蓝色循环、特征问题与黄金仪式：Coda 创始人的增长与团队手册

Mehrotra says every business should draw a diagram of how its ecosystem works. Coda's came from Matt Hudson, who runs data and finance, about three or four years into the company. In the Black Loop, someone makes a doc and shares it with a group, and some of the recipients make their own docs, so the cycle repeats. Mehrotra calls it the Microsoft Loop because that is how documents naturally spread.

That loop shaped Coda's pricing. Document products usually charge both editors and makers, but Coda charges only when someone creates a doc. Mehrotra explains that he wanted no friction on the share edge and "no dollar signs" there. He compares it to YouTube, where nobody would share anything if they had to pay for each viewer.

In the Blue Loop, a maker publishes a doc to the world, much like building a website: they pick a URL and decide whether Google can index it or whether it appears in the gallery. He calls it the YouTube Loop because publishing a doc feels like publishing a video. Almost a third of users come through it, often without hearing of Coda first. They arrive looking for a way to run an offsite or win at Ticket to Ride, and many people start with a problem, not a blank page.

Mehrotra says that thinking in funnels is natural but misleading, and almost every product has some kind of loop, whether through built-in sharing or advocacy. In Coda's case, the Black Loop is table stakes for the category, while the Blue Loop is its own invention. A user described the Coda Gallery as halfway between Medium and an app store. His advice for finding your own loop is to look at what you told the last few candidates you pitched.

Users arrive three ways. Someone shared a doc with them, which gives the best activation and retention. Someone found a published doc or template, which is second best. Or they started from the top with a blank cursor, where roughly one in five reach the activation moment.

Each entry point needs a different mindset because in the first two the conversation about Coda is owned by someone else, such as a teammate or a publisher. Mehrotra contrasts Steve Jobs's control-everything view with platform thinking, where your connection to the end user runs through other people. At YouTube, that meant being surprised by things you could not control, and he compares it to Airbnb learning to let hosts market their own places.

Mehrotra says he expected his book *Rituals of Great Teams* to take six months, and he is now nearly two years in and about half done. It started when he mentioned a Bing Gordon line on Reid Hoffman's *Masters of Scale*: great companies have a small list of golden rituals, which are named, known to every employee by their first Friday, and templated. Examples include Amazon's six-pagers, Google's OKRs and Salesforce's V2MOM.

Coda's own golden ritual is Dory and Pulse. In Pulse, everyone writes their opinion and the others stay hidden until all are done, which forces people to be clear and unbiased. In Dory, people put questions on the table and the group up- and down-votes them to decide what to discuss. New hires tell Mehrotra that a question they asked outvoted the CEO's, and that it felt open.

After the podcast he began hosting dinners for people to swap rituals and has now interviewed over 1,000 people, from Nike and Disney to small startups. He found that people love sharing how they work, and that small details decide whether a ritual succeeds. He quotes Dharmesh Shah's idea that companies build two products, one for customers and one for employees, and the second is culture. Rituals are how people describe a culture, Mehrotra says, which is why he calls them a mirror of it.

Mehrotra shares some favorites. Arianna Huffington's "Reset" is a personal one-minute video with a breathing exercise, and meetings begin by spinning a wheel to choose whose video to play. At Gusto, the whole interview loop joins the offer call to say why the candidate is great, even if they voted no. He says it gives the candidate a standout experience and shows the company how much hiring matters.

He steals ideas freely and calls it the cheapest form of research. Pulse could feel like voting and leave a decision-maker with 30 pieces of feedback and no sense of what mattered. From Coinbase's RAPID decision ritual, Coda added a role column and grouped the table by role. The person running the meeting also pre-fills what they need from each approver, such as whether there is budget.

Mehrotra is writing the book in the open. He shares each finished chapter with the Rituals of Great Teams Brain Trust, a few hundred people who co-edit. Many contribute their own variations, and he calls it a cheap way to get good editors.

Asked how rituals form, Mehrotra says some are great at 100 people and terrible at 1,000, and some suit wartime and not peacetime. He mentions Facebook's "lockdowns", which everyone understands to mean dropping certain work. Coda's Dory and Pulse grew organically from a product manager tired of waiting on Zoom for everyone to speak.

When a company wants to create a ritual on purpose, he recommends *Switch* by Chip and Dan Heath. It describes a rider on an elephant on a path, and you can direct the rider, motivate the elephant or shape the path. Mehrotra maps these to Bing Gordon's tests: tell people in onboarding, give the ritual a name, and make it easy with a template.

Mehrotra tells the story of the "Modern Family question" at YouTube in 2008. People searched YouTube for the show every week, but it wasn't on YouTube, though ABC.com streamed every episode. The product side wanted to link out to serve users, while the business side feared that partners would never post good content if YouTube linked away.

The night before an executive offsite in Half Moon Bay, he read an analysis from Google's shopping team. Users said they chose Amazon because they valued consistency over comprehensiveness. He reframed the debate as whether online video in a decade would favor consistency or comprehensiveness. The team decided on consistency, and the link-out question became easy.

The term eigenquestion borrows from eigenvectors in linear algebra, but Mehrotra says the math doesn't matter. It is the question that, when answered, also answers the most subsequent questions. Instead of ranking questions by importance or order, rank them by how many others they eliminate, the same idea as ranking a to-do list by leverage.

The same answer led YouTube to take back its iPhone app, which Apple's team had built and which could not play almost half the catalog. Mehrotra drove to Cupertino to tell Apple. He told them YouTube would rather be on fewer phones with a consistent experience than on all of them with an inconsistent one. Apple agreed to brick the old app on a set date and send people to the new one, and Mehrotra says it had about 80% share within six months.

Coda's central eigenquestion is whether it is a doc or an app, and the answer is doc, which is why the company's name is "doc" backwards. Mehrotra says the skill is learnable but hard, so people should practice on frivolous problems, as with basketball or piano. His interview question asks candidates to bring a teleportation device to market, then limits them to two questions.

A strong answer asks whether it is safe enough for humans and whether it is costlier to buy or to run. Those two questions produce four quadrants with different strategies, such as replacing airports if it is expensive to buy but cheap to run. Kids often find good eigenquestions, he says, and adults lose the skill, as when his daughter asked why three gas stations sit on one corner.

In 2011, Larry Page reorganized Google into product units, and Mehrotra, who ran YouTube product, was elected to lead a group keeping the PM function consistent. When the eight product leaders tried to identify levels from anonymized level guides, nobody could. The only thing that ordered them was scope. A scope scale also failed because Search is one huge product while the Ads team counts every SKU as a product, scope is an input rather than an output, and a scope-only system would discourage risky projects.

They settled on PSHE, from his former boss Quentin Clark: Problem, Solution, How, Execution. A junior PM is handed the problem, the solution and the how, then executes. More senior people get a problem and a rough solution and figure out the how, then a problem and bring back solutions, and finally tell the company which problems to work on.

Careers start at the execution end and grow in scope, and late careers are about ever-bigger problems. In the middle, the difference isn't scope, and Mehrotra calls this confusing stage the "trough of dissolution". The same logic works for other roles. For salespeople, quota attainment mostly reflects negotiating a quota and territory, while the best can sell anything in any region. Mehrotra values P over S over H over E, and he notes that spotting the right problems is closely related to finding eigenquestions.

Mehrotra says Coda levels everyone on something like PSHE, alongside Radford levels, and the same lens guides interviews. The best way to assess it, though, is references. Strong H-level product managers run efficient meetings and keep everyone informed, and counterparts love them. But when he asks who picks the problems or comes up with the best solutions, the answer is sometimes the designer or the CEO.

References have perverse incentives, since people dislike criticizing others and worry about legal issues. In the ideal case he doesn't say who he is asking about and just asks who on the team is best at a thing. Otherwise he offers four personas: someone who finds the problems, someone who creatively solves them, someone who gets a team moving, someone who executes a playbook. He doesn't reveal which he prefers.

He also asks how excited the reference would be to hire the person again, and what he should have asked. A reference check leads him not to hire "all the time", so he tries to do them early to spare candidates a late rejection.

Mehrotra's ranking of signals puts the reference check at the top, because people who worked with the candidate for years know far more than 30 minutes of artificial scenarios. Next is anything that feels like real work. At the start of each interview loop, the candidate presents to the loop and other employees, spending about half the time on whatever they choose and half on a prompt. Interviews come last, and when signals disagree they are weighed in that order.

He also balances questions by "home court", "away court" and "neutral court". Home-court questions ask what you would do at the interviewer's company, so they test whether the candidate lands on the interviewer's answer. Away-court questions about past work can't be verified. Neutral questions like the teleporter need no company knowledge, and the presentation gives candidates a chance to set an upper bound on themselves.

In the lightning round, he recommends *Switch* and Scott McCloud's *Understanding Comics*, because diagrams crystallize ideas. His second-favorite interview question has candidates pick a product they didn't work on, design its one-page dashboard, and plan a me-too version with a quarter of the time and team. He says his top reads are Lenny's newsletter and Ben Thompson's Stratechery, and he points listeners to the Rituals of Great Teams Brain Trust.