Zendesk’s Director of Product urges product leaders to take a human approach to AI adoption
Discover how Zendesk’s product teams use AI to accelerate delivery while keeping human leadership, judgment, and team development at the center.
Mirza Besirovic
Director of Product at Zendesk
Journey: Mirza Besirovic transitioned from poetry translation to AI product leadership over a twenty-year career.
AI Adoption: Zendesk expands roles and accelerates development using AI, increasing potential remits across teams.
Limitations: AI use is limited in personnel matters, emphasizing a human touch for team development and growth.
Productivity: Early AI results show increased development speed, though adoption remains inconsistent across teams.
Leadership: Product leaders must become more technical, embracing AI tinkering and supporting team adaptation.
Mirza Besirovic is the Director of Product at Zendesk, where he leads 15 product teams focused on agentic systems.
We asked Mirza to tell us about his journey with AI adoption within Zendesk's product organization. He told us how sporadic adoption can become more consistent across teams.
An unconventional product leadership journey
Twenty years ago, I was a poetry translator. A series of accidents and choices led me to become an AI product leader.
Continue Reading for Free
Create a free account to finish this article, plus get ongoing access to timely insights and practical resources.
Have an account? Log In
The tech part of my journey started at a network for scientists where, years ago, we built datasets for all sorts of ML experimentation. In the past 15 years, I've worked across travel tech, fintech, streaming, data science, and for the past few years in conversational AI and automation.
Today, I'm Director of Product at Zendesk, leading a group of just under 15 product teams that own a portfolio across AI agents and agentic systems, RAG and retrieval platforms, analytics, integrations, and AI-powered onboarding and self-service. We are a distributed team mainly located in Europe, working globally with our San Francisco headquarters, partners in APAC, and customers worldwide.
I joined Zendesk via the Ultimate.ai acquisition in 2024 as Head of Product. Since then, we've launched a couple of generations of generative bots, conversational automation products, and an autonomous agentic system automating millions of customer service conversations on text and voice channels for global brands like Netflix, Vimeo, Levi's, and many others.
In the past 18 months, my team has scaled AI agents to over $100MM in ARR and expanded our reach to all corners of the world.
How AI expands roles in product development
Because of AI, we've expanded each individual contributor's potential remit: designers build GitHub repos, PMs submit pull requests, and engineers drive product decisions. AI, especially coding agents, expanded everyone's ability to step outside their comfort zones, learn, and lean into adjacent domains. This change allowed us to bridge gaps and move faster while retaining the desired quality level.
For example, we had a Figma prototype for a new testing experience, but lacked a fully-formed team to own this work. Instead, one of our incredible product designers used her skills and a coding agent to build a working product.
AI, especially coding agents, expanded everyone’s ability to step outside their comfort zones, learn, and lean into adjacent domains.
Enabling our team to run better data analysis also improved product quality. While we continue to work closely with data science and product analytics, PMs can self-serve at a much higher rate than ever before to produce high-quality analysis that influences product decisions.
Additionally, we're experimenting with team formats and org design. We've seen at least one team of three AI-native ICs leverage coding agents to deliver the output of a much larger team.
Where product leaders should limit AI usage
My team leverages AI across the entire product lifecycle.
And in my role as a product leader, I've built a second brain and work copilot. What started as an Obsidian vault has transformed into a personal repo to brainstorm ideas and develop strategies.
I synthesize data much faster than ever before, manage conversations across an extensive portfolio, and debate with a strategic sparring buddy that performs various pre-trained roles, retains memory, and understands my work style.
But I limit AI's use when it comes to people: my team, their development, career growth, and our working relationship. While I might ask my copilot to report on activities (pulling from internal systems), I rely on good ol' human instinct to lead and develop my team.
Some things require a human touch, after all.
Why early results with AI are positive
Some teams have increased development speed 3-4x — but these are early results and not yet statistically significant. We are waiting to connect increased output and velocity to lagging indicators and long-term outcomes before reaching a verdict.
Early qualitative signs are positive too: higher quality UX decisions, better research material, faster turnaround, and fewer handoffs between teams. AI acts as organizational glue more than expected. Personally, it enables me to work across a larger portfolio and stay on top of things.
What to do when AI adoption is uneven
That said, adoption is uneven, and it's not yet commonplace or fully mainstream. Like everyone, we're at the beginning of that vaunted 10x productivity journey. It hasn't moved as fast as the hype suggests, because change is a human concern. A lot of folks still need to retrain and incorporate AI into their workflows. It takes time.
We've done a few things to make adoption more consistent:
- Built an internal AI gateway that gives everyone access to a set of coding agents, either via a TUI or a more user-friendly browser version, both of which are compliant with our internal infosec policies
- Started an AI product guild and several generative AI exchange formats, programs, and get-togethers to showcase how people use AI, the tools they build, and the problems they solve
- Started redesigning workflows for producing lifecycle artifacts like PRDs, prototypes, etc., to work with coding agents
- Built a series of MCPs and integrations to enable teams to connect to internal systems and access data securely and compliantly
Adoption is uneven, and it’s not yet commonplace or fully mainstream.
And here are a few tips:
- Ensure equal access for technical and non-technical folks: tokens, CLI, desktop apps.
- Encourage diversity of LLMs, agents, and tools. Let your team experiment to find what works best in your environment.
- This requires a budget. Secure it, and work with your engineering counterparts to draft a plan. Prepare for Finance and Ops involvement as AI becomes an operational expense, not IT COGS.
- Sharing is caring. Have your team showcase their work.
- Help kick things off. Build artifacts and workflows yourself to solve repetitive PM tasks (drafting product briefs, writing strategy docs, etc.).
Why product teams must become technical
I've become more bullish on PMs becoming more technical.
I used to believe a general understanding of basic software development terms and concepts was enough. But with AI, I now believe we all need to be more technical. This doesn't mean great product people can't come from all backgrounds. I'm a linguist by training, and before tech, I was a poetry translator.
But we need to adopt a more universalist approach to learning. Being technical is not the same as becoming a developer. It is, however, about curiosity, understanding what powers the product experiences we build, tinkering, and adopting a builder mindset. Because it's never been easier to learn all things technical.
How a second brain can assist product leaders
The second brain I mentioned earlier is an AI-powered personal OS that helps me work as a Product Director across my org, using Claude Code as the chief interface.
Inputs flow into it and it handles the pipeline: processing transcripts, extracting knowledge, and cross-referencing them against a large body of memory files.
The system triages for priority, maps dependencies, and pre-processes relevant data into audience-specific formats. For example, casual Slack bullets for my team, structured updates for leadership, or narrative paragraphs for strategy docs. Effectively, the system replaces what would traditionally require a chief of staff and an analyst.
It maintains my to-do list, tracks strategic threads I manage, and synthesizes an extensive reading backlog into indexed references, allowing me to prioritize my time effectively. I use my second brain to keep track of ongoing work and quickly synthesize vast swathes of data when I need information on the fly.
Data recency and quality are continual concerns. Stale data can bias LLM-powered workflows and produce negative outcomes. I filter, clean, and maintain incoming data, but since I also connect to third-party systems where I don't always control the data, I implement quality gates for the output. A rule I've heard people say is that you should know what output you want before providing the system input, and test accordingly. Not exactly foolproof, but with enough fiddling with instructions and memory, the system produces high enough quality.
I've crossed that threshold where I can rely on 80-90% of the output my second brain provides.
Why product leaders and their teams must tinker with AI
Here's my advice to CPOs:
- Embrace it. Start exploring AI in your work and private contexts. Start by building a personal copilot, structuring your data, letting it reason over that data, and uncovering new patterns. Start pet projects and build that idea you've had for years. Product leaders are by nature creative tinkerers, but many of us did not start out as engineers, so we always felt constrained. This has changed rapidly and drastically.
- Do not mandate AI adoption without tools, resources, or structural support. If you do, you're effectively tilting at windmills. Enthusiasts will get there with or without you, but the rest of the organization will fall behind.
- Keep an eye on your team and support them on this journey. Many concerns and worries exist, not to mention LinkedIn fearmongering, about what AI's advent means for product managers and other folks in tech and business. Have honest conversations with your team and help structure learning plans. I want to do more to help my team carve out more time for thinking, experimenting, and tinkering.
- You need to be diligent about security. Also, the cost of your users generating AI is quickly bringing the token honeymoon to an end.
Follow along
You can follow Mirza Besirovic's work on LinkedIn, his You Are The Product newsletter, and his website.
More expert interviews to come on The CPO Club!
