工艺重于指标：Linear 如何用一名 PM、不做 A/B 测试做出软件

Saarinen describes Linear as a project and issue tracking system for software teams, started in 2019 by three founders who had seen clunky, legacy tools at several companies. He names customers including Block, Vercel, Ramp, Retool and Mercury, plus thousands of others from freshly graduated YC startups to public companies. He adds that Linear has been profitable for the last two years and has a net negative lifetime burn rate, meaning it has more cash in the bank than it has raised.

Craft, he says, starts with hiring people who care, because friction and small paper cuts make a collaboration tool annoying. It does sometimes push timelines, for example when he looks at a feature about to launch and says it does not feel right, but he says the delays are usually days, not rewrites. The team deliberately avoids spending lots of time designing and polishing before building, since building reveals how the feature should really work.

In practice, teams put a feature into the production app within about the first week where only staff can see it. Next they find a customer or opt-in group, which can range from one to ten, and tolerate a janky experience while saying it is not finished. Often they choose a large company, such as Vercel for the roadmap feature, to explain how they work. Full polish comes only at general release.

Saarinen says people mistakenly think of craft as perfecting things early, and that perfection mindsets make it hard to ship anything. Linear balances this by pushing things out quickly and fixing them quickly. There is no formal review of everything shipped, but there is a check before it goes to everyone.

Saarinen says the craft culture is not just his: his two co-founders are engineers who often obsess over details even more than he does, such as where an animation should go. He argues that craft is only possible if the company values it, and advises founders from different corporate backgrounds to align on quality before hiring, since one company cannot run on several cultures at once.

Linear has only one PM, Nan Yu, who is head of product. Saarinen says specialized roles can mean the person who decides is not the person looking at the button all day, so Linear gives project teams of engineering and design the ownership. His example is an engineer, Andreas, who unprompted built dynamic safe zones for submenus so users can move diagonally to a submenu instead of needing to hover pixel-perfectly.

Nan joined as a contractor to help shape the insights data tool and later took on defining direction and helping teams align. Saarinen says he steers but is not in every meeting or writing every spec. Every project has a lead, who can be an engineer or designer, as an assignment rather than a title, and that person communicates scope and progress. Not everyone wants this, so Linear hires people interested in broader scope.

Lenny notes that this requires a very high hiring bar and engineers who actually want to do project communication. Saarinen agrees and says interviews must test product skills as well as engineering skills, by asking how candidates think about products, whether they form opinions, use judgment and can articulate them. He adds that Linear aims to make PMs' jobs easier too, since engineers actually keep the tool updated, and he expects more PMs as the company grows, each focused on a specific area.

Asked whether design can be a differentiator against incumbents, Saarinen says the more a domain matters, the more design matters. When a new paradigm like mobile or web appears, the first products do not need to be well designed, but once there are hundreds of email clients, any new one must be pretty good to be taken seriously. He notes that the early Google or YouTube websites were very basic and would get no attention today.

He adds that this is not always fair, because a good product without a designer may be dismissed. Still, design is never the reason a company succeeds on its own: the product has to be better or different, and design enables it, much as good technology does. He compares it to packaging at Apple, which sets expectations before you use the product, and says landing pages do the same early on when nobody knows you.

When people face many choices, he says, they often pick whatever looks highest quality, and a strong brand can make the product almost secondary, as with Nike. On building brand, Saarinen says it must be authentic and is not a logo or color palette. It accumulates from what a company does, how it speaks, how it treats customers and how it builds its site and product, and he cites Brian Chesky's attention to brand at Airbnb.

Saarinen still runs the design team, and he or another co-founder or head of product acts as sponsor for projects. Early on the team reviews demos and aims to get the main concept right. Before launch, he goes into the product and tries different states, which is how he spotted janky animations and scrolling in threaded comments and had the release pulled back until they were fixed. For newer ideas such as project updates, the team launches a scoped version knowing it is not fully figured out and learns from use.

Decisions, he says, come from understanding the problem and judging whether the solution is right and good enough, not from A/B tests or goals to move a number. Linear sometimes has telemetry, but has no feature-level metric goals, and only occasionally a company-level one such as weekly active users. He explains that Linear's users differ in size and culture and the product is a system of many parts, so optimizing one number does not make sense, and success is customers agreeing the problem is solved.

The Linear Method is a published set of principles. Saarinen believes productivity software should be opinionated, since flexible software makes teams spend time figuring out ten ways to use a feature, and his design mantra is to design something for someone. Cycles are Linear's optional take on sprints: a fixed timeframe, run on an automated schedule, with a chosen set of priorities, so that when something changes the team can explain what was decided and what came up.

Saarinen describes Linear as a mix of magic and science. The science is talking to users: every project starts with some user research, founders have weekly customer calls, and everyone is encouraged to answer questions in shared customer Slack channels. Stats are used to answer a specific question like "I wonder what is going on," not to pick a number to raise.

The magic is what happens when everyone has built up customer and product understanding: discussions become better informed, and people can use intuition rather than data to back decisions. The team shares notes, videos and sessions, and Saarinen says it is usually enough if one or two people deeply understand a given area. He adds that teams often lean on data out of fear of making the wrong choice, and that Linear accepts mistakes can be fixed, as long as people and not data made the choice.

Lenny contrasts this with Ramp's velocity-and-metrics approach, and concludes there are many ways to do it if you commit fully. Saarinen says Amazon and Apple are both successful in different ways, and that the domain matters. Linear, he says, is in the retention and trust business, wanting customers to adopt it early and stay forever, which requires consistent quality and respecting customers rather than a transactional mindset.

Lenny praises Saarinen's focus, and Saarinen credits advice from his 2012 YC experience: founders should talk to customers, build the product and exercise, and if they are doing something else, it is probably the wrong thing. Exercise, he says, keeps you healthy and prevents burnout. He also asks whether something is important to do now or later, and says doing this podcast earlier would have been less interesting because Linear was not yet at the right scale.

The same applies to the product. Linear knew it would eventually need a SOC 2 certificate but said no until later, and told customers so, and many of them were fine with it. Internally the team borrows from role-playing games: the main quest is building a great product for customers, and ideas like making T-shirts are side quests that do not advance it. Saarinen applies this to himself as well.

Saarinen says that in his experience, a small team of high-quality people is easier and faster than a large team of average ones. Linear did not hire anyone in roughly its first year, then added a few people at a time, with a guideline never to more than double in a year. Across roles, the team looks for taste and broader perspective: engineers with product sense, marketers who are good storytellers, operations hires who understand HR.

He says people who are a bit more than their title make the company easier to manage, because they pick up things easily and rarely say "it's not my job." Linear aims for fewer specialized roles and fewer levels, so new people can own large areas from the start rather than needing to climb levels to gain scope.

After standard interview loops, every hire does a paid work trial in which the candidate acts as a mini contractor on a deliberately vague problem. An engineer might get a feature to build, scope it down, build a version in the code base and present it. Candidates can join meetings, use Slack and Notion and chat with teammates, and Saarinen says both sides learn how they would work together. Few people have declined, and the company tries to schedule trials on weekends or holidays.

To assess product sensibility, Saarinen asks people to go deep on past projects: why a decision was made, whether they agreed, what they would do differently. He says a vague "I didn't like it" is not enough, while strong candidates can reason and keep going deeper.

Linear officially started in April 2019, after some months of prototyping. The founders built a local-first data structure, with data living in the client and syncing to the backend, because they wanted the app to be fast and found nothing off the shelf. They announced the company in mid-April with a waitlist site, could use the product themselves by May, and by June or July had a couple hundred users across about 10 companies.

The product stayed in private beta for almost a year. The waitlist included a survey about current tools, company size and motivation, and Saarinen invited smaller companies using the integrations Linear supported, while looking for genuinely interested people. Within about a month the list had around 4,000 people, and initially he read the survey in a spreadsheet and emailed invite links from his personal address.

Linear invited only about 10 people a week at first, then more, in cohorts. Inviting everyone at once would mean everyone hit the same early bugs and sent the same feedback, so the team fixed problems before inviting the next cohort. After a year Linear launched publicly with pricing and several hundred companies, and almost all of them subscribed to the paid plan.

The first 10 companies were a few friends' startups and mostly waitlist signups, and none paid at first. Later the team added a settings page with a slider letting people choose what to pay per seat, with a maximum of $20, and some chose $28 or $1 in tests. Saarinen says it did not produce enough data to set pricing, but it gave confidence there was a market under $20.

Saarinen says Linear is always somewhat paranoid about product-market fit, asking whether it really has it and with whom. The team felt it early with smaller early-stage companies where founders still run the product and care about shipping speed. The first two years focused on being the default tool for startups, and the last two years on strengthening fit with larger companies of hundreds or a thousand people, where he acknowledges a Fortune 500 firm would not yet be well served.

Lenny describes it as a spectrum and a map where you slowly acquire territory. Saarinen agrees, noting that B2B has customer sizes, domains and categories, and recounts Zoom founder Eric telling him they doubled down on a customer type that worked, like universities, instead of chasing everyone. Linear sees similar pull today from AI companies, and earlier from crypto.

On launching, Saarinen had maybe 10,000 followers from design talks and blog posts, which helped but was not huge. He credits an announcement that was direct and authentic rather than corporate, with each founder writing their own reasons. He says founders should ask who the first audience is, where they are and what language they use. The small angel round with friends also made it feel like a real company with people to answer to, and those friends helped spread the word.

Because Linear is fully remote across Europe and the US, happy hours do not work well, so the team runs a quarterly baking and cooking competition inspired by The Great British Baking Show. People buy ingredients on the company card, cook along on Zoom, help each other when stuck and post photos in Slack, with friendly competition on presentation. Saarinen sees it as another craft activity, and says that as a designer he tends to do well on visuals, since no one can taste the results over Zoom.

On leadership, Saarinen says the surprise of becoming CEO was how many different things land on you, from compensation to marketing plans to offsites, while trying to stay focused. He says he has not fully figured it out, but hiring and delegation to other leaders help, and he enjoys learning areas like finance and legal. Looking ahead, Linear is building a feature called Asks, a Slack integration that lets people outside the product team send requests to a team's triage in Linear and get replies in Slack without being Linear users.

In the lightning round, Saarinen recommends *The Timeless Way of Building* by Christopher Alexander and *Zen and the Art of Motorcycle Maintenance*, both for their thinking about quality. His favorite interview question is what a candidate is most proud of and why, and his motto is "go slow to go fast." He says the best Finnish food to try is creamy salmon soup and invites listeners to find him on Twitter and try Linear with his team's help.