I’ve been at Asana for years. So when I moved into the General Counsel role, I thought the transition would be straightforward. Same company, same people, same mission. I know where the bathrooms are.
That confidence lasted about four days.
Here’s the thing nobody tells you about internal promotions: there’s no onboarding. You’ve already been onboarded. You’re supposed to know. So instead of a structured first week, you get a calendar that fills itself, a Slack DM count that doubles, and the creeping suspicion that something important is about to fall through a crack you haven’t found yet.
The crack is always there. A contract about to expire. A project that stalled six months ago for a reason nobody wrote down. A commitment your predecessor made that lives only in their memory and a calendar invite they deleted on their way out.
Welcome to leadership. Don’t drop the ball.
Start From Somewhere, Not From Scratch
On one of my first days in my new role, I opened Asana and let our AI Chief of Staff tool, Asana Dash (soon to be publicly available), surface what actually needed my attention. Not the firehose — the firehose would have been overwhelming — but what had momentum, what had gone quiet, and what was waiting on me specifically.
Buried in that list was a draft blog post, set aside a year earlier with a small note from the original author to circle back when Asana’s AI tech has evolved.
That note alone didn’t mean much to me, just a stranded draft from someone who was previously handling the work. If institutional memory didn’t live in Asana, there would be no thread to pull, no context for why it mattered, no way to tell if it still did.
So I put Asana’s AI capabilities to work: I asked our Government Affairs-dedicated AI Teammate, Regula-Tori Clause or “Tori,” to pull the project history, the prior comments, and the related decisions that had accumulated around the work over time. Instead of searching across multiple systems and threads myself, I could quickly see why the draft existed, what conversations had shaped it, and what had changed since it was paused. Tori didn’t hand me a file. Tori already understood our legal priorities, our voice, and what we were trying to accomplish, so she was able to hand me a solid update to the draft in ~two minutes.
Thirty total minutes of iterating in Asana, and the post was finished. Thanks to Asana’s context-rich AI capabilities, I had started from somewhere instead of from scratch.
AI Needs a Shared System of Work
What we’ve built at Asana is the operating system for human-agent teams — a platform where AI Teammates can work inside the same shared environment as the rest of us, with the same context, the same permissions, and the same accountability. They track priorities, surface what’s slipping, and take action on routine work transparently. They can get new human teammates up to speed quickly. And they don’t replace human judgment; they free people from their “work about work” so they can spend their time on the things that actually need it.
The same infrastructure that has helped teams coordinate work for nearly two decades — our Enterprise Work Graph® — turns out to be particularly valuable in powering the next generation of AI enablement. With the backbone of our infrastructure, we are able to give agents and AI-powered workflows the organizational context to provide better outcomes. It’s a living map of every person, task, project, goal, and dependency across an organization, connected on a single shared plan. When an AI Teammate like Tori operates inside that system, it isn’t guessing. It already knows what your team is doing, what’s been decided, and what still needs someone to act. We call this new stage of our product evolution the operating system for human-agent teams.
Two things make that possible. The first is multiplayer: every workflow, app, and agent runs in the same shared space, where humans and agents act on the same plan and see the same context. Multiple people can train, guide, and improve any agent, and the whole team can see what each agent is doing in the same project. Nobody has to ask what the agent did. The second is shared memory. Every task completed, every piece of feedback given, every preference your team sets gets carried forward — within a governance framework that controls what gets retained, who can see it, and what an agent can act on. AI Teammates learn from the work itself, so nobody repeats the context they already gave, and every workflow starts smarter than the last.
With our recent acquisition of StackAI, that capability now integrates across the external tools where work happens — procurement, HR, finance, compliance, etc. When coupled with Asana’s AI Teammates, it means AI that can act, with permissions and an audit trail, across multiple systems where the work actually lives.
The Case for Risk-Based AI Policy
Here’s where my role makes me look at this differently than most.
As a lawyer, I care less about whether an AI can produce an answer than whether I can understand where that answer came from. Institutional memory documented in Asana isn’t valuable to me because it makes work faster; it’s valuable because it makes decisions traceable. When an AI Teammate completes a task in Asana, I can see what it did, why it did it, what data it touched, and who authorized the action.
That distinction matters far beyond our walls. Right now, much of the policy conversation about AI treats AI as a single category — the same regulatory posture applied whether the system is an autonomous model acting on its own or an AI that operates inside an enterprise platform with built-in human oversight, scoped permissions, and a full audit record. Those are very different risk profiles. They deserve very different treatment.
An AI Teammate at Asana operates with three things that change the regulatory calculus: Context that comes from the organization’s own work, not a generic prompt; Checkpoints where a human reviews and approves before the AI acts on anything consequential; and Controls that govern data access through the same permissions framework as human users. That combination — context, checkpoints, and controls — is what makes AI auditable, accountable, and ultimately safe to deploy in regulated settings. It’s also what should inform how policymakers think about which AI systems require heavier guardrails and which ones already have them built in.
Smart AI policy starts from that distinction. Not every AI system carries the same risk. The systems built with traceability, human oversight, and access controls from the foundation are demonstrably different from systems where those things were bolted on later — or never added at all. Policy that recognizes this difference encourages responsible deployment. Policy that treats every AI system identically discourages it.
AI Adoption Is Organizational Change
Across the 180,000-plus organizations using Asana, the pattern is consistent. Teams don’t have to figure out which agent to build or how to onboard it — there are thirty prebuilt AI Teammates ready to work on day one, preapproved and preauthorized, with the right skills for the function. Marketing teams use them to draft campaign briefs and review creative against brand guidelines. IT teams use them to categorize and route tickets and build a knowledge base that gets smarter every time something is resolved. Operations and PMO teams use them to compress sprawling project data into something a leader can actually act on. Different functions, same underlying capability: an AI that holds context across the work, the team, and the systems they touch.
The organizations getting the most out of this aren’t the ones with the most advanced models. They’re the ones who treated AI adoption as an organizational change, not a technical one. They defined governance before they scaled deployment. They started with structured, repeatable work — places where a human checkpoint is natural — before moving into more ambiguous territory. They onboarded AI Teammates the way they would onboard any new team member: set permissions once, give the right context, and show people how to work alongside them.
Continuity Depends on Institutional Knowledge
There’s a version of this story that sits beyond my role at Asana — one I keep coming back to because the parallels are hard to miss.
Institutional knowledge is fragile. It lives in people, and people leave. Administrations change. Staff rotate. Contractors come and go. A program that took a decade to build gets handed to someone new with a two-paragraph email and a shared drive that hasn’t been organized since 2019. The new person isn’t the problem — they’re capable, they’re motivated. They’re also going to spend their first six months doing work about understanding the work, while the actual work waits.
That’s a systems problem. And it’s a problem AI can actually solve, if the AI in question knows your organization.
For a government department running a decade-long program across rotating staff and changing leadership, an AI that holds the institutional memory isn’t a productivity story. It’s continuity. It’s the difference between a transition that loses momentum and one that doesn’t.
AI Should Build on Organizational Memory
The best day-one experience I can imagine isn’t a faster day one, it’s one where the anxiety of not knowing is replaced by the confidence of having context. Where you don’t spend your first month auditioning for the job you already have. Where the calendar still fills itself, but at least you know what’s on it and why.
That’s what AI built on organizational memory delivers. Not a magic trick. Not a chatbot with a personality. A Teammate that knows immediately how to get you up to speed on what’s happening when you walk in.
I felt that on one of my first days as General Counsel. Most people in a new role won’t, yet.
That’s the part we’re trying to fix.