Building an AI Agent to Automate Cross-Functional Project Updates

At a fast-moving SaaS company, I was a senior manager of marketing operations leading campaign and platform operations. My team supported marketing roles on repeatable campaigns such as webinars, as well as larger cross-functional projects involving teams outside marketing. Asana was our operational source of truth: I managed project templates, assigned tasks, and set due dates so work could move consistently from planning to launch. Slack, however, was where people actually communicated—in public channels, threads, and DMs.

That split created a persistent gap between planned work and active conversations. Tasks were missed or went overdue, context stayed buried in messages, and I spent at least three hours a week searching for blockers and chasing updates. The trigger was recognizing that the problem was not a lack of project structure; it was that the structure lived somewhere people were not talking. I needed to connect Asana’s planning system to Slack’s existing habits without asking a fast-moving marketing team to change them.

The broader need was to make project-status conversations proactive, especially across cross-functional campaigns, without asking marketing teams to adopt another place to work. My objective was narrower and personal: define the problem, prototype a reusable Asana–Slack workflow, and launch it beyond my own projects so other teams could use it without me manually chasing updates.

I owned the solution end to end, but I was not a project manager by title, and there was no existing foundation connecting Asana’s tasks with Slack’s conversations. I also had to prove the workflow safely: I began with a human-in-the-loop review, tested it across roughly 10 webinar campaigns and one cross-department project, and only then made it available to the AI Champion group and selected coworkers. Success meant the agent could start relevant, actionable conversations automatically, produce genuine updates or insights—not just post messages—and reduce the manual work involved. My baseline was at least three hours a week spent searching for blockers and chasing people; I expected the workflow to remove most of that effort, while recognizing that fewer overdue tasks would be useful but was not yet measurable.

I started with a simple principle: meet people where project conversations already happened instead of asking them to return to Asana. I considered trying to solve the gap inside Asana, including with Asana AI and ChatGPT, but those approaches did not create a reliable connection between the project plan and the Slack conversations where work was actually moving.

First, I created a one-to-one relationship between an Asana project and a Slack channel by adding a project-level field and building a guided agent workflow. When I requested a channel, it applied a standard naming convention, posted a welcome message explaining the channel’s purpose, identified task assignees from the Asana project and invited them, and created a Slack canvas containing the relevant links. This gave each project a clear conversational home and reduced the need to search across DMs, threads, and documents.

Next, I built a scheduled agent that scanned each Asana project weekly for overdue and soon-due tasks. It formatted the findings for quick review in Slack, mentioned the relevant assignees, and tagged the project owner so they could add context, answer questions, or escalate an issue. I began with a human-in-the-loop workflow: the agent drafted each message, and I reviewed it before sending. I tested and refined the workflow across about 10 webinar campaigns and one cross-department project, adjusting the messages and presentation until they produced real updates and insights rather than noise.

Once I trusted the behavior, I made the agent available to the entire company. I also handled a limitation in the orchestration layer—its inability to archive completed channels—by creating a backup Asana task so channel cleanup would not be forgotten.

I used three tools because each solved a different part of the workflow. Asana remained the source of truth for tasks: project templates defined repeatable work, assignees, due dates, and dependencies. I added a project-level field linking each Asana project to its Slack channel so the systems could be connected.

Slack was where project conversations already happened. The workflow created standardized channels, invited task assignees, built a canvas with relevant links, and started formatted conversations about overdue and soon-due work. That made timeline changes—such as a landing-page deadline or speaker-confirmation dependency—visible to the stakeholders who needed to act, instead of leaving them buried in a DM.

Dust was the load-bearing tool. I used it to build the guided channel-creation workflow and the scheduled agent, with permissions to read Asana and take actions in Slack. It could map Asana assignees to Slack users, customize how updates were displayed, mention the right people, and run the scan on a schedule.

I tried Asana’s built-in AI and used ChatGPT to help configure the prompt, but neither could reliably connect Asana users to Slack users, create the Slack workflow, customize the message format, or trigger it when I needed. Dust could. I first reviewed every generated message before sending it, then made the workflow autonomous after testing it across about 10 webinar projects and one cross-functional project. I would use this combination again, while keeping a backup Asana task for limitations such as channel archiving.

The first measurable result was time saved: I reduced the manual work of searching Asana for blockers and chasing updates from a baseline of at least three hours a week. The workflow produced value even before it became fully autonomous, because it connected Asana’s source of truth to the right Slack users and turned project status into a message that could start a conversation. Once scheduled, it ran every Tuesday at 8:00 a.m., often surfacing overdue or soon-due work before people had started their day. Project owners responded more quickly to timeline changes—for example, around landing-page deadlines and speaker confirmations—because updates were visible to all relevant stakeholders instead of getting lost in DMs. The workflow also moved me and my team out of private chasing and into shared project conversations. I did not measure overdue-task volume, broader adoption, or long-term project outcomes, so I can’t claim those improved. The result I can defend is faster escalation and less time spent finding and requesting information.