↓ scroll · space · N for notes · P pause · R restart

Julie
Torrisi-Coffey

Design leader · Experiences, systems, AI across global platforms

Hello I’m Julie,

20 years in design, 11+ in product, building high-impact brand systems and scaling creative orgs, including almost five years at Meta leading design systems, end-to-end experiences and AI.

Today I’ll walk you through two projects where I set the visual direction and saw it through to ship: Facebook’s brand refresh in product, and the Comments redesign for 3 billion people. Both come down to hierarchy, density and attention: deciding what earns space on a busy screen.

I’ll also show you how I use AI to deliver design on spec and on brand, and why the taste call stays with people.

TLDRI’m the design voice at the leadership table, with the technical depth to back it up. I love making complex products simple and modern.

Julie laughing on a video call with a friend, both in colorful home studios.

What sets me apart

I’m a designer who covers the whole stack.

  1. 01BrandI trained formally in graphic design and grew up in brand and advertising. For the last 11+ years I’ve been in product design, delivering end to end.
  2. 02SystemsI build design systems for desktop, native apps and watches.
  3. 03Native prototypingAnd I build in native code, and I design the logic AI models use to keep improving the experience.

Most designers go deep in one or two of those. I do all three.

Talk track · voice-over only
  • 01 Brand. Graphic design degree, then nine years in brand and advertising at R/GA and agencies in New York and London. I say that; I don’t show it. The proof is the brand chapter: Facebook’s brand refresh, in product.
  • 02 Systems. “A system, not a style guide.” is the system. “What actually shipped.” is the reach: iOS, Android, Web, Mobile Web and FB Lite. “Build once, ship everywhere.” is how it held. Watches: at Tin Can I worked with the team that built new Android watch hardware, and it shipped. Told, no visual. My Xcode prototype project also targets watchOS.
  • 03 Native prototyping. The native chapter: “Design for the OS it ships on.”, “I use AI to prototype in the code it ships in.”, and “Same finish line.” Then the AI logic: “AI and product didn’t share a language. I built one.” and row 03 of Three fires, the creator framework.
  • Most designers live in one or two of these rings. I work in the middle.

What you get

Six superpowers.

01I get called when it’s on fireWhen something’s stuck and no one’s sure how to fix it, I get the call. I join the team, learn how it works, fix what’s broken and hand it back stronger.
02Influence without authorityI influence through the craft of what I deliver. I ask teams for input before I make a recommendation, so they adopt the design, and it makes their part of the product look better.
03Brand to system to native codeI set the brand, build the system and design straight to native code for iOS, Android and watches. I onboard AI like a brilliant new hire: house rules, the design system and a short brief. It takes the busywork. The design stays mine.
04I multiply every team’s workI build systems teams keep using after I leave: one bar for every milestone, decision logs, early prototypes, reusable artifacts and AI brought in upstream.
05I hold the bar under pressureWhen partners push for more on the screen, I protect what people came to do. Content stays the hero, and AI shows up when you ask for it, not before.
06I raise the whole design orgDesigners come to me to get better at presenting, and directors send theirs to me before they ship. I help them build the confidence that drives alignment fast. I post videos from design exhibitions I visit in New York to the design org channel, and invite everyone to share how design shows up in their life.

Leadership

How I lead.

Design leaders want me at the table for the decisions that build a stronger foundation for design.

01Low ego, outcome driven.Engineers trust my work as much as designers do. I hand iOS, Android and web engineers product changes in the code their platform already runs on, not just a spec.
02Hands on the pixelsI lead by example, with my hands on the pixels. I deliver the work, not just an opinion, with the design rationale and the technical reasons behind every call. Designers see the bar and raise their own.
03I treat partners like clientsI speak the language of every discipline, so each team gets the same clear direction in the terms it already uses. I treat partners like clients and advocate for them when they’re not in the room.
04Steady through turbulenceWhen leaders change, I keep the team moving. I set simple measures of success, like zero regressions, build trust, and coach designers to document their own impact so their work gets seen.
05I grow the bench.I grow designers who want to get better through hands-on mentorship. I show them, I don’t just tell them, and I share what I know often so others share too.
06Feedback in the momentI give clear feedback in real time. If a spec misses a detail, you hear it from me that day, not in a review. I treat my team’s time like executive time: weekly rhythms, strict agendas and room to do the work.

The machine needs rules

Chapter 1 · My AI workflow

Teach the machine the rules.

AI is a brilliant new hire who has never met your team. My first job is onboarding it.

My AI workflow

Every tool has one job.

Home baseClaudeOne project per body of work, with skills, a single source of truth, and scheduled tasks that run while I’m offline.
DesignFigmaConnected to AI so it reads and writes with real variables and components, never raw values.
My design kit: tokens in Figma, synced to code in GitHub.
BuildClaude Code and CursorClaude Code and Cursor for production-fidelity prototypes, built against the live design-system library.
SwiftUI in Xcode and Compose in Android Studio when it has to feel native, checked on device.
ConnectConnectorsConnectors to my calendar, mail and docs, so context comes to me.
Zapier for push alerts on emails I want to answer fast, and my daily top 3.
Quality checkEvaluationTest inputs that pass or fail, so a change is better, not just different.
A second agent that grades the first. A person approves.

Most used tools in my /stack

  1. Claude
  2. Claude Code
  3. Figma AI + MCP
  4. Cursor
  5. Zapier
  6. Xcode + SwiftUI
  7. GitHub
  8. Jetpack Compose
Talk track · voice-over only
  • The tools change every month. The setup doesn’t: one home, one system, one person who makes the call.

Production-fidelity prototyping

I use AI to prototype in the code it ships in.

On the live design system: SwiftUI, Compose and the browser.

01I write the spec first.The problem, every state, what done means.
02I branch the team’s repo.Claude Code or Cursor gets the spec, the team’s components and the design tokens.
03System-constrained AI.It can only assemble from the live library. No raw colors, sizes or fonts.
04I run it on a real phone.When something’s off, I fix the spec or the rule, then AI rebuilds it.
05I check it.Against the spec, state by state, and against the brand, value by value.

Drift isn’t an engineering problem. It’s a design gap.

My POVDesign for the OS it ships on.Craft is platform-specific. A sheet springs one way on iOS and another on Android. Back, scroll, type and haptics all follow the OS. When design hands over one generic version, engineers have to make those calls themselves. Every call they make up is a place the design drifts. So I design for each OS, and I prototype in production code, with the real components.
Talk track · voice-over only
  • Most teams prototype with AI from a blank canvas. I constrain it to the live library, so it can only build with what engineering already ships.
  • AI makes it easy to build something that looks right. I make it build something that is right: on spec, on brand, on the device.
  • Hand an iOS engineer a design that wasn’t made for iOS, and they’ll fill the gaps with their own small decisions. Each one is reasonable. Together, that’s drift.
  • My job is to make those decisions before anyone writes code, for the platform it ships on.
  • It doesn’t ship as-is. It ships as the spec.
  • Tools: Xcode (developer.apple.com/xcode) · Android Studio (developer.android.com/studio) · Cursor (cursor.com)

Design engineering

How does this impact craft?

I use AI to prototype in the code it ships in. Drift isn’t an engineering problem. It’s a design gap. When design hands over one generic version, engineers have to make those calls themselves. Every call they make up is a place the design drifts. So I design for each OS, and I prototype in production code, with the real components.

I block a day on the engineering team’s calendar, and we work back and forth, pushing to code and reviewing changes until it ships live. For big launches, I sit with the engineers in person.

01Real tokens and componentsBuilt on the design team’s components and tokens, so nothing gets rebuilt by eye.
02Every state, real contentTesting on real data to see real names, real comment lengths, errors to see where it breaks.
03Motion in numbersReal OS specific direction on duration, easing and spring values, never “make it smooth.”

AI drafts. I review. An engineer reviews before anything reaches production code.

Toolbox
  • FigJam
  • Figma
  • Dev Mode
  • Code Connect
  • SwiftUI
  • Compose
  • Cursor
  • Claude Code
  • Jira or Linear
  • GitHub
  • TestFlight or Android test track
  • Slack
Talk track · voice-over only
  • Craft counts when it shows up in the product, not in the file we polish for review. So we review the work live, in the code it ships in.
  • Reviewed live, until it ships: built on the team’s own components; every state, empty, loading, error, long text, no network, with accessibility notes.
  • That’s where you feel a spring that’s off, or the error state nobody drew. AI drafts the tickets and acceptance criteria from my spec; I review every one.

One brand, shipped

Chapter 2 · Case Study 1 · Meta, 2024–2025

Facebook's brand refresh, in product.

The mandate

The press release was the easy part.

In 2023, Facebook launched its first major brand evolution in over a decade. Announcing it was the easy part.

Inside the company, every product lead was handed the same mandate, bring this brand into the app. Every surface. Every team. Every visual.

Everyone hit the same wall. Where do we start?

Press release Meta Newsroom “Redefining Facebook’s Brand Identity”

What I didWrote a design brief and started researching the implementation of visual assets. The first thing I discovered, Facebook had no map of its own visual asset system. Before replacing a visual, an engineer had to find it.

Zoom out

Two decades of visual systems. Zero designed to talk to each other.

Early Utility, 2004
2004Early Utility
Papercraft, 2016
2016Papercraft
Cultural Moments, 2016
2016Cultural Moments
Alegria, 2017
2017Alegria
PVS, 2017
2017PVS
Illustration 2.0, 2018
2018Illustration 2.0
Campfire, 2020
2020Campfire
Photography, 2020
2020Photography
CX Territories, 2021
2021CX Territories
Flex, 2023
2023Flex

TypographyToo many styles, sizes too similar with loose hierarchy

IconographyGrey backgrounds, filled and outlined

Where I started: the account-restore screen with a one-off illustration, a long headline and three custom icons.

IllustrationIllustration looks dated a lot faster than iconography so we need to use it with care. No guidance on when or how big

ColorDull and flat

  • TypographyToo many styles, sizes too similar with loose hierarchy
  • IconographyGrey backgrounds, filled and outlined
  • IllustrationIllustration looks dated a lot faster than iconography so we need to use it with care. No guidance on when or how big
  • ColorDull and flat
Talk track · voice-over only
  • Here are all these visual styles, side by side.
  • When I started, I went through the product and kept finding different illustration styles. None of them really felt the same or consistent.
  • Historically, a redesign meant replacing every asset by hand, one engineer at a time. I could have done that again. Instead, I used the refresh to build the system underneath it, so future updates are easier for engineers, designers and everyone who touches an asset end to end.
  • What I did: I cataloged 20 years of illustration systems into one view, then built a system so the next refresh is a swap, not a hunt.
  • There was no real guidance: when do we use an image, what type, where it goes, what size.

My direction

No one agreed on what our assets should look like, or where they should live.

FLEX was the most recent attempt at the problem. It was an ambitious system, but when six assets were viewed together, they didn’t yet read as one brand, and leadership held off on approving it for rollout. Without an approved illustration system in place, execution remained blocked across surfaces.

Problems

BusinessOne illustration, four versions. Marketing, product, brand and engineering each owned a slice of how visuals shipped. No one owned the whole thing.
IllustrationFLEX drifted into inconsistency over time, producing assets that no longer read as one brand.
IconographyLoose guardrails meant icons varied across the product, with no shared standard for style or construction.
MotionFile creation and storage practices forced a bad tradeoff: assets were either too heavy, lagging load times, or too low-quality to ship. The styles weren’t always consistent.

How do we systematize illustration, iconography, and animation so they hold up over time?

Talk track · voice-over only
  • Mapped who owned what. I sat with each team and wrote down where their assets came from, who approved them and how they shipped.
  • Found the overlap. Teams were solving the same problem separately. I pulled them into one plan instead of four.
  • Unblocked the designers. Brilliant people, new to product scale. I ran the workshop to give them the constraints up front, so they could ship without guessing.
  • Moved us from a house of brands closer to a branded house. One Facebook, one visual language.
“We all own brand.”
Tom AlisonHead of Facebook

To get there

I put everyone who disagreed in one room.

Nobody agreed on where our assets should live or what they should look like, so I ran a 2-day sprint with Brand and Product. I picked speakers who disagreed, owned the deck, and had each partner make their case.

Then I ran blind polls: which asset would you use for this message?

The answers were all over the place. That was the proof we needed.

What I didI did the audit, ran the workshop, built the case, and got it funded.

Building the visual data layer

I ran the visual audit and built the data layer that made the problem visible.

Before proposing any framework, I audited every user-facing visual across Facebook myself. I sorted them into 11 categories, turning a messy image dump into a structured product schema.

What I did

Audited every visual by hand and turned an image dump into 11 categories.

Talk track · voice-over only
  • The 11 categories, top wins when two messages collide: 00 Content first (always first); 01 Errors & warnings (interrupts); 02 Trust, safety & policy (interrupts); 03 Deprecations & changes (interrupts); 04 Product & feature announcements (can wait); 05 Onboarding reminders (can wait); 06 Incentives & promotions (can wait); 07 Best practices (can wait); 08 Education & guidance (can wait); 09 Status & utility (quiet); 10 Activity (quiet); 11 Milestones & recognition (quiet).

The pilot

First, ask if the moment needs a visual at all.

Empty states were the first full test of the rules. The audit had found 160 different illustrations, from different teams and years, all saying "there’s nothing here yet."

160 variants → 10 solutions, covering 88% of empty-state use cases
Empty states are 1 of 11 categories, and the pilot for the whole system.

What I didRan the pilot and measured it against the audit, variant by variant.

Talk track · voice-over only
  • Pre-workshop survey, a Facebook product designer: ““There's no guidance for when to use an asset for what. Everyone makes it up.””
  • On every page, I ask how much information the user needs to make a decision or take action, and I cut everything else.
  • Grey boxes and icons were all visual assets we had to replace or update.
  • Wrote the decision rules: unship, replace, or intake.

The case

The audit and the workshop
became the business case.

I wrote the business case, the strategy and the proposal. Every part came from what the audit found and the constraints the workshop agreed on together.

Approved June 2025Strategy briefEndorsed by the Head of Facebook. Funded to bring the brand refresh into the app.

What I didTurned the evidence into a plan leadership approved and funded.

What shipped

Bringing the brand into the app.

We had too many sizes, colors and styles of icons, illustrations and motion, all inconsistent across the product. So I started where people look every day.

  1. Sorted visuals by communication category, then started with the highest-traffic surfaces
  2. Cut aspect ratios and sizes down to one consistent set
  3. Removed assets where content should come first
  4. Gave each asset type rules for when to use it, so teams know what to find and what replaces it

TypographyToo many styles, sizes too similar with loose hierarchy

IconographyGrey backgrounds, filled and outlined

Where I started: the account-restore screen with a one-off illustration, a long headline and three custom icons.

IllustrationIllustration looks dated a lot faster than iconography so we need to use it with care. No guidance on when or how big

ColorDull and flat

  • TypographyToo many styles, sizes too similar with loose hierarchy
  • IconographyGrey backgrounds, filled and outlined
  • IllustrationIllustration looks dated a lot faster than iconography so we need to use it with care. No guidance on when or how big
  • ColorDull and flat

What I didStarted where traffic was highest, cut the variables, and gave every asset type a rule.

End-to-end flow · the rules

Anyone can see what needs to change.

After. A system badge replaces the one-off art. Containers come in three sizes: 24, 16, 12.

Fewer choices, so every discipline can build it right without a review.

TypographyClear hierarchy

Iconographyone style, used everywhere.

After: one system badge, filled or outlined icons, container sizes 24 · 16 · 12.

Illustrationrules for size, type, and when to skip it.

Colorbrighter, with fewer colors, so the eye knows where to go next.

  • TypographyClear hierarchy
  • Iconographyone style, used everywhere.
  • Illustrationrules for size, type, and when to skip it.
  • Colorbrighter, with fewer colors, so the eye knows where to go next.

End-to-end flow · the rules

Anyone can see what needs to change.

After. A system badge replaces the one-off art. Containers come in three sizes: 24, 16, 12.

Fewer choices, so every discipline can build it right without a review.

TypographyClear hierarchy

Iconographyone style, used everywhere.

After: one system badge, filled or outlined icons, container sizes 24 · 16 · 12.

Illustrationrules for size, type, and when to skip it.

Colorbrighter, with fewer colors, so the eye knows where to go next.

  • TypographyClear hierarchy
  • Iconographyone style, used everywhere.
  • Illustrationrules for size, type, and when to skip it.
  • Colorbrighter, with fewer colors, so the eye knows where to go next.

What I didMade the rules visible in the UI, so anyone at Facebook could look at a screen and know exactly what to update, without me in the room.

In production

A system, not a style guide.

Visual debt is findable for the first time. When design works, you almost don’t notice it. People’s content should shine, and everything around it should get out of the way.

BeforeFinding an asset meant asking the engineering team that owned the surface to dig it up. Per asset, per surface, per team.
AfterAssets are tokens in the design system on iOS, Android, Web, Mobile Web and FB Lite, named for what they’re used for. Find it, replace it, swap it everywhere at once.

A one-time refresh became a process anyone can repeat. The next brand refresh shouldn’t need another audit.

What I didTurned visual assets into tokens teams can find, replace and swap at scale.

Adoption

Adopted, not mandated.

Teams don’t resist a system they helped build.

I set a high bar for craft. When the work makes their product look better, teams adopt it without a mandate. Design directors sent their designers to me for review before they shipped.

Once the pilot proved it, I handed ownership to the Facebook Foundations team to roll it out across all 11 categories.

What I didScoped it, built it, proved it, then handed it off so it runs without me.

The outcome

Six random assets. Read as One brand.

I wrote the business case, the strategy and the proposal. Every part came from what the audit found and the constraints the workshop agreed on together.

$1.2Msaved every year
78%faster from design to engineering
300+designers adopted it
60%fewer aspect ratios

*$1.2M: the yearly cost of making one-off assets instead of reusing existing ones, counting everyone involved end to end.

Talk track · voice-over only
  • “Selecting 6 random visual assets should reveal a cohesive look, as if they’re from the same brand.” A design director, Design Shop.
  • I ran the empty-states pilot end to end, then handed the 11-category rollout to Foundations.

One comment experience

Chapter 3 · Case Study 2 · Meta, 2025–2026

Modernizing Facebook comments.

As design lead for Comments, I led a seven-month modernization of public conversation for 3 billion people. 9 partner teams. Ship approval from Tom Alison, Head of Facebook.

The opportunity

Facebook was losing the next generation.

By 2025, Facebook’s comment participation rate had declined 12.5% year-over-year.

Instagram, TikTok and YouTube all trending the other way, and their conversations were where young adults went.

Risk

Doing nothing, and losing young adults a little more every year.

Instagram’s comment sheet.
YouTube’s comment sheet.
TikTok’s comment sheet.

The directive

“The behavior we should try to reinforce for creators is whatever leads to more authentic interactions.”
Mark ZuckerbergCEO of Meta

The CEO's directive was clear, build a thriving ecosystem for authentic creator conversations.

The research

“Instagram is mean but Facebook is worse, like I need a restraining order bad!”
Real Facebook user2025 research

What I didTook over comments in late 2024. While on the Creators team we cared a lot about the quality of public conversations.

Context

This wasn’t a product problem. It was a system problem.

Comments read as toxic and the UI was dated. No one owned comments, so every team changed the comment sheet its own way, and it worked differently depending on where you were in the app. Every new feature broke two old ones.

01Legacy UI debtPatched so many times that every new feature broke two old ones.
02Toxic signalRanking surfaced negative and spam content at the top. People felt unsafe posting in public.
0330+ duplicate effortsTeams solving the same problem in isolation, shipping conflicting features into one surface.
04Organizational churnRe-orgs and manager rotations. Every quarter brought a new plan.

What I didShowed leaders that no one owned comments. Reframed 30+ comment workstreams into 10, then wrote the strategy brief that set one direction.

Pillar alignment

Find what nine teams already agree on.

Nine teams. One shared foundation they hadn’t seen yet. Any feature built for safety, clarity or control was a bet every team could stand behind, and no leader could veto it. Anything else got deferred or killed.

Consume, react, write came first; everything else moved to a long press.

What I didRan the comment sprint that got nine teams to agree on three jobs: safety, clarity, control.

My leadership style

I lead through influence, not org charts.

I wrote the business case, the strategy and the proposal. Every part came from what the audit found and the constraints the workshop agreed on together.

9partner teams aligned
15designers directed
5manager changes in a year
Execfinal ship approval
Quarterly digestevery test and ship decision in one update, so new leads had full context on day one.
Weekly XFN syncteams running public tests scheduled around each other, so signal stayed clean.
One results feedevery result posted live, before it could become a surprise.

What I didFounded and ran the rhythm nine teams built on. No mandate. Teams came because it was the one place comment decisions got made and stuck. Directors started sending their designers to me for sign-off.

Talk track · voice-over only
  • I had five managers in a year. The plan was solid enough that none of them had to worry about this work. I kept the work moving, and it shipped earlier than expected.
  • Nine teams were running tests in comments at the same time. Overlapping tests muddy the signal, so every week I lined up what was live where, what each test was trying to learn, and what the results meant. The quarterly digest let a new lead walk in with full context on day one.
  • Engineers started bringing me broken issues outside my area, and I worked with our Engineering Director to get the right people on the parts of the product costing us revenue.
  • People kept assuming I was on the Foundation team. That’s the biggest compliment, because it’s the bar I hold for craft, whatever team I’m on.

Failing fast

I didn’t show leaders options. I showed them the survivors.

Every comment pattern was pressure-tested against the three jobs and the product constraints, and cut before leadership ever saw it. Cut patterns never reached a live test: they broke the design system and didn't scale.

What I didExplored and pressure-tested the patterns myself. Leads only saw survivors, because a failed pattern costs a test slot. Every test that ran was worth running.

  • Failed
    Safety test
    Clarity test
    Control test
    CUT
  • Passed all three jobsMove to test
Failed patterns

Presentation + strategy

Each milestone earned the next.

Four milestones, each gated on one bar, hold metrics flat, no lift required.

That bar is what unlocked structural change, the team didn't have to chase a win on a redesign.

Every milestone ran on the same thread, so leads reviewed the design, not the content.

People run to the comments. We put them in a bottom sheet so they could immerse themselves and join the conversation without leaving the post. One swipe and you're back on the post in Feed.

Legacy experience
Legacy comment sheet: grey bubbles, the old byline and reaction row.
  • Users expect the best comments first. Ranking didn't show them
  • One screen showed two comments
  • Two places to scan meta info, top and bottom
  • Everywhere else, Like is an icon. Here it's a word, on the left
  • A thread line at the top level, with nothing to connect
  • Every comment sat in a box, taking up height
  • Every reaction shown, every time. Noise instead of signal
  • Bubbles, badges, counts and icons all competing
  • Creators couldn't pin comments and Audiences couldn't see their replies
  • Images ran the full width of the thread
Milestone 1
M1, the live Android test: bubbles removed, one byline.
  • indent replies
  • no thread L0
  • Explore Creator badge “author”
  • removed bubbles
  • reduced image width
Milestone 2
M2, refine and isolate: reactions and replies tightened.
  • meta info on one line
  • replies color change no indent
  • Explore Creator badge with more visibility
  • increase comment text size
  • removed bubbles
Milestone 3
Ship candidate
M3, the ship candidate: thumbs up and down on every row.
  • replies pattern exploration
  • reduced to one dominant reaction with a number
  • Explore Creator reaction
  • move reactions to right rail, added thumbs down to enhance ranking
North star
recommendation
North star: filter tabs, pinned comment and the full system.
  • pin comment
  • View reply pattern
  • Creator or author reply pattern
  • filters for faster replies to increase engagement
  • move reactions to right rail, added thumbs down to enhance ranking
  • Small Creator reaction, on the reaction line
  • simplified composer

What I didSet the single go/no-go bar and ran every gate against the same evidence, so leadership could decide with confidence.

Zero relitigated decisions

Same content. Only the design changes.

Comments has a lot of moving parts. Every milestone runs on the same thread, so leads review the design, not the content.

Launch criteria were agreed up front, so rotating leads walked in with full context and decisions stayed made. The path to ship was visible from week one.

  • Legacy comment sheet, same comments as every other milestone.

    Legacy product

  • Milestone 1 comment sheet, same comments as every other milestone.

    Milestone 1

  • Milestone 2 comment sheet, same comments as every other milestone.

    Milestone 2

  • Milestone 3 comment sheet, same comments as every other milestone.

    Milestone 3

  • North star comment sheet, same comments as every other milestone.

    Northstar

Holding the bar

The VP of Engineering hesitated. I came back with data, not pressure.

Early metrics dipped, as they do after any big visual change. I brought history: a past video-player redesign launched flat, then recovered. Short-term dips are recoverable. Long-term participation loss from a degraded experience is not. Leads held working prototypes in their hands. We shipped.

What I didBuilt the data case myself and kept the launch bar where it was.

“Metrics for our current comment experience are strong.”
Lars BackstromVP of Engineering at Facebook
“I've been using the new design all summer. It's much better than the previous experience.”
Tom AlisonHead of Facebook
Talk track · voice-over only
  • We did qualitative and quantitative research with young adults. A lot of them talk to text, and we didn’t support it. They reply with memes they’ve screenshotted, kept in a curated folder on their phone they’re proud of, and they wanted a recents option to reuse a favorite.
  • Phase 1 of the composer shipped as a starting point. The Feed team didn’t want to build on the old composer’s infrastructure, so I gave them the evidence to build new infrastructure instead, on a team I had no authority over.

The roadmap, as it actually happened

Read the data, killed the redundant test, shipped.

01 Inherited ambiguity: nobody agreed what comments were for. I wrote the strategy brief and ran a comment sprint first.

02 My director lead left in the first weeks. I set scope across nine teams anyway, so the plan didn’t depend on one person. Then my product design partner left and the org re-shuffled. M1 went live on Android and validated the direction.

03 Android already told us what an iOS M2 test would. I made the call to skip it, briefed the leads, and pointed the team at ship readiness. Head of Facebook approved the ship.

04 Cross-platform build to ship, through another re-org. No re-opened decisions. No regression at launch.

05 The design system rollout (M4) carried the new components to every surface. The H1 ’26 roadmap, and the Xcode prototype platform so any designer could build on live data.

What I didOwned the plan end to end: set scope across nine teams, made the call to skip the iOS M2 test, and took the ship to the Head of Facebook for approval. Designed every comment component in the library and delivered redlines and specs to engineering for every platform. Caught broken type tokens that were rendering the wrong font.*M1–M4: milestones 1 to 4. H1: the first half of the year.

Stabilizing while leading the program

Four fires. Four ships. Four clean handoffs.

I was in the room with Tom Alison, giving clear input so leadership could make informed decisions.

Asked for byThe problemWhat I didThe input I gaveResult
01Creator MessengerTom Alison, Head of Facebook
Creators couldn’t message followers on Facebook.Shipped full inbound and outbound messaging on iOS and Android under a hard deadline.Aligned legal, safety, Messenger and product on one scope, so the deadline held.H1 goal met.
02Facebook ProfilesChris Cox, Chief Product Officer
The profile infrastructure didn’t work for public creators.Built trust with the owning team and moved in weeks instead of quarters.Used executive alignment to clear the approval blockers.Quick fixes shipped. Experience stabilized. Handed off.
03Creator FrameworkWendy Owens, VP, Head of Design
The creator journey framework had no logic for AI to use.Added entry and exit criteria to each stage and connected the framework to how the AI learns.Introduced the AI engineering team to the framework directly.Framework adopted. Cross-team connection made. Handed off.
04Android CommentsMark Zuckerberg, CEO
Comment text on Android was the wrong size, and the redesign wasn’t built to spec.Flew to Menlo Park and sat with engineering to build the full redesign to spec, down to comment threading.Found the broken comment text-style token. One fix cleared many downstream issues.Text size fixed. Redesign built to spec. Handed off.

What I didPulled in by four executives while leading Comments: diagnosed each problem, gave clear input, shipped, and handed off.

Talk track · voice-over only
  • Messenger: creators were using Instagram instead.
  • Profiles: the team that owned the profile infrastructure didn’t want to change it.
  • Framework: two teams were solving related problems without talking to each other.
  • Android text size: I was already fixing it when Mark flagged it. Sitting with engineering, I found the broken token, and fixing it cleaned up a lot downstream. That’s the attention to detail you get working with me.
  • No one team owned comments. I didn’t need to own it to move it forward.

Before and after

Fewer boxes. Clearer conversation.

Leadership didn’t have to take the redesign on taste. Every change traces to a job it serves: safety, clarity or control. It set the standard past comments: Feed followed the new comment action row when they simplified the action row under every post.

What I didDesigned every one of these changes and wrote the rationale leaders used to approve them.

  • FiltersSort the thread by who's talking: followers, top fans, stars.
  • PinnedThe creator's pick sits on top, marked with a pin.
  • Every rowNo bubbles. Reply, reactions, up and down on one line.
Legacy comment sheet: grey bubbles, Top Fan label, byline and reaction row on every comment.
New comment sheet: filter tabs, a pinned comment, no bubbles, up and down votes on every row.
BeforeLegacy sheet AfterNorth star

Outcome

What actually shipped.

iOS · Android · Mobile Web · Web · Facebook Lite. One system, every surface.

0engagement regressions at ship
+20%daily comment volume, 367.6M → 441.4M daily average
+17%comments published to Facebook lifted the whole ecosystem
73Mdaily views on Pinned Comments at launch
10.79%increase in positive comments
42%drop in toxicity
24.18%decrease in negative comments

Features I designed end to end · all shipped

Pinned Comments
Comment composer
Sharing via DM
Anonymous commenting
The Blueprint components cover card over the Comment component sheet: comment, reply and thread variants on default, on media and on color surfaces.

What I didDesigned the component library and every permutation, and drove the cross-platform launch to iOS, Android, Mobile Web, Web and FB Lite.

Proof of prototyping in native code · a precedent

Any discipline. Live data. No engineer required.

In December 2025 the first version of Facebook’s master prototype shipped across all of Facebook design; by March the fifth version carried the final comment update. Real feeds, real profiles, live AI models. One line of code to start building. Instagram adopted the pattern.

What I didDirected two designers: one built the comment components into the final library, one owned documentation and guardrails. Reviewed their work weekly, kept a standing touchpoint with the director, and recognized all designers and engineers publicly at ship.

Talk track · voice-over only
  • At launch, I recognized every designer and engineer publicly.
  • I hold a high bar for what ships: beautiful, functional and good for the business. All three, or it’s not a win.

Live today

Still live in production.

The modern comment composer, live for everyone. Shown here from a live recording: the keyboard, then the sticker tray.

Meta’s December 2025 newsroom post announced the simplified commenting experience, new controls for creators and group admins, and anonymous flagging for everyone.

North star

One keyboard, everywhere.

My north star recommendation: bring Messenger’s composer into Facebook, so every Facebook user learns one way to use the keyboard. Shown here from my Messenger keyboard prototype.

I don’t just build a prototype. I build it so it actually records your voice and every feature works, so I can see where it breaks and refine it before it goes to an engineer.

Live recording of the shipped comment composer: the keyboard, then the sticker tray.
Composer live today
Recording of my Messenger keyboard prototype in Facebook comments: the Messenger composer, the plus menu and the photo tray.
North star · Messenger keyboard

What I didMade the case that an 8-year-old composer was the real risk, got it funded, designed and delivered Phase 1, then handed it to a designer on the Feed team and stayed in every review through ship.Made the case that an 8-year-old composer was the real risk, got it funded, designed and delivered Phase 1, then handed it to a designer on the Feed team and stayed in every review through ship. Recommended parity with Messenger to the head of Facebook, it was approved but I left the company before it was implemented.

Precedent

Designed for comments. Adopted by Feed.

Feed took its icon modernization from the comment redesign and cut the action row under every post from as many as 4 lines to 2.

What I didDesigned the comment action row and its icons. When Feed simplified the action row under every post, they built on my design.

A Facebook Feed post with the redesigned action row: Like, Comment, Send and Share on one line, with the comment redesign's icon style.
  • FeedTook its icon style from the comments redesign.
Feed UFI redesign
Talk track · voice-over only
  • Feed went from as many as 4 lines to 2 under every post. Product constraints ruled out 1, so never say Feed went to one line.

Result

The goal was never to just redesign comments. The goal was to make great conversations the default, forever.

Three things I do consistently well that make great work inevitable:

  1. Turn fragmented work into one system that ships.Scattered teams become partners working toward the same goal.
  2. Build the operating rhythm that protects the work from organizational chaos.Decisions stay made, even when the org changes around them.
  3. Make products other teams want to build on.Adoption is a choice, not a mandate.

Thank you.

julietorrisi.com · linkedin/julietorrisi

“Julie was key to executing the Comments Modernization and initiated the Comments Composer Modernization. Her behaviors led to faster, more polished product launches.”
Engineering Director · CreatorsOn the comments modernization
“You've pulled into a ton of fragmented work and aligned several disjointed PMs.”
Partner · CommentsOn the comments strategy, 2025
“Thanks for leading an incredible comments workshop… driving this collaborative effort across FB teams towards a northstar.”
Partner · CommentsOn the north star workshop, 2025
“It's literally bringing clarity to a question I've asked myself for years… the conductor driving this train forward.”
Design manager · BrandOn the Visual Asset Framework, 2024
“Your ability to navigate through ambiguity and drive meaningful conversations with both internal teams and XFN partners has been instrumental.”
Design manager · IconographyOn Visual Asset Framework 2.0, 2024
“The nomenclature, definitions and overall work will prove to help so many future conversations and allow faster alignment on all workstreams.”
Design director · Design ShopOn the Visual Asset Framework, 2024
“Your efforts and time saved our team a ton of work and unblocked us so we could pursue explorations in a timely manner.”
Animator · Design ShopOn the surface audits, 2024
“Frank conversations… resulted in us having a quick alignment and shared POV.”
Partner · BrandOn the brand debt strategy, 2024
“I appreciate your expertise and willingness to jump into something with lots of moving pieces.”
Partner · FacebookOn the loading screen project, 2024
“Thanks for treating me like a product designer. The past few weeks… has given me a huge confidence boost.”
Visual designer · Design ShopOn working together, 2024
“Being an inclusive partner… Our team felt seen and heard.”
Design director · Design ShopOn the glossary work, 2024
“Encouragement creates a safe space.”
Crit pod member · Design ShopOn the crit pod I led
“As a novice, I learned so so much, this is really going to help speed up my process.”
Illustrator · Design ShopOn my Figma 101 workshop, 2024
“Thanks for sharing the secrets of how designers have such polished presentations on zoom calls.”
Partner · CommentsOn presenting, 2025