Who are you and what do you do at Camunda?
My name is Lana, and I'm a senior strategic program manager in the Corporate Strategy and Execution team at Camunda. I work at the intersection of strategy, product, and go-to-market, helping turn executive priorities into measurable outcomes. Most of my work focuses on enterprise transformation, AI first operating models and bringing new strategic initiatives from idea to execution.
What did the work look like before and why did you decide to try AI on it?
As any program or project manager can probably relate to, a large part of my work happens in environments with a high level of ambiguity and a high mental load. You're driving multiple strategic initiatives in parallel while continuously aligning stakeholders, monitoring progress, preparing executive updates, communicating transparently, identifying risks, and adapting plans as priorities change.
A lot of that quickly turns into administrative overhead, duplicate maintenance and constantly moving information between documents, presentations, and tracking systems. At some point I found myself thinking there has to be a smarter way to do this.
I wasn't looking for AI to make decisions for me or replace strategic thinking. I was looking for a way to reduce the cognitive overhead so I could spend more time where I actually create value: solving problems, connecting people, making decisions and driving execution. That's what got me started integrating AI into my daily work.
What did you actually build or change and what's the impact?
The biggest change was how I operate. AI has helped me drive more strategic initiatives in parallel, including company-wide AI adoption, the ProcessOS product proof point and the great process re-engineering, the global ProcessOS beta rollout, feature adoption and Agentic Orchestration GTM Program migration, the Company handbook migration, and executive strategy support.
For the Quote to Cash re-engineering initiative, which later became the foundation for the ProcessOS beta roll out, I worked with forward deployed engineers to reconstruct the current process using existing documentation, system knowledge, and subject matter expertise. Claude Code helped identify knowledge gaps, prepare stakeholder interview guides, test process hypotheses, consolidate findings, and improve documentation. This accelerated discovery and allowed us to focus more on solving the business problem.
I also use Claude Artifacts to turn fragmented information from documents, Slack conversations, repositories, and presentations into interactive visualizations and executive-ready overviews. When our GitHub-based knowledge base for the global ProcessOS beta rollout became stale, I moved toward a more automated approach.
Scheduled agents now read meeting transcripts, Slack activity, documents, and email, while helping me maintain task boards, schedule focus time, and draft daily recommendations. Nothing is auto-sent; I review everything first. This reinforced that automation is only as good as the data it reads, so my agents now verify whether issues have already been resolved before flagging them.
Impact: I successfully managed 5+ major strategic initiatives in parallel, reduced preparation time for many strategy documents, executive updates, and first drafts by roughly 50 to 70 percent, and shifted more of my time from documentation and coordination toward stakeholder alignment, strategic thinking, and execution.
Primary tools:
- Claude Code
- Claude Artifacts
- GitHub
- Glean
What went wrong, and where does it still let you down?
One example was our company handbook migration. I got excited about what AI could do and trusted it a bit too quickly. We moved fast, but that also led to unnecessary back and forth because I didn't validate enough before acting on the output. That was a good reminder that AI isn't a replacement for judgment. It's incredibly fast, but it's still my responsibility to decide when its suggestions are good enough to move forward. Especially in areas where I'm not the specialist, trusting AI blindly was not a good call. Bringing in an engineer early, aligning on direction, and then letting AI accelerate execution would have worked much better.
Something I'm still figuring out is where AI is actually worth using. Sometimes I catch myself trying to automate a task I could just finish faster myself. The challenge isn't using AI everywhere, it's recognizing where it creates real leverage and where it just adds complexity.
That said, for the parts of my role that are genuinely repeatable, like daily status tracking, meeting follow up and calendar management, I've recently started handing those fully to scheduled agents that operate independently. The highly contextual, judgment heavy work stays a close collaboration between me and AI rather than something I hand off outright.
If you were starting from scratch tomorrow, what would you do differently?
I would spend much more time teaching AI how I work before asking it to help me work. If I were starting over, I'd spend the first one or two weeks feeding Claude as much context as possible about my role, my responsibilities, the types of decisions I make and the way I like to operate. Then I'd ask it to look at my work and figure out where AI should accelerate me, where automation actually makes sense, and where human thinking should stay in the driver's seat. Instead of discovering good workflows gradually over months, I'd intentionally build my own AI co-pilot from the start, one that's built around how I work instead of expecting a generic assistant to fit my job.
And a few weeks ago, I actually did exactly that. I spent real time walking an AI through my role, my mandate and how I actually work, before building anything. What came out of it was noticeably better than anything I'd gotten from one off requests.
What can other teams steal or try right now?
Treat each major initiative as a long-lived AI project, not a series of disconnected chats. Create a dedicated Claude Project with a living context document covering the goals, stakeholders, decisions, terminology, operating model, and current status. This reduces context switching and gives AI the context to support updates, communications, and decision-making.
Before automating anything, verify that the underlying data is current. Agents should check whether an issue has already been resolved before flagging it as a risk. More broadly, don’t ask AI to do your job. Ask it to make you better at your job by challenging assumptions, structuring your thinking, spotting blind spots, and accelerating learning.
