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

Engineers here are shipping roughly eight times more code than a year ago, so writing code is no longer the choke point—it is ambition, initiative, and what you choose to build. With that in mind, I’m thrilled to have Fiona Fung, who leads Claude Code and CoWork at Anthropic and previously shaped TypeScript and Visual Studio, launched Facebook Marketplace, and ran large orgs at Meta and Instagram.

I’ve lived several step changes, from hacking in VIM at IBM to discovering modern IDEs on Visual Studio, where dogfooding taught me to obsess over developer experience. We moved from printing CDs to shipping online, and now with Claude, coding speed is so high that verification and product quality are the true constraints.

Paint the picture: what does an AI‑pilled engineering team look like in 2026?

Everyone becomes a builder. I keep a Claude Code remote session enrolled across our repos with access to metrics and Slack, then run monthly reviews to trace what shipped, how it performed, and where incidents clustered; routines now watch feedback channels and even draft PRs before I’ve had coffee.

Where does feedback come from, and how are you accelerating code review with confidence?

It pours in from internal users, email, socials, partners, and we post it in Slack for Claude to triage. Reviews sped up once we codified what good looks like—checklists, specs, and content rules live in the repo—so automated checks gate most changes while experts still review the tricky parts; even test‑driven flows feel fun again with Claude writing the tests first.

Who are you hiring now, and how has ambition shifted with AI?

We want deep systems experts where it matters and creative product builders who iterate from real feedback; trust the model, then verify. Claude raises the ceiling—an engineer who was not a mobile specialist shipped a solid mobile feature by pairing with it.

Who is thriving in this transition, and how do those struggling get unstuck?

A growth mindset wins: keep learning and turn fear into a next action you control; that is how I funded school by taking a weekend bank‑teller job. Do something a little scary often, because that is where the learning curve lives.

You care about closing the AI gap, especially for small businesses—what’s working?

Start with your own life‑changing use case and demo it. CoWork saved me from expense hell, and with friends we used it to surface lost menus in a messy folder and price dishes against local comps—practical wins that spark adoption and great product feedback.

Anthropic keeps spotting big openings early—coding, knowledge work, model personality. What enables that?

We chase latent demand and iterate fast when users bend the product in new ways; that is how Claude for Small Business emerged. People use tools in ways you did not plan—listen closely and make the path smooth.

What’s the next frontier for how engineers work—agents, routines?

Workflows are going async with routines that spawn agents and prompts on a schedule, raising the abstraction so managers review outputs and gradually grant more autonomy as verification hardens.

Agency matters, but so does accountability; and how are you measuring ROI without getting lost in token spend?

We pair high agency with high accountability and a clear hypothesis. Measure outcomes, not motion—metrics evolve, so I do listening tours and adjust, like when Marketplace moved from counting sellers to accounting for power sellers because that better predicted buyer success.

Tips for balancing speed with quality at this velocity?

Front‑load detection with tests and monitoring, and label experiences as bad or sad so teams set their own thresholds and goals; even a cheeky swear‑word dashboard helps catch frustration early.

You require managers to ship as ICs and to dogfood—why?

Starting as an IC builds product feel, trust, and judgment before people management; using what we build keeps leaders grounded, and Claude gave me the confidence to ship again with strong tests and onboarding help.

Do skills atrophy when AI writes most code, and how do you fight the loneliness of agent‑driven building?

Still learn architectures and dependencies so you can verify and improve systems; we run pair‑programming lunches and hackathons because watching each other’s Claude flows teaches a ton and restores team energy.

Which other roles changed most, and what should managers now expect from engineers?

PMs and other coding‑adjacent roles now build, so verification across disciplines is key; expect most commits to be Claude‑assisted and push engineers to strengthen product sense and self‑unblock across functions.

You often favor anecdotes alongside data—why?

Using the product daily and meeting customers reveals blockers data can hide, like a marketplace feed failing on slow mobile networks; combine those insights with metrics to prioritize the right fixes.

What big questions are still open for you?

We are rethinking how specialized iOS and Android orgs need to be, how far to push automated reviews, how to verify overall experience quality, and how to manage the rising context‑switching load from many async agents.

Where does hiring and training go from here?

Demand stays strong, but we need new paths—think apprenticeships—to teach the important layers without years of typing; abstractions keep rising, models improve quickly, and the thing that keeps me up is preserving a healthy, open, one‑team culture through hypergrowth.

Culture under hypergrowth is hard, so normalizing honest status and surfacing what is not working seems critical; before we wrap, any operating advice you live by?

Give explicit permission to kill processes that no longer serve the team. We switched to just‑in‑time monthly planning with a lightweight sheet and weekly check‑ins, and we are exploring how to automate even that.

On planes I rewatch three comfort films. Amélie captures the Paris wonder I felt as a teen art student, and two Studio Ghibli staples—Spirited Away and Nausicaä—left a deep mark, with Nausicaä shaping how I think about leadership.

Do you have a recent product you love?

I swear this is not an ad, but Sweet Sisters Body Care, a small island maker of organic hair and skin products, changed my life. A stubborn nose rash vanished when I switched from my old shampoo, and a week of hotel shampoo on the road reminded me to pack travel sizes.

Love the local pick, and I’m excited you’re doing those global Code with Claude events, with Tokyo wrapping this run. What mantra do you lean on at work or in life?

At work I repeat keep it simple, and in life I try to choose kindness. During the lockdowns a direct report let me move a one on one so I could FaceTime my ailing grandmother, and that small grace meant everything.

In big meetings people sometimes hear you knitting; what is going on there?

Knitting pairs with programming in my head, purls and knits like zeros and ones, and after years of practice I can stitch without looking so it keeps my hands busy while I listen. My grandmother taught me when I was eight, so every project feels like time with her, and yes, the yarn stash is real.

If work ever took care of itself, I’d open a yarn shop named for my grandmother and make it a little community hub.

You and your team are doing remarkable work and the growth shows it; thanks for coming on.

I feel lucky every day to work with this team, and I’m grateful for the chance to share; thanks for having me.

Before we wrap, where can people find you and how can listeners help?

I’m on LinkedIn and would love honest feedback on what’s working, what’s not, and ideas for new use cases, like a friend who had an AI assistant draft a shed plan. If you can, walk someone new to AI through a workflow that could help them.

Thanks for listening. Subscribe on Apple Podcasts, Spotify, or wherever you get shows, leave a rating or review to help others find us, and explore past episodes at Lennyspodcast dot com.