LLenny's Podcast
← All episodes
Fiona Fung (Manager of the Claude Code and Cowork Teams)21 June 2026

What happens after coding is solved?

10Frameworks
15Insights

Frameworks in this episode

Productivity3 steps

Bad vs Sad Quality Tiers

Classify every failure as bad (irrecoverable) or sad (recoverable pain) so teams triage quality across many surfaces.

Read →
Mindset4 steps

Fear-to-Agency Reframe

When AI change triggers fear, lean in and ask what's within your control — turn it happening to you into happening for you.

Read →
Leadership4 steps

High Agency, High Accountability

Pair the freedom to solve problems your own way with clear ownership of the hypothesis and the outcome.

Read →
Strategy4 steps

Just-in-Time Planning

Shrink long roadmaps to a lightweight monthly priority list and grant explicit permission to kill dead processes.

Read →
Innovation4 steps

Latent Demand Mining

Watch for people jumping through hoops to make your product do something, then make that the smooth path.

Read →
Leadership4 steps

Leader Dogfooding for Product Pulse

Live and breathe your product as a real user (or meet customers) and trust the anecdote over the dashboard.

Read →
Leadership3 steps

Manager-as-IC Onboarding

New managers ship as individual contributors first — and keep doing it — before and while they manage people.

Read →
Leadership4 steps

Manager Visibility via an Always-On Agent

Enlist a standing AI session across all repos and channels to stay on top of 8x throughput and drive conversations.

Read →
Strategy4 steps

Output-Toward-Outcome Measurement

Don't forsake motion for progress — keep asking whether the metric you climb still serves the outcome.

Read →
Productivity4 steps

Spec-as-What-Good-Looks-Like Verification

Check your definition of good into the repo so AI code review can automatically validate work against it.

Read →

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Hot Take· 4

Hot Take11:00

"Make New Mistakes" — Fiona's Rule Against Zero-Defect Thinking

Fiona shares a personal management saying: it's fine to make mistakes as long as they're new ones, because that means you're learning. Aiming for zero mistakes is a signal you're moving too slowly or being too cautious.

  • Mistakes are acceptable if they're new — repeated ones mean you're not learning
  • A zero-mistake target implies moving too slow or being overly cautious
  • Frames failure as a byproduct of healthy velocity

I have this saying, "Make new mistakes."

Fiona Fung · 11:00

just make new ones so that we're always learning cuz if you aim to make zero mistakes, like that probably means you're not, you know,…

Fiona Fung · 11:30
#management#culture#learning#velocity
Hot Take16:00

TDD Is Fun Again When Claude Writes the Failing Test First

Fiona always found test-driven development valuable in principle but painful — like eating your broccoli first before the thrill of shipping. Now that Claude can write the failing test, run it, and then implement the fix, the 'tax' of test generation is automated and old principles like TDD become newly efficient.

  • TDD was sound in principle but felt like a tax you had to pay first
  • Fiona's first Claude Code bug fix used a full TDD loop: write failing test, then fix, then pass
  • Automating test generation revives long-standing principles that models now make efficient

it was almost like you have to eat the broccoli first.

Fiona Fung · 16:00

Yeah, like that's it's so unfair. It just writes the test for you first.

Lenny · 16:30
#tdd#testing#claude-code#workflow
Hot Take19:00

The New Bottleneck Is Ambition, Not Ability

With coding no longer the bottleneck, the constraint has shifted to how ambitious you can be. Fiona describes an engineer who isn't a mobile specialist shipping a mobile feature with Claude as a partner — the tools have lifted the ceiling of what anyone is able to do.

  • Everything is possible in theory; the question is how big you can think
  • A non-mobile engineer shipped a mobile feature by partnering with Claude
  • The mindset shift: from 'that's hard and complicated' to 'that's so possible'

It's like the it's lifted the ceiling of what and anyone is able to do.

Fiona Fung · 19:00
#ambition#agency#productivity#mindset
Hot Take40:00

Don't Mistake Motion for Progress When Measuring AI ROI

Fiona walks through the failed hunt for a productivity metric — lines of code, throughput, time-to-land-PR — and warns that token-maxing and tool-usage metrics measure action, not outcome. Her Facebook Marketplace example: an early region had few sellers but buyers found what they needed because of power sellers, so the team fixed the metric rather than blindly following it.

  • Lines of code, throughput, and time-to-land-PR all break as productivity proxies
  • Measuring tool usage measures the action, not whether the outcome improved
  • Zoom out: what problem are we solving, and what's a good way to measure that?
  • Marketplace: low seller count but power sellers met the real goal — the metric was changed, not obeyed

there there was another saying I really like, don't forsake motion for progress.

Fiona Fung · 41:00

in this area the number of sellers is low, but actually people are like people are finding items that they're looking for which is what…

Fiona Fung · 43:00
#metrics#productivity#roi#measurement

Explainer· 2

Explainer13:00

Routines: Automating the Manager's Morning Feedback Ritual

Fiona's old morning ritual was coffee plus manually scanning feedback channels for gaps she could help close. Since Anthropic launched routines a month or two ago, she has an agent that automatically surveys feedback channels overnight, summarizes the themes, and even drafts PRs she can review when she wakes up.

  • Old workflow: manually read feedback channels each morning for maker-time opportunities
  • Routines run on a schedule and generate the prompts, not just execute them
  • Wakes up to a summary of themes plus ready-to-review PRs
  • Feedback comes from internal channels, emails, LinkedIn, socials, and partnerships, all posted to Slack

I think maybe a month or two ago we we launched routines. And that's also completely changed. Like now I just have a routine that…

Fiona Fung · 13:00
#routines#automation#workflow#feedback
Explainer45:00

"Bad vs Sad": A Simple Framework for Reading Quality Across Surfaces

To make sense of quality across many product surfaces, Fiona's teams classify issues as 'bad' — a very bad, irrecoverable error like a crash where you lose work — or 'sad' — a recoverable pain point like UI flickering. Stacked-up sads can accumulate into a bad. Each team gets agency to define what counts as bad vs sad for the surfaces they own.

  • 'Bad' = irrecoverable error (e.g. a CLI crash where you lose work)
  • 'Sad' = recoverable pain point (e.g. flickering)
  • Enough stacked sads can escalate into a bad
  • Beats raw reliability/perf dashboards for zooming out on overall experience; each team defines its own thresholds

bad is like a very bad irrecoverable error. And sad is something that's kind of like a pain point recoverable, but it's interesting when you…

Fiona Fung · 45:30
#quality#framework#reliability#product

Story· 2

Story10:00

How Fiona Uses a Claude Code Session to Manage Her Whole Org

Fiona runs a Claude Code remote session enlisted across all her team's repos, with access to Slack channels and metrics, giving her full visibility into everyone's work. Where she once used such sessions only to generate PRs and bug fixes, she now uses them to prep monthly retrospectives and have grounded conversations with the people she supports about what shipped and how it did.

  • The session is enlisted in every repo and has access to Slack channels and tracked metrics
  • Used monthly to review focus areas, what shipped, how products did, and feedback channels
  • Shifted from generating PRs/bug fixes to enabling real conversations with reports
  • Surfaces themes across incidents to spot quality hot spots that used to be a manual process

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…

Fiona Fung · 10:00
#management#claude-code#visibility#quality
Story56:30

AI Coding Got Lonely — So the Team Started Pairwise Programming Lunches

As everyone on the Claude Code team started working heads-down with their own agent, it began to feel lonely. In response the team introduced pairwise programming lunches and hackathons, and discovered that because everyone uses Claude Code and Cowork so differently, watching a colleague work teaches you something every time.

  • Working solo with an agent all day created a lonely experience
  • The team started pairwise programming lunches and hackathons to reconnect
  • Everyone uses the tools so differently that watching others is genuinely instructive
  • Lenny likens it to 'parallel play' — working next to each other while building your own things

after a while we felt uh it could start being a lonely experience cuz we all started just working with our agent so much. So…

Fiona Fung · 56:30
#team-culture#collaboration#loneliness#pair-programming

Takeaway· 7

Takeaway15:00

Check Your Spec Into the Repo So Claude Can Review Against It

Fiona's key unlock for automated code review is giving Claude an explicit framework of what "good" looks like. Any statement of quality — a content-design scale, a spec — should be checked into the repo and kept in sync with the code, so Claude Code review can validate every change against it.

  • Claude reviews reliably when given an explicit framework to validate against
  • Check specs and 'what good looks like' statements into the repo
  • Keep the spec updated alongside the code, frequently
  • Human reviewers still handle areas needing deep subject-matter expertise

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

Fiona Fung · 15:00

any time you have like a statement of what good looks like, get that into the repo and then Claude code review can make sure…

Fiona Fung · 15:30
#code-review#specs#quality#claude-code
Takeaway17:00

The Two Profiles Fiona Hires For Now

For the Claude Code team, Fiona hires two profiles: creative builders with product sense who own a product end-to-end, and deep systems experts for the hard parts that still need verification. She realized early the team had strong product generalists but was missing distributed-systems depth.

  • Profile 1: creative builders/'dreamers' with product sense who iterate, polish, and own end-to-end
  • Profile 2: deep subject-matter/systems experts for the parts that need verification
  • The team was strong on product generalists but short on systems and distributed-systems depth
  • 'Trust but verify' — invest human expertise wherever verification is critical

the two profiles that you now look for when you're hiring are creative builders with product sense and deep systems experts for the hard parts.

Lenny · 17:00

it's all about trust but verify. The models are really good, but there are definitely a lot of areas that still need verification.

Fiona Fung · 17:30
#hiring#team-building#engineering#systems
Takeaway20:00

Why Some Engineers Thrive on AI and Others Resist: Growth Mindset and Fear

Fiona attributes thriving in the AI transition to a growth mindset — always learning, and accepting that what made you successful may no longer serve you. She sees frustration as often rooted in fear, and her advice is to lean in and ask what's within your control, reframing change as happening 'for you' rather than 'to you.'

  • Growth mindset = always learning; past success habits may need to change
  • Frustration often masks fear and a feeling that everything is outside your control
  • Antidote: find one action within your control; shift from 'happening to me' to 'happening for me'
  • Do something scary occasionally — you'll dip while learning, but that's how you keep growing

for anything that you know, that that there is a fear my advice is kind of lean in and ask, "Okay, is there some what…

Fiona Fung · 21:00

the cave you fear contains the treasure you seek.

Lenny · 25:00
#mindset#fear#career#growth
Takeaway38:30

High Agency Comes With High Accountability

Agency is a core value on the Claude Code and Cowork teams — everyone gets the freedom to propose how to solve a problem. But Fiona pairs it with accountability: freedom to build is balanced by owning the outcome and having a clear hypothesis of what you're trying to solve.

  • Everyone on the team owns ideas for how to address a problem
  • High agency is deliberately coupled with high accountability
  • Accountability ties back to having a clear hypothesis of what you're solving

it's really high agency and then we say with high agency is also high accountability. So it's all about like making sure folks have that…

Fiona Fung · 39:00
#agency#accountability#culture#management
Takeaway49:30

Every Manager Starts as an IC — and Keeps Doing PRs

Fiona has new managers begin as individual contributors before taking on people responsibility, giving them maker time to learn the codebase and build rapport as a teammate first. Even for herself, the PRs she does matter less for what they fix than for keeping her in the flow and in daily touch with a fast-changing product.

  • New managers start as ICs to learn the codebase before shouldering people responsibility
  • Doing the work first builds genuine rapport with the team
  • Fiona's own PRs are about staying in touch with the product, not the fix itself
  • Leaders who don't live and breathe the product lose touch and feel for it

even me doing PRs it's less about what it is I'm fixing, it's more about me using the the product every day just to keep…

Fiona Fung · 51:00
#management#player-coach#leadership#dogfooding
Takeaway1:03:30

Dogfood Relentlessly — and Trust the Anecdote Over the Data

Fiona's defining habit is using the product she builds constantly, which surfaces things dashboards miss — a scam vector on Marketplace, floor-height bugs in VR she uniquely reproduced. She and Lenny land on the Bezos principle: when you have both the data and an anecdote, trust the anecdote, and if you can't use the product, meet customers instead.

  • Constant dogfooding keeps a leader's pulse on the real product experience
  • Personal use surfaced a Marketplace scam vector and VR floor-height bugs others missed
  • Bezos: with data and an anecdote in conflict, trust the anecdote
  • If you can't use the product, do customer visits — a Chile trip revealed slow-LTE feed loading blocking growth

that's how you keep your pulse on the product that you're building and don't get too lost in metrics and dashboards only or presentations.

Fiona Fung · 1:05:30

I think it was Jeff Bezos that said if you have the data and you have an anecdote, trust the anecdote over the data.

Lenny · 1:08:00
#dogfooding#customer-research#product#leadership
Takeaway1:23:00

Explicit Permission to Kill Processes — and Just-in-Time Planning

A pillar of the Claude Code and Cowork culture is explicit permission to kill any process that no longer serves the team. Fiona applied it to her own planning: a six-month roadmap went stale within three months, so she moved to 'JIT planning' — a lightweight monthly priority spreadsheet checked weekly to confirm it's still current.

  • Team norm: explicit permission to kill noisy, expensive, or dreaded processes
  • Test any process by asking whether it still serves its purpose
  • A six-month roadmap became stale in three months as the landscape changed
  • JIT (just-in-time) planning: a lightweight monthly priority sheet, re-checked each week

one other thing that's really big on Cloud Code and Core team culture is explicit permission to kill processes that no longer serve us.

Fiona Fung · 1:23:00

I I call it JIT planning now, like just-in-time planning.

Fiona Fung · 1:24:30
#planning#process#culture#management