The Leo library
Everything Leo has written down, in one place.
Articles, notes, case studies, videos and podcasts on running a business with Leo, filed in folders. Search every word, or narrow it by what it is and what it is about.
FolderGetting startedWhat Leo is, and your first session with it.2 itemsFolderHow Leo worksHow Leo learns a business, decides what it can say, and joins people up.3 itemsFolderGetting work doneHow Leo turns what it learns into work that gets done.2 itemsFolderConnectionsWhat Leo reads in each tool you connect, and what it does with it.2 itemsFolderPrivacy and securityWhere your data goes, who can see it, and what Leo keeps private.2 itemsFolderBig questionsStraight answers for people deciding.2 itemsFolderPremium Puzzles, in the openA real business of one, running on Leo.1 itemFolderCase studiesWhat Leo saw in real organisations' own public records, and what happened next. Every claim links to its source, so you can check it yourself.28 items in 6 foldersArticleWhat Leo is, in two minutesLeo is smart AI that learns how your business runs, shows you where the time and money go, and offers to build the fix.Getting started · Getting started · How Leo works · 1 min readArticleYour first session with LeoTalk, connect or upload, then watch your business take shape. What happens in the first ten minutes, step by step.Getting started · Getting started · 2 min readArticleAn early read or a recommendationLeo tells you what it can see from day one, and labels it early. A firm recommendation waits for five things to be true.How Leo works · How Leo works · 1 min readArticleWhy Leo is built on game theoryA business is not one mind. It is people whose goals only partly overlap, and Leo is designed around what each of them has reason to do.How Leo works · How Leo works · 3 min readArticleOne brain gets you 5x. Connected brains get you 100xGiving everyone their own AI makes each person faster. Joining every person, agent and tool into one brain changes the shape of the business.How Leo works · Buying decisions · How Leo works · 1 min readArticleThe board: Leo drafts, you say goWhere what Leo found turns into work. Three things at a time, a name on every step, and nothing built until you say go.Getting work done · The board · 1 min readArticleFrom go to done: how work Leo drafts gets builtYou say go, your own agent collects the work, does it and reports back. And a routine, once built, keeps running without you.Getting work done · The board · Agents · 3 min readArticleHow connecting a tool worksConnect a tool and Leo reads it for itself. What it can see, what it can't change, and how to switch it off.Connections · Connections · 1 min readArticleLeo with ShopifyConnect your store and Leo reads orders, products, customers and stock, with the full history. What it looks for, and what it asks you.Connections · Connections · Online shops · 1 min readArticleWhat Leo keeps private, and what it sharesWhat you say about yourself stays with you. What you teach about the work goes on a page with your name on it, and any line comes off when you ask.Privacy and security · Privacy and security · 1 min readArticleWhere does our data go, and who can see it?Every business gets its own Leo, with its own database in one region: Sydney today, or the US or the UK if your business needs it. Only the people you invite can see inside it, nothing is used to train AI, and you can delete it all. Here is the whole path, including the two services that sit outside your region.Privacy and security · Privacy and security · 5 min readArticleDo I need an AI consultant?What a consultant actually does for you, which parts Leo does, and when you still want a person in the room.Big questions · Buying decisions · 2 min readArticleWhy not just point Claude Code or Codex at our company files?A coding agent starts every session cold and reads whatever it happens to open. Leo answers from a picture of your business that your own people taught it, and tells you where each part came from.Big questions · Buying decisions · Agents · 4 min readArticleOne owner, seven routines, sixteen toolsPremium Puzzles is an online puzzle store run by one person, with Leo as its operating brain. Here is what that looks like from the inside.Premium Puzzles, in the open · Premium Puzzles · Agents · 2 min readArticleWhat Leo saw at GitLab, 1. The payroll that was typed by handIn January 2020, adding a new hire to GitLab's US payroll took 28 hand steps between two systems. Leo found it on its first read of the handbook. Three years later the steps were still there.Case studies / GitLab · Finance · People · 4 min readArticleWhat Leo saw at GitLab, 2. One fault, three hand fixes, and the team that removed themIn January 2020 GitLab's support team ran a two hour Salesforce to Zendesk sync by hand two or three times a week to correct one fault. Leo joined three separate hand fixes into one problem. 27 days later, GitLab's team removed them, and the commit says why.Case studies / GitLab · Sales · Customer support · 5 min readArticleWhat Leo saw at GitLab, 3. A promotion copied into five places, and the path nobody offeredIn 2019 an approved promotion at GitLab was retyped into five places from the HR system. Leo offered two ways to stop the copying. GitLab chose a third in 2022: it replaced the HR system. Leo's card records the path it did not offer.Case studies / GitLab · People · 4 min readArticleWhat Leo saw at GitLab, 4. Who owns which customer, kept in four placesAt GitLab in 2019, who owns which customer was kept in four places that had to agree, and no single page said so. Leo joined the pages. Through 2021 the territory page was edited 86 times, one move at a time. Then the handbook copy went.Case studies / GitLab · Sales · 4 min readArticleWhat Leo saw at GitLab, 5. The investor update built by hand from 14 sourcesEvery month in 2019 GitLab's investor update was built by hand from 14 sources, and it went out before the books closed. Leo found the clash by reading two pages side by side. Two years later the monthly update was gone: GitLab reported quarterly as a public company.Case studies / GitLab · Reporting · Finance · 3 min readArticleWhat Leo saw at Apache, 1. The report 211 project leads rebuild by handEvery quarter, each of Apache's 211 project leads writes a board report. Leo heard from every one and saw the same thing nearly 90 times: last quarter's report carried forward by hand, with figures the foundation already holds retyped beside it.Case studies / The Apache Software Foundation · Reporting · 4 min readArticleWhat Leo saw at Apache, 2. One security team, dozens of teams to chaseIn 2019 Apache's Security Team chased projects by hand every month, with 7 to 13 issues over a year old at any time. Leo heard from the team and 30 project leads and laid out three ways forward. In 2023 the same chase was still running.Case studies / The Apache Software Foundation · Engineering · Operations · 4 min readArticleWhat Leo saw at Apache, 3. The project nobody shuts downIn 2019, retiring a dead Apache project waited on someone finding the time. Leo heard the same stall from 12 project leads and suggested a standing signal. In May 2023 the board made every report name the project's state, 'Considering moving to the Attic' included.Case studies / The Apache Software Foundation · Operations · 3 min readArticleWhat Leo saw at Apache, 4. The backlog one team had already solvedIn 2019 dozens of Apache projects watched their issue backlogs grow, while one, Beam, had already fixed its own with a shared triage queue. Leo saw both in different leads' reports. In 2022 the projects took another path, and Leo's card keeps their reason.Case studies / The Apache Software Foundation · Operations · Engineering · 3 min readArticleWhat Leo saw at Apache, 5. The promotions that wait for yearsIn 2019 dozens of Apache project leads said they were looking for new leaders, while years passed with no additions, seven in one case. Leo suggested a standing monthly review. In 2023 one community chose the opposite, deliberately, and said why.Case studies / The Apache Software Foundation · People · 3 min readArticleWhat Leo saw at Kafka, 1. One more channel, added to a brain Leo already hadLeo knew Apache Kafka from four board reports a year: 11 lines, 6 processes, no flows. We added the project's dev list and JIRA for 2019. In 69 minutes it held 837 lines, 28 processes and 25 flows, and had corrected what the board picture got wrong.Case studies / The Apache Software Foundation / Kafka · Engineering · How Leo works · 5 min readArticleWhat Leo saw at Kafka, 2. The red build nobody ownedIn 2019 Kafka's developers filed 193 tickets for tests that fail for no reason, and the board reports never mentioned it. Leo drew the work from twelve developers' mail and JIRA. In 2024 Kafka adopted a quarantine for flaky tests, one of the ways Leo had laid out.Case studies / The Apache Software Foundation / Kafka · Engineering · 5 min readArticleWhat Leo saw at Kafka, 3. The release manager's playbook, from the release managersKafka had a written release process in 2019. Leo drew the one release managers actually ran, from their own mail and JIRA, including a step about flaky tests that Kafka's release page only wrote down in January 2024.Case studies / The Apache Software Foundation / Kafka · Engineering · Operations · 5 min readArticleWhat Leo saw at PostHog, 1. The copy and paste that lasted three yearsIn May 2021 PostHog's handbook said to copy customer conversations from Slack into the CRM by hand, 'until we have a way to integrate the two'. Leo named the integration. The workaround went on 24 June 2024, 1,137 days after it was written down.Case studies / PostHog · Sales · 3 min readArticleWhat Leo saw at PostHog, 2. The support queue that was a week out of datePostHog aimed to answer paying customers in under 15 minutes, while their priority was set by hand once a week. Leo suggested setting it from billing. The weekly step went 158 days after the handbook Leo read.Case studies / PostHog · Customer support · 3 min readArticleWhat Leo saw at PostHog, 3. Always hiring, and asking every timePostHog asks the whole company before opening any new role, and says why. Leo saw that step repeated for engineers they were always hiring. By 2025 engineering was one standing opening, and the step was still there: a choice, not an oversight.Case studies / PostHog · People · 2 min readArticleWhat Leo saw at PostHog, 4. Every flight, handled twicePostHog offsets the carbon from every team flight. In 2021 each flight reached the same person twice, once to repay it and once to offset it. Leo suggested one record. PostHog moved repayment into its card tool instead.Case studies / PostHog · Operations · Finance · 2 min readArticleWhat Leo saw at Sourcegraph, 1. The reason for late invoices, written down and left aloneIn 2021 Sourcegraph's handbook named why invoices ran late: 'missing contact or PO information'. Leo suggested carrying those details on the closed deal. When the handbook was archived in June 2024, the page was word for word the same.Case studies / Sourcegraph · Finance · Sales · 3 min readArticleWhat Leo saw at Sourcegraph, 2. Every incident update, written five timesDuring an outage, Sourcegraph's incident messenger wrote each update for the status page and then again for four Slack channels, every 30 minutes. Leo suggested writing it once. Fifteen days later, the repeat posting was gone.Case studies / Sourcegraph · Engineering · Customer support · 3 min readArticleWhat Leo saw at Sourcegraph, 3. The upgrade done by hand, again and againSourcegraph upgraded each hosted customer by hand, with a short outage every time. Leo laid out three ways forward: script it, keep each customer's settings apart, record the change first. By November 2022 all three were done.Case studies / Sourcegraph · Customer support · Engineering · 3 min readArticleWhat Leo saw at Sourcegraph, 4. Five approvals in a row, and then a sixthA Sourcegraph job offer passed five approvers one after another, then was typed again for signature. Leo suggested approving in parallel. In 2023 Sourcegraph added a sixth approver. Some processes are slow on purpose, and Leo's card records that too.Case studies / Sourcegraph · People · 2 min readArticleWhat Leo saw in Enron's mail, 1. Where is the signed contract?We gave Leo the email eleven people at Enron sent in early 2001, and nothing else. Seven lawyers each wrote about the deal in front of them. Leo joined them into one problem none of them could see alone: nobody was sure where the signed contracts were.Case studies / Enron · Sales · Operations · 4 min readArticleWhat Leo saw in Enron's mail, 2. The same terms, typed again on every tradeNine trades with one customer carried three different credit thresholds, because each confirmation was typed fresh. A lawyer asked how to margin $1.3MM of exposure. Leo heard the same document from four sides, from the email alone.Case studies / Enron · Operations · Sales · 3 min readArticleWhat Leo saw in Enron's mail, 3. Gas still flowing on deals that had endedIn early 2001 gas kept flowing at Enron meters after the deals behind it had ended: about 29 catch-up requests in 13 weeks, each found after the fact. Leo counted them, showed its working, and suggested checking before the month instead.Case studies / Enron · Operations · Finance · 3 min readArticleWhat Leo saw in Enron's mail, 4. The playbook nobody wrote, kept up by the next quarter's mailLeo drew 63 processes from three months of eleven people's email in 47 minutes, for US$22. Given the next three months, it did not start again: it updated 47 pages, added 26, and caught the lawyers deciding to drop confirmations, a change it had not suggested.Case studies / Enron · Operations · How Leo works · 6 min readArticleWhat Leo saw at Zulip, 1. A review playbook from nothing, in 73 minutesZulip used review labels on its pull requests for three years before anyone wrote down what they meant. Leo read 20 people's chat and GitHub, and nothing else, and wrote the playbook in 73 minutes for US$29.75, with some things the written one still leaves out.Case studies / Zulip · Engineering · How Leo works · 7 min readArticleWhat Leo saw at Zulip, 2. How a report in chat becomes a tracked issueZulip's bug reports have lived in a chat channel since 2015. The page on turning them into GitHub issues was written in June 2023, and its author called it quite a project. Leo wrote one from 20 people's messages in 73 minutes.Case studies / Zulip · Engineering · Customer support · 7 min readArticleWhat Leo saw at Zulip, 3. The playbook keeps up when the work changesWhen Zulip's summer contributors arrived in 2023, pull request review grew three new stages. Leo's page grew them too, from what people did, in 37 minutes for US$17. Nobody edited it.Case studies / Zulip · Engineering · How Leo works · 5 min read
Nothing here says that yet. Try fewer words, or another word for it.