Albert Cheng
Growth and monetization leader, formerly at Duolingo and Grammarly, now at Chess.com
Where to hear and read about Albert Cheng2 episodes and tests
Lenny's PodcastFinding hidden growth opportunities in your product | Albert Cheng (Duolingo, Grammarly, Chess.com)
SubversiveHow Chess.com Became the World’s Top Chess AppAlbert Cheng is a growth and monetization leader who has worked at YouTube, SoFi, Duolingo and Grammarly, and now leads growth at Chess.com. The thing most worth learning from him is to alternate between finding insights and spreading each win across the whole product, on top of tracking and a culture that make experiments compound.
A big part of my job on joining the company is to help the growth be less lumpy.
How Chess.com Became the World’s Top Chess App
Albert Cheng: the story and the lessons
1Albert Cheng's path from classical pianist to growth leader
Cheng's childhood piano routine built the habits growth work needs. His Taiwanese immigrant parents loved classical music, and a 90-minute countdown clock sat on the family's upright Yamaha every day. He has perfect pitch, which let him tell right notes from wrong ones quickly.
He considered a music conservatory but chose engineering school because his intrinsic motivation for music had weakened. He sees music and growth as the same loop: constant mistakes, tight feedback, resilience, and creativity inside a structure.
He moved from code to users to business across top consumer apps. He studied computer engineering, then moved from engineering to sales engineering and product, then to growth over the last five years. His employers include YouTube, SoFi, Duolingo and Grammarly, and he joined Chess.com about a year before the second interview.
At YouTube he worked on streaming and gaming features used by over 20 million people. He says each step added to his understanding: first the code, then the user, then the business.
Each company taught him a different way to win. Duolingo ran a tight, consistent experimentation process and wrote a playbook called the Green Machine. Grammarly mixed self-serve, enterprise and core product, and Chess.com is run by people fanatical about chess.
He sees Duolingo's real strength as motivation and habit, and Grammarly's as integration across many apps. Grammarly's hybrid go-to-market model taught him what product-led growth means in practice, a term he never heard in his first decade.
2How Cheng finds and spreads growth wins
Cheng alternates between exploring insights and exploiting them. Exploring is finding the right mountain to climb, and exploiting is focusing resources on climbing it. Too much exploring is scattershot, and too much exploiting saturates into local maximizing.
The idea reached him through his Grammarly engineering partner Nirmal, who had taken Reforge classes. Most people treat it as a macro strategy, but he applies it to a single insight.
He found that 80% of game reviews happen after a win. Chess.com built game review so people could study their losses. Data showed 80% of reviewers use it after a win, so the team flipped the loss experience to show brilliant moves and an encouraging coach message.
Game reviews grew 25% and subscriptions 20%. Typical experiment win rates are only 30 to 50%, so intuition alone leaves gains unclaimed.
He spreads each win across the product because users carry one mindset. After the win, other PMs audited their own areas for the same pattern, such as puzzles, with their success ratings, copy and button colors. The original PM only has to state the hypothesis and finding clearly.
Companies split the product into around 50 pieces, but users do not change psychology between features. He says a hugely losing experiment is just as valuable to share as a winner.
A rising share of flat tests tells him to go back to exploring. He uses an experiment explorer tool to look across all running experiments for links between hypotheses and learnings. When more tests stop reaching significance, he calls the team back to brainstorm.
He frames it as an accordion: lean in while low-hanging fruit lasts, then explore again when the hit rate falls.
Cheng avoids chasing a 60 to 80% hit rate. He says a high hit rate pushes teams into small-ball tests that do not move the needle. Roughly half of tests fail to validate the hypothesis, and newer areas like a fresh monetization team win more than retention work.
"I think a lot of companies are too afraid to take bigger swings and fail."
3How Cheng nearly doubled Grammarly's upgrade rate
Cheng first fixed tracking to see why free users rarely paid. Over 90% of Grammarly users are free, so the free-to-paid path was the key lever. The team instrumented suggestion types and paywall views and found the paywall appeared only after a user accepted all suggestions.
Very few free users did that, because they picked and chose instead.
He sampled paid suggestions inside free users' writing. Grammarly mixed a limited number of paid suggestions, such as tone, clarity and full sentence rewrites, into free users' writing. Upgrade rates nearly doubled.
The fear was that giving features away would hurt subscriptions, and it did not. Free users had only seen spelling and grammar fixes, so they saw Grammarly as a small tool. He compares it to a reverse trial tied to usage instead of time.
He keeps core value free and samples premium on top. For mission-driven consumer products, free is the lowest friction and lets word of mouth and network effects do the growth work. He calls free core plus sampled premium the sweet spot he has seen.
Reverse trials work better in B2B, where lock-in makes people feel the loss when the window closes. Consumer products rarely have that lock-in.
4Cheng on retention, returning users and beginners
Cheng treats retention as what makes growth pay off. Without retention, a consumer subscription must win payment on day one, which forces paid acquisition and aggressive upselling. He calls a day-one retention of 30 to 40% solid for a consumer app.
Duolingo and Chess.com grew by word of mouth in a market they helped expand, not by bidding against rivals.
He designs a separate experience for resurrected users. Mature products pile up dormant and sporadic users, who can number in the hundreds of millions at Chess.com. New and resurrected users there are about the same size.
At Duolingo, contact sync sent a push when a best friend started using the app, and returners redid a placement test so they started in the right spot.
For a passive tool like Grammarly he focused on activation. Nobody opens Grammarly daily, so installation and the aha moment carry users for a long time, and daily active counts mislead. He had staffed a growth team on retention, then realized it was getting in the way and handed the metric to the core product team.
Retention there depends on the quality and frequency of suggestions.
He found that losing a first game cuts retention by 10%. Over 75% of new Chess.com users say they are beginners, and fewer than a third win their first game. Retention after a loss is 10% worse than after a win.
Tests include a guided learn-to-play start in place of a live game, and hiding ratings for the first five games so people do not see them plummet.
Cheng keeps removing cold moments from Chess.com. A Slack channel collects harsh spots such as blunder callouts and red buttons. After a loss the coach now says something encouraging and the game-over modal shows positive glyphs.
Positivity has kept improving retention over the past year, though he expects the well to run dry eventually. Staff who play a lot overlook how harsh chess feels to newcomers.
He grew Duolingo sharing by improving moments users already screenshotted. A virality team briefly tracked screenshots and found streak milestones and funny challenges were shared most, while top-three leaderboard finishes were not. Illustrators and animators were then staffed on those moments.
He says leaning in can "5x or 10x" growth from sharing.
5How Cheng builds an experimentation culture at Chess.com
He audited tracking and metrics before asking teams to experiment. Chess.com ran almost no experiments before 2023, and experimentation began in earnest in 2024. His first job was auditing the platform, the metrics and which teams were comfortable with testing.
Teams that doubt the data are hard to get onto the experiment wheel. He says the system matters as much as any single experiment, and cites a company whose retention was configured backwards for three months, so every positive result was actually negative.
Cheng set a made-up target of 1,000 experiments a year. Chess.com went from almost none to about 50, then about 250, with the 1,000 target ahead. He admits the number is invented, and says its value is the conversation about what must be true to hit it.
Reaching it needs no-code tests, lifecycle marketing, app store screenshots and keywords, engineers launching tests from their own ideas, and fast analysis. Running 1,000 tests without learning would be a waste.
He wins the company over by making wins and losses visible. Experiments are posted in Slack and all-hands with hypothesis, result, impact and learning. He shows with real data how a metric move changes revenue or active users, and repeats the message as a leader.
Goals set in a vacuum fail. "People have to see it working." He also says he needed the co-founders on board, since being at odds with their approach would have been a problem.
He tells new teams to buy an experimentation tool first. Grammarly used Statsig, while Duolingo and Chess.com built their own about 15 years ago, before tools existed. He generally advises against building in-house from day one.
He cites an Atlassian report that about 40% of product teams run no experiments at all. His advice: crawl, then walk, then run.
6How Chess.com grows and why Cheng wants it steadier
Cheng joined a company already carried by creators, events and a boom. Chess.com has passed 200 million registered users, with 40 to 50 million monthly actives and about 1.5 million subscribers. The pandemic and The Queen's Gambit more than quadrupled new registrations, and usage settled much higher than before.
Years of work on product and infrastructure let it catch the wave, and it now runs north of 20 million games a day. Much of the roadmap went to keeping the service reliable under the surge.
Chess.com brings in beginners through creators and crossover events. YouTube is a top three channel in the "how did you hear about us" survey, and social is projected at about three billion impressions this year. PogChamps has non-chess personalities like Mr. Beast and Rainn Wilson play each other.
The survey is directional, not exact attribution, but a channel can climb from 4% to 10% in a week after a viral post. When Gukesh became world champion, registrations in India passed those in the US.
He skips generic paid ads and keeps paid for targeted fixes. Chess.com owns the domain, ranks high organically, and its small tests of generic Facebook and Google ads did not pass the LTV and CAC test. Paid may be used to retarget lapsed subscribers.
"We just need to make chess in the ethos, right?" He points to Duolingo, which spends little on paid and used TV in Japan to lift awareness.
He cut events to fewer, better produced formats. After the boom, Chess.com dialed down expensive tournaments with big prize pools. It kept the Speed Chess Championship, weekly Titled Tuesday, and two or three big events a year.
The aim was never to manufacture a boom, but to test which marketing strategies worked.
Cheng's job is to turn lumpy viral waves into compounding growth. He calls it the 1.01 to the hundredth effect: many small wins in retention and conversion. The path to 100 million daily players, up from about 10 million today, needs beginner onboarding and mainstream social content.
The game is to grow the daily active base, then convert it better. The free game review loop drives upgrades, since people pay for unlimited reviews.
He wants a daily-active north star to focus a team that says yes to everything. Chess.com's passionate staff all play and watch chess daily, which makes prioritization hard. He plans to bring the Duolingo model, with daily actives as the north star and retention as the hero driver.
He calls the passion a shadow of a superpower.
7How Cheng uses AI in growth and in the product
Cheng uses Slack bots to answer data questions for the whole company. Chess.com is training bots to give first-pass answers to ad hoc questions, such as subscribers in South Africa, that used to wait on an analyst. The bot does the analysis, not only the SQL.
People skip questions that feel embarrassing or bothersome, so an easy tool brings an explosion of questions.
He prototypes core screens with AI so exploring gets cheaper. The team built prototypes of onboarding, the home screen and the chessboard with tools like v0 and Lovable. Others in the company layer their ideas on top.
A team reacts to something it can click around in, which he says "makes a world of difference." Handoffs between tools are still clumsy, and product managers do not yet ship code to production.
He applies the right technology to the right feature. Game review pairs a chess engine that evaluates every move with a language model that explains it in plain language, in the user's own language, with audio. Language models are weak at chess itself and hallucinate moves.
The planned virtual coach will have several personalities across game review, puzzles and lessons, and let players ask what to train. Engines stay for analysis and fair-play detection.
8Cheng's hiring rules and the lesson of Chariot
Cheng hires for agency and clock speed over matching experience. The best performers he saw were fast, high-agency learners who cared about the mission, even without deep domain experience. He says "sometimes that experience could be a crutch" while AI changes the ground.
He reads soft signals: the candidate's questions, whether they went deep into the product, their references and how they set up the interview. He once read only from his rubric and now weighs these signals more.
He does best at a mid-sized company, 500 to 1,000 people. Big companies ship slowly, and tiny startups win users one by one. He wants to see across the company, still "look at the pixels," and work at a daily and weekly pace.
He favors companies 10 to 20 years old that are durable, ideally profitable, with good leaders and at key inflection points.
Chariot Direct taught him to start from the user problem. Chariot ran 15-person commuter shuttles on fixed routes in San Francisco, which riders loved. The team then added dynamic routes for better utilization, a solution searching for a problem.
The team also missed that drivers and operations were users of the marketplace, and it did PR before validating demand. The sunk cost pushed them to see it through. He says the lessons carried into every product he has built since.
Finding hidden growth opportunities in your product | Albert Cheng (Duolingo, Grammarly, Chess.com)
Teaches how to find and scale growth wins by alternating between exploring new insights and exploiting proven ones, with examples from Duolingo, Grammarly and Chess.com. Worth hearing for concrete results: Grammarly's upgrade rate nearly doubled from sampling paid features for free users, and Chess.com's game review change lifted subscriptions 20%.
Cheng's childhood piano routine built the habits growth work needs. His Taiwanese immigrant parents loved classical music, and a 90-minute countdown clock sat on the family's upright Yamaha every day. He has perfect pitch, which let him tell right notes from wrong ones quickly.
He considered a music conservatory but chose engineering school because his intrinsic motivation for music had weakened. He sees music and growth as the same loop: constant mistakes, tight feedback, resilience, and creativity inside a structure.
He moved from code to users to business across top consumer apps. He studied computer engineering, then moved from engineering to sales engineering and product, then to growth over the last five years. His employers include YouTube, SoFi, Duolingo and Grammarly, and he joined Chess.com about a year before the second interview.
At YouTube he worked on streaming and gaming features used by over 20 million people. He says each step added to his understanding: first the code, then the user, then the business.
Each company taught him a different way to win. Duolingo ran a tight, consistent experimentation process and wrote a playbook called the Green Machine. Grammarly mixed self-serve, enterprise and core product, and Chess.com is run by people fanatical about chess.
He sees Duolingo's real strength as motivation and habit, and Grammarly's as integration across many apps. Grammarly's hybrid go-to-market model taught him what product-led growth means in practice, a term he never heard in his first decade.
Cheng alternates between exploring insights and exploiting them. Exploring is finding the right mountain to climb, and exploiting is focusing resources on climbing it. Too much exploring is scattershot, and too much exploiting saturates into local maximizing.
The idea reached him through his Grammarly engineering partner Nirmal, who had taken Reforge classes. Most people treat it as a macro strategy, but he applies it to a single insight.
He found that 80% of game reviews happen after a win. Chess.com built game review so people could study their losses. Data showed 80% of reviewers use it after a win, so the team flipped the loss experience to show brilliant moves and an encouraging coach message.
Game reviews grew 25% and subscriptions 20%. Typical experiment win rates are only 30 to 50%, so intuition alone leaves gains unclaimed.
He spreads each win across the product because users carry one mindset. After the win, other PMs audited their own areas for the same pattern, such as puzzles, with their success ratings, copy and button colors. The original PM only has to state the hypothesis and finding clearly.
Companies split the product into around 50 pieces, but users do not change psychology between features. He says a hugely losing experiment is just as valuable to share as a winner.
A rising share of flat tests tells him to go back to exploring. He uses an experiment explorer tool to look across all running experiments for links between hypotheses and learnings. When more tests stop reaching significance, he calls the team back to brainstorm.
He frames it as an accordion: lean in while low-hanging fruit lasts, then explore again when the hit rate falls.
Cheng first fixed tracking to see why free users rarely paid. Over 90% of Grammarly users are free, so the free-to-paid path was the key lever. The team instrumented suggestion types and paywall views and found the paywall appeared only after a user accepted all suggestions.
Very few free users did that, because they picked and chose instead.
He sampled paid suggestions inside free users' writing. Grammarly mixed a limited number of paid suggestions, such as tone, clarity and full sentence rewrites, into free users' writing. Upgrade rates nearly doubled.
The fear was that giving features away would hurt subscriptions, and it did not. Free users had only seen spelling and grammar fixes, so they saw Grammarly as a small tool. He compares it to a reverse trial tied to usage instead of time.
He keeps core value free and samples premium on top. For mission-driven consumer products, free is the lowest friction and lets word of mouth and network effects do the growth work. He calls free core plus sampled premium the sweet spot he has seen.
Reverse trials work better in B2B, where lock-in makes people feel the loss when the window closes. Consumer products rarely have that lock-in.
Cheng treats retention as what makes growth pay off. Without retention, a consumer subscription must win payment on day one, which forces paid acquisition and aggressive upselling. He calls a day-one retention of 30 to 40% solid for a consumer app.
Duolingo and Chess.com grew by word of mouth in a market they helped expand, not by bidding against rivals.
He designs a separate experience for resurrected users. Mature products pile up dormant and sporadic users, who can number in the hundreds of millions at Chess.com. New and resurrected users there are about the same size.
At Duolingo, contact sync sent a push when a best friend started using the app, and returners redid a placement test so they started in the right spot.
For a passive tool like Grammarly he focused on activation. Nobody opens Grammarly daily, so installation and the aha moment carry users for a long time, and daily active counts mislead. He had staffed a growth team on retention, then realized it was getting in the way and handed the metric to the core product team.
Retention there depends on the quality and frequency of suggestions.
He found that losing a first game cuts retention by 10%. Over 75% of new Chess.com users say they are beginners, and fewer than a third win their first game. Retention after a loss is 10% worse than after a win.
Tests include a guided learn-to-play start in place of a live game, and hiding ratings for the first five games so people do not see them plummet.
He grew Duolingo sharing by improving moments users already screenshotted. A virality team briefly tracked screenshots and found streak milestones and funny challenges were shared most, while top-three leaderboard finishes were not. Illustrators and animators were then staffed on those moments.
He says leaning in can "5x or 10x" growth from sharing.
He audited tracking and metrics before asking teams to experiment. Chess.com ran almost no experiments before 2023, and experimentation began in earnest in 2024. His first job was auditing the platform, the metrics and which teams were comfortable with testing.
Teams that doubt the data are hard to get onto the experiment wheel. He says the system matters as much as any single experiment, and cites a company whose retention was configured backwards for three months, so every positive result was actually negative.
Cheng set a made-up target of 1,000 experiments a year. Chess.com went from almost none to about 50, then about 250, with the 1,000 target ahead. He admits the number is invented, and says its value is the conversation about what must be true to hit it.
Reaching it needs no-code tests, lifecycle marketing, app store screenshots and keywords, engineers launching tests from their own ideas, and fast analysis. Running 1,000 tests without learning would be a waste.
He wins the company over by making wins and losses visible. Experiments are posted in Slack and all-hands with hypothesis, result, impact and learning. He shows with real data how a metric move changes revenue or active users, and repeats the message as a leader.
Goals set in a vacuum fail. "People have to see it working." He also says he needed the co-founders on board, since being at odds with their approach would have been a problem.
He tells new teams to buy an experimentation tool first. Grammarly used Statsig, while Duolingo and Chess.com built their own about 15 years ago, before tools existed. He generally advises against building in-house from day one.
He cites an Atlassian report that about 40% of product teams run no experiments at all. His advice: crawl, then walk, then run.
Cheng uses Slack bots to answer data questions for the whole company. Chess.com is training bots to give first-pass answers to ad hoc questions, such as subscribers in South Africa, that used to wait on an analyst. The bot does the analysis, not only the SQL.
People skip questions that feel embarrassing or bothersome, so an easy tool brings an explosion of questions.
He prototypes core screens with AI so exploring gets cheaper. The team built prototypes of onboarding, the home screen and the chessboard with tools like v0 and Lovable. Others in the company layer their ideas on top.
A team reacts to something it can click around in, which he says "makes a world of difference." Handoffs between tools are still clumsy, and product managers do not yet ship code to production.
He applies the right technology to the right feature. Game review pairs a chess engine that evaluates every move with a language model that explains it in plain language, in the user's own language, with audio. Language models are weak at chess itself and hallucinate moves.
The planned virtual coach will have several personalities across game review, puzzles and lessons, and let players ask what to train. Engines stay for analysis and fair-play detection.
Cheng hires for agency and clock speed over matching experience. The best performers he saw were fast, high-agency learners who cared about the mission, even without deep domain experience. He says "sometimes that experience could be a crutch" while AI changes the ground.
He reads soft signals: the candidate's questions, whether they went deep into the product, their references and how they set up the interview. He once read only from his rubric and now weighs these signals more.
He does best at a mid-sized company, 500 to 1,000 people. Big companies ship slowly, and tiny startups win users one by one. He wants to see across the company, still "look at the pixels," and work at a daily and weekly pace.
He favors companies 10 to 20 years old that are durable, ideally profitable, with good leaders and at key inflection points.
Chariot Direct taught him to start from the user problem. Chariot ran 15-person commuter shuttles on fixed routes in San Francisco, which riders loved. The team then added dynamic routes for better utilization, a solution searching for a problem.
The team also missed that drivers and operations were users of the marketplace, and it did PR before validating demand. The sunk cost pushed them to see it through. He says the lessons carried into every product he has built since.
How Chess.com Became the World’s Top Chess App
Chess.com grew to 200 million registered users with almost no paid advertising, riding creators, events and a pandemic plus Queen's Gambit surge. Its growth chief explains how it now shifts from lumpy viral waves to steady product-led growth through high-velocity experimentation, with a target of 1,000 experiments a year.
He moved from code to users to business across top consumer apps. He studied computer engineering, then moved from engineering to sales engineering and product, then to growth over the last five years. His employers include YouTube, SoFi, Duolingo and Grammarly, and he joined Chess.com about a year before the second interview.
At YouTube he worked on streaming and gaming features used by over 20 million people. He says each step added to his understanding: first the code, then the user, then the business.
Each company taught him a different way to win. Duolingo ran a tight, consistent experimentation process and wrote a playbook called the Green Machine. Grammarly mixed self-serve, enterprise and core product, and Chess.com is run by people fanatical about chess.
He sees Duolingo's real strength as motivation and habit, and Grammarly's as integration across many apps. Grammarly's hybrid go-to-market model taught him what product-led growth means in practice, a term he never heard in his first decade.
A rising share of flat tests tells him to go back to exploring. He uses an experiment explorer tool to look across all running experiments for links between hypotheses and learnings. When more tests stop reaching significance, he calls the team back to brainstorm.
He frames it as an accordion: lean in while low-hanging fruit lasts, then explore again when the hit rate falls.
Cheng avoids chasing a 60 to 80% hit rate. He says a high hit rate pushes teams into small-ball tests that do not move the needle. Roughly half of tests fail to validate the hypothesis, and newer areas like a fresh monetization team win more than retention work.
"I think a lot of companies are too afraid to take bigger swings and fail."
Cheng keeps removing cold moments from Chess.com. A Slack channel collects harsh spots such as blunder callouts and red buttons. After a loss the coach now says something encouraging and the game-over modal shows positive glyphs.
Positivity has kept improving retention over the past year, though he expects the well to run dry eventually. Staff who play a lot overlook how harsh chess feels to newcomers.
He audited tracking and metrics before asking teams to experiment. Chess.com ran almost no experiments before 2023, and experimentation began in earnest in 2024. His first job was auditing the platform, the metrics and which teams were comfortable with testing.
Teams that doubt the data are hard to get onto the experiment wheel. He says the system matters as much as any single experiment, and cites a company whose retention was configured backwards for three months, so every positive result was actually negative.
Cheng set a made-up target of 1,000 experiments a year. Chess.com went from almost none to about 50, then about 250, with the 1,000 target ahead. He admits the number is invented, and says its value is the conversation about what must be true to hit it.
Reaching it needs no-code tests, lifecycle marketing, app store screenshots and keywords, engineers launching tests from their own ideas, and fast analysis. Running 1,000 tests without learning would be a waste.
He wins the company over by making wins and losses visible. Experiments are posted in Slack and all-hands with hypothesis, result, impact and learning. He shows with real data how a metric move changes revenue or active users, and repeats the message as a leader.
Goals set in a vacuum fail. "People have to see it working." He also says he needed the co-founders on board, since being at odds with their approach would have been a problem.
Cheng joined a company already carried by creators, events and a boom. Chess.com has passed 200 million registered users, with 40 to 50 million monthly actives and about 1.5 million subscribers. The pandemic and The Queen's Gambit more than quadrupled new registrations, and usage settled much higher than before.
Years of work on product and infrastructure let it catch the wave, and it now runs north of 20 million games a day. Much of the roadmap went to keeping the service reliable under the surge.
Chess.com brings in beginners through creators and crossover events. YouTube is a top three channel in the "how did you hear about us" survey, and social is projected at about three billion impressions this year. PogChamps has non-chess personalities like Mr. Beast and Rainn Wilson play each other.
The survey is directional, not exact attribution, but a channel can climb from 4% to 10% in a week after a viral post. When Gukesh became world champion, registrations in India passed those in the US.
He skips generic paid ads and keeps paid for targeted fixes. Chess.com owns the domain, ranks high organically, and its small tests of generic Facebook and Google ads did not pass the LTV and CAC test. Paid may be used to retarget lapsed subscribers.
"We just need to make chess in the ethos, right?" He points to Duolingo, which spends little on paid and used TV in Japan to lift awareness.
He cut events to fewer, better produced formats. After the boom, Chess.com dialed down expensive tournaments with big prize pools. It kept the Speed Chess Championship, weekly Titled Tuesday, and two or three big events a year.
The aim was never to manufacture a boom, but to test which marketing strategies worked.
Cheng's job is to turn lumpy viral waves into compounding growth. He calls it the 1.01 to the hundredth effect: many small wins in retention and conversion. The path to 100 million daily players, up from about 10 million today, needs beginner onboarding and mainstream social content.
The game is to grow the daily active base, then convert it better. The free game review loop drives upgrades, since people pay for unlimited reviews.
He wants a daily-active north star to focus a team that says yes to everything. Chess.com's passionate staff all play and watch chess daily, which makes prioritization hard. He plans to bring the Duolingo model, with daily actives as the north star and retention as the hero driver.
He calls the passion a shadow of a superpower.
He applies the right technology to the right feature. Game review pairs a chess engine that evaluates every move with a language model that explains it in plain language, in the user's own language, with audio. Language models are weak at chess itself and hallucinate moves.
The planned virtual coach will have several personalities across game review, puzzles and lessons, and let players ask what to train. Engines stay for analysis and fair-play detection.