The winners this week weren't defined by what they added. They were defined by what they refused to do. Diya Jolly decides what not to decide. Crystal Ammari kills ideas before a single engineer is hired. Vonny Laing designs for the users everyone else skips. Walt Disney refused to oversaturate the thing that made him money. e.l.f. won by not competing where it couldn't. In a world where adding anything is nearly free, the scarce discipline is the "no". This is a masterclass in the art of not.

💡 Top tip for this week's roundup - read the TL;DR, then jump to the section closest to your current problem. "Decide what not to decide" and "test it before you build it" are the load-bearing ones.

Remember, we've built an ever-growing library of our top podcast summaries. Check it out here.

Featured in this round-up:

🕒 Estimated reading time: 10 mins. Time saved: 16+ hours! 🔥

Not your topic this week? Try these five instead:

———————————————————————————

Start with the person whose whole job turns out to be subtraction. Diya Jolly, CPO and CTO of Xero (ex-Okta, ex-YouTube, ex-Google), argues that most product orgs ship mediocre products for one reason: their leaders think the job is deciding the roadmap. It isn't. The job is three things - "setting direction", "understanding customers deeply" enough to choose which bets to make ("cuz you can't make every bet"), and re-cutting resource allocation "every single quarter."

Everything good flows from what you refuse to touch. Her delegation model is three buckets: reversible-and-small goes down; big-but-certain goes down with a trusted lieutenant watching; only big-and-ambiguous stays with you. And she is militant about protecting that narrow remit - one meeting-free day a week, two hours a day, a browser she can't open - because "the senior job is few big irreversible calls, not more email." The discipline isn't doing more. It's fiercely guarding what you don't do.

The one place she says to add risk, not subtract it: the AI-era roadmap. The old rules of thumb have broken, so "you now have to take more roadmap risk, not less" - building what customers can't yet articulate. When Xero shipped automatic bank reconciliation, customers resisted hard: "No, no, no, no. You can't do this with a machine." The team built it anyway. It hit 97% accuracy and saved small businesses around 22 hours a month.

Key takeaways:

Adam Mosseri, Head of Instagram, describes a shift that is subtraction at the level of the org chart. Instagram's canonical squad - a dozen specialist engineers plus a PM, designer, data scientist and researcher - is giving way to pods of four to six generalists. The new keystone role is "product staff": a PM who now does slices of design, data science and research, pulling in a senior specialist only when the work genuinely demands one.

His claim about why is the interesting part - it isn't only AI. "Smaller teams simply decide faster." Fewer people to coordinate means less design-by-committee and quicker, often better, calls. Functions aren't vanishing, they're being pared back and reshaped: the strongest designers and researchers convert into product staff; the roles Mosseri says he'll deliberately hire fewer of are large-org leadership and mid-level specialist ICs.

And when you strip a team down, what's left has to carry more weight. The best leaders, he argues, are curators - they don't need every idea to be theirs, they build environments where strong ideas surface and mesh. Grit, quick learning and self-awareness stay the baseline; curiosity and a willingness to look stupid become the premium. The move is removal, not accumulation.

Key takeaways:

Sam Ladner - Senior Principal Researcher at Workday and a corporate ethnographer - draws the sharpest line of the week around what not to automate. Offload the labour, she says, not the sense-making: "AI is a gift for transcription and boring coding, but the moment you hand it cognitive sense-making you have broken the research."

Her reframe is the quote of the week: "Qualitative research is not correlation hunting. That's not what it is, and it never has been. It's closer to unriddling than it is to predicting." Machines "don't think, and they certainly don't make intuitive leaps." And the thing you can't subtract is the outlier: "Your explanation as to how this entire corpus of data came to be must explain the outlier. And if you can't do that, you're not doing quality research."

She saves the hardest "don't" for AI moderation and synthetic users: automating the human-to-human contact "is telling human beings, you don't give a shit about them... you still haven't done the work. And when I say the work, I mean caring." Sometimes the honest move is to subtract the pretence and just run a survey.

Key takeaways:

Katie Burke, COO of Harvey and a long-time HubSpot people leader, treats her job as knowing which complaints to ignore. Her most useful tool is a filter: 20% of your people will always be unhappy, so learn to tell "protein" complaints - the ones that actually block winning - from "sugar" ones, and let the sugar go. The people who need watching aren't the loudest single grievances but the "frequent flyers" who orbit every minor one, "because they recruit others to the noise."

The role runs on what you withhold as much as what you share. A CPO passes on thematic feedback, "never who said what", because "one breach of confidence and you never get the honest conversation again." And titles are a subtraction test in disguise: candidates are on best behaviour in interviews, so "if someone fights hard over a title now, expect them to be difficult later" - a signal to not hire.

Where she refuses to cut is speed of response to AI: if the models improve 60%, "your operations have to keep pace", and people leaders should lead the "people vs agents" conversation rather than react to it. The posture underneath: demanding and kind together. "You can be liked or respected; pick respected."

Key takeaways:

Vonny Laing, UX Lead at the UK's Student Loans Company, offers the most concrete "don't" of the week: don't design for the average user. Invert the pyramid - build for underserved and edge-case users first, "and everyone below them inherits the clarity", the halo effect that curb cuts and captions proved decades ago. Public services, she argues, can't only design for what people want; they have to design for what people will need in two, five, ten years, because wants and needs are not the same thing.

Her sharpest prompt refuses the happy path entirely: not "what's the happy path", but "what is the worst thing that could happen to a user in, or because of, this service?" And she refuses the expensive-research excuse - "your users are already screaming what's broken", in chat logs, complaint logs and call listening. A trestle table, a box of cheap chocolates and a train fare buys insight "no repository or licence can."

On synthetic users she is blunt, and it's a subtraction warning: LLMs give "a facsimile of a person and the warm feeling you've done research", but real lived experience "isn't in the data to begin with." Don't mistake the feeling of research for the thing itself.

Key takeaways:

Crystal Ammari, a product optimisation leader who's run experimentation at Nike and Disney, tackles the purest subtraction discipline there is: not building the thing until you know it's worth building. Her four-question workshop - what do we want, what do we know, what can we do about it, where should we start - "turns a blank page into a scored backlog" in 90 minutes, ranking every idea against effort, priority and traffic so the weak ones fall away before anyone commits.

The technique that lands hardest is "dry testing" - testing a feature before it exists. She recalls a fake button at Disney: "there was somewhere around 4 million-ish people and only 106 people clicked... I will never forget that 106 because it was just a huge, stark, obvious evidence that this is not something people wanted." Nobody was hired. Nothing was built. The idea was killed for the price of a button.

Her reframe of outcomes turns not-building into a win: "not calling them wins and losses, but savings and gains. So if a test won, you gained something... but if a test lost, really, we saved ourselves." And the real blocker to testing was never appetite - it was a "glorified waterfall" annual plan with no room for it. Her unlock: "just get something live."

Key takeaways:

Evan Spiegel, CEO of Snap, has spent 15 years refusing to defend the wrong thing: "software is not a moat." Features get cloned overnight - which is what AI is now teaching everyone - so the mistake is pouring energy into defending them. Don't. Defend the hard-to-copy stuff instead: distribution, which "is the scarce resource" in the AI era, plus creator ecosystems, an AR developer platform, and a vertically integrated hardware stack.

His counterintuitive design principle is also a subtraction: don't connect people to all their friends, connect them to their best friend, "because most of the value sits in the closest relationships." Snapchat beat network effects by narrowing, not widening.

Even his innovation model is about restraint. "Loonshots" pairs a tiny flat design team running fast with a large operational org, and the leader's job is the relationship between them. Design sign-off is a deliberate bottleneck - nothing ships without it - precisely to stop incoherent additions reaching the customer. "If you want to have a good idea, you have to have lots of ideas", then cut ruthlessly to the ones that cohere.

Key takeaways:

Chad Peets and former Snowflake CRO Chris Degnan bring subtraction to hiring, and their central point is a "don't": don't hire on the signal everyone trusts. "Someone from Salesforce or ServiceNow has probably never opened a new logo" - they rode an inevitable product. Skip them. The rep you actually want is the one "who won with an inferior product at a company no one has heard of."

Their craft is about removing the wrong risk. Quota-setting is risk-setting: "set quotas too low and you overpay; set them too high and your A-players quit." Booked contracts beat monthly on-demand revenue, because "no contract means no moat, and a competitor can flip your customer overnight." And on performance, the subtraction is unsentimental: "when there is doubt, there is no doubt" - a healthy sales org attrits its bottom 10% a year.

The framing device is the "$100M CRO bubble" - frontier AI labs paying sums the market has never seen, which both guests call unsustainable. The trap it creates is exactly the one they warn against: comp inflation makes the big-logo, big-number signals look even more attractive, right when you should trust them least. Hire for what someone actually did, and cut the rest.

Key takeaways:

Joey Shamah, co-founder of e.l.f. Cosmetics, tells a founder story that is really about refusing the obvious fight. e.l.f. sold cosmetics for a dollar - but the discipline wasn't racing to the bottom on price. "We weren't the first person to sell lipstick for a dollar. We weren't the cheapest guy selling lipstick for a dollar. We were just creating a brand around it. We were creating an emotional affinity." A $6 product cost about the same to make as their $1 one; e.l.f. simply refused to pay for the marketing, spokespeople and shelf space that inflated the price.

The discipline was validated the hard way, through a wall of "no". Dollar General, Family Dollar, Target and others all rejected e.l.f., convinced a cheap product would cannibalise their margins. Then a buyer at HEB tried it and emailed the morning after: "I was wrong. This brand is incremental and impulsive." Cheap prices made customers buy more, not switch - and that single data point rewrote the pitch.

The line that let e.l.f. grow up came from a Target buyer: "Price is what you pay and value is what you get." It let Shamah move from $1 to $3 to $15 "without betraying customers", because the promise was always value, never a price war it would eventually lose. (A false rumour that Bloomingdale's was buying e.l.f. and raising prices didn't hurt - orders jumped from 300 a week to 18,000 a day.)

Key takeaways:

For the deepest version of the pattern, Acquired went back ninety years. In Ben Gilbert and David Rosenthal's telling, Walt Disney's empire is a study in what to withhold. He "lost his entire company to a contract loophole once", and "every durable thing Disney built traces back to that lesson": own the IP, and don't give away the thing that compounds.

The discipline that made the flywheel work was refusing to flood the core. Scarcity and quality in the primary medium - the films - while covering the earth everywhere else with merchandise and licensing (which "out-earned film rentals within a year"). "Marvel's fatigue is what breaking this rule looks like." And the vault was subtraction as strategy: re-release every seven years, capture each new generation, "never dilute."

The most humbling detail for anyone chasing a quick add: "99.95% of Disney's value was created after Walt died." He didn't try to monetise everything at once. He obsessed over the core product, protected its scarcity, and let the durable business emerge - an asset "with more runway than he had time."

Key takeaways:

———————————————————————————

Ten conversations, one instruction that cuts against every "add more" reflex in tech: the skill is subtraction. When adding anything is nearly free, the scarce discipline is the deliberate, defended "no".

Concretely, a few refusals worth borrowing this week. Audit your calendar against Jolly's test and hand back the reversible decisions that aren't yours. Draw Ladner's line - automate the labour, never the sense-making. Steal Ammari's fake button and kill your riskiest idea before you staff it, then call the kill a "saving". Apply Burke's filter and stop chasing the unhappy 20%. Skip the big-logo hire, per Peets and Degnan, for the one who won with a worse product. And borrow Disney's discipline: protect the core by refusing to flood it.

The thread that ties it together, from Jolly's three buckets to Disney's vault: everyone can add now, so addition stopped being the edge. The builders pulling ahead are the ones with the nerve to leave things out - the decision undecided, the feature unbuilt, the complaint unchased, the price war unfought. Not the most, but the right, defended by an unfashionable "no".

Want more of the same? Try these five:

That’s a wrap.

As always, the journey doesn't end here!

Please share and let us know what you liked or want changing! 🚀👋

Alastair 🍽️.