DoistOS: How Doist Built an AI Workplace for Everyone

Going all-in on AI created an unexpected problem at Doist: a widening capability gap.

When Doist gave the whole team access to AI tools and invited people to rebuild their workflows, the most technically advanced people raced ahead. Engineers at the top started building personal agents and custom workflows. Non-technical people in support, design, and operations were left behind.

"There were a lot of challenges coming from this," Warrington said. "It made it very difficult to balance how we distribute work, how we're going to staff cross-functional projects."

The insight that changed Doist's approach came from a simple reframing. It's easy to be impressed by the top performers. But what actually shows how leveraged your organization is isn't the ceiling; it's the average. It's the median Doister, not the AI-native engineer.

Doist made company-wide AI adoption a formal goal for the first half of 2025. They called it "company-wide leverage." Everything that followed, including Doist OS itself, was built in service of that goal.

For leaders navigating this same gap in their own organizations, AI champion programs offer a proven structure for closing it, using peer champions embedded in teams rather than top-down mandates or technology alone.

The grassroots activities that moved the floor before any tool was built

Doist ran a set of structured, repeatable activities to raise baseline AI adoption before Doist OS existed.

• AI lightning talks: Short monthly demos where an internal AI "pro" showcases something they've built, followed by Q&A. The goal is making advanced usage visible to the whole team, not just the engineers.
• AI office hours: Every two weeks, a pro opens their calendar for a one-hour slot where anyone can drop in for small-group troubleshooting and learning.
• Dedicated AI channels: Doist created specific spaces in Twist for AI discussion, rather than letting it scatter. "We said: this is where these conversations go. Everybody is expected to participate. It's not something that a few AI nerds are really excited about."
• Parallel retreat tracks: At Doist's two annual in-person retreats, they run dedicated AI sessions with two parallel tracks: one for pros, one for beginners.
• Top-down consistency: Leadership commits to discussing AI adoption in every leadership meeting. "How do we raise the floor? How are we making sure everybody across the org is benefiting from this?"

This combination maps closely to what structured AI champion programs do at larger organizations: create visible peer-to-peer learning, lower the friction for late adopters, and make AI adoption a shared cultural norm rather than a personal initiative.

Doist OS is a shared AI workspace that connects every tool where the company's knowledge lives, so any team member can ask questions and take actions across all of them from a single interface.

It connects Todoist, Twist, Outline (the handbook), GitHub, and Google Workspace, including Drive, email, Docs, Calendar, and Sheets. Through a CLI, it accesses all of those sources simultaneously, plus the shared context of the company.

The original goal was simple: discoverability. A company survey revealed the number one team pain point was just finding content. "If I have a question, I could go to our 1,800-page handbook. I could go to Twist. I have my email. There's so much to potentially access. How do I find something?"

What they found once they built it was that they could go much further: converting what you find into action and creation was the real unlock.

Warrington's live demo illustrated the difference. Asked about the company's sabbatical policy, Doist OS doesn't just pull up the handbook page. It sees that someone named Nadia has two open Todoist tasks related to updating that policy and has an active conversation about it in Twist. In the same prompt, Warrington could ask it to suggest a revision, push a message to Nadia, and add a task to her Todoist with a deadline. All of that would have taken 15 to 20 minutes across multiple tools. In Doist OS, it takes two minutes.

How Doist OS works under the hood

Three elements make Doist OS smart rather than just searchable.

First, a personal profile in the system. When you onboard into Doist OS, it learns your role, what you're working on, and how you like to communicate. It personalizes responses around your context without you having to re-explain yourself every session.

Second, a "soul" document. This is the personality and values layer of Doist OS. It knows the company's communication standards, principles, and culture. When it responds, it's guided by that.

Third, an address book and index. This is what keeps costs and token usage reasonable. "When I ask about the sabbatical policy, I don't have to say 'check the handbook for this content,'" Warrington explained. "It knows this is likely in the handbook, but there may also be conversations happening in Twist." The index means Doist OS isn't combing through thousands of files on every query. It knows where to look.

The entire system is synced to a GitHub repo. Any change anyone makes is pushed through to everyone's local setup and back to the central repository, keeping it current across the whole company.

Doist OS ships with 55 pre-built skills across 10 domains, and anyone in the company can add new ones.

A skill is a repeatable workflow. "Slash create daily briefing" pulls everything that matters for your morning. "What did I miss while I was out of office?" consolidates a week of Twist threads, flags what needs your decision, and prioritizes what requires action today versus later. "Review this section of the handbook for outdated content" scans a portion of the 1,800-page document, identifies conflicting pages, and surfaces tasks that need to be assigned.

When Warrington demonstrated the handbook review skill live, Doist OS found two conflicting documents on PTO policy, identified pages that overlapped, and recommended consolidation. From that output, he could have asked it to create a Todoist project, split the tasks across three people on the people team, spread deadlines over a month, and link each task directly to the handbook page that needs updating. One prompt. Tasks created, assigned, and contextualized.

The skills system also solves the governance question: when does a personal workflow become a shared company capability? At Doist, that line is drawn by the squad that governs Doist OS. Any new skill created during the first three months required Warrington's review. Every change to the system requires peer review from at least one other person and is visible to the whole company. "It's all in a transparent space. Any change that's made, it's viewable who made the change and why."

The rule that kept humans in the loop

As AI adoption scaled, Doist introduced one clear accountability rule: if you wouldn't put your name to it, don't post it.

"There was a point where people were producing something and just pushing it to the rest of the team, and that information was not correct," Warrington said. The rule became: you are still the human in charge. You need to fact-check the output and make sure what you're sharing with others is accurate. That's your responsibility, not the system's.

This isn't complex governance. It's a cultural norm. "If you don't understand it and you can't put your name to it, then you shouldn't post it. That's prompted people to say, 'Hey, Doist OS, I don't fully understand this part, explain it to me like I'm five.'"

The measure of success Warrington chose wasn't a technical metric. It was a product-market fit survey.

Doist uses the same PMF survey for internal tools that they use for products shipped to their 40+ million users. The question: would you be very upset if this tool were taken away? Doist OS scored higher than any product or feature in the company's history.

Their original goal was 50% of the team as daily active users, targeting specifically the non-technical half, not the engineers who would have adopted AI regardless. They exceeded that number.

What Warrington describes as the "real big unlock" isn't the automation or the search. It's the shared context. "Doist OS is getting smarter day by day. It's knowing where stuff is. It has the shared context from across the team. It knows what I was working on recently and it flagged for me that somebody else was also working on this."

What makes Doist an AI-native organization: what it means for yours

The Doist story is not primarily a story about a clever tool. It's a story about organizational design.

Most organizations approach AI as a productivity layer: give people access to ChatGPT, run a training session, and measure adoption. An AI-native organization goes further. It restructures how knowledge is created, stored, and accessed. It gives AI shared context rather than individual prompts. It measures company-wide leverage, not just usage at the top.

Doist got there faster than most because they already had the foundation: 15+ years of documentation-first, async-first culture. But the principles translate.

‍You don't need 1,800 pages of handbook content to start. You need a decision to treat your existing knowledge base as an AI asset, connect it, and make it shared.

Key takeaways

• Build the floor, not just the ceiling. Company-wide AI leverage is determined by the median user, not the most advanced ones. Design AI systems for the people least likely to adopt on their own.
• Use your existing knowledge base as your first AI asset. Every company already has documentation, communication history, and structured data that can become a shared AI workspace. The question is whether it's connected.
• Make advanced usage visible. Lightning talks, office hours, and dedicated channels let the whole team see what's possible without requiring everyone to figure it out independently.
• Give shared tools a shared identity. Doist OS works because it knows the company, not just the individual user. A "soul" layer and shared context turn a personal AI into an organizational one.
• Keep humans accountable for outputs. Peer review handles governance at the structural level. Individual accountability handles quality at the output level. Both are necessary.

Chase Warrington is Head of Operations at Doist, the creators of Todoist and Twist, used by 40+ million people worldwide.

Watch the full recording here.

To go deeper on building AI capability across your organization, explore the AI Leader Advanced program.