected ways. Those unexpected behaviors can reveal entirely new product opportunities.
Examples she shared:
- small businesses using Co-work for expense management
- restaurant owners using it to find menus in cluttered folders
- using AI for market comparison and pricing decisions
Notable quotes and ideas
- “High agency is also high accountability.”
- “Make new mistakes.”
- “What is within my control?”
- “Don’t forsake motion for progress.”
- “Keep it simple.”
- “In a world where you can be anything, be kind.”
Practical recommendations for teams
- Move from manual management to agent-assisted routines
- Define success criteria in the repo with specs, tests, and evals
- Measure outcomes, not just output
- Keep planning lightweight and adaptive
- Preserve culture through open, honest conversations
- Use the product constantly, or meet customers directly
- Be willing to delete outdated processes that no longer serve the team
Final thought
Fiona’s core message is that AI hasn’t eliminated the need for great engineers and leaders—it has shifted the job upward. The winners in this new world will be the people and teams who can combine ambition, taste, systems thinking, and accountability while using AI to move faster than ever before.