Director of Product, AI Agents at Zendesk · Berlin

Mirza
Beširović

Building AI products and the orgs that ship them

I lead Zendesk AI Agents. I help founders and product leaders turn AI strategy into products, teams, and durable businesses.

Work with me→ Book me to speak→

Selected work

The platform, the pricing, the org, the acquisitions

My portfolio across AI products, commercial models, organizational design, and M&A integration.

Impact

$100M+Scaled AI Agents into a core growth business at Zendesk.
4×Customer-base growth in 18 months after the Ultimate acquisition.
80%+Automation rates across mature customer deployments.
2Product integrations following the Solvemate and Ultimate acquisitions.
01

The platform

Zendesk AI Agents

When I joined Ultimate (later acquired by Zendesk), the product was a deterministic bot executing decision trees. LLMs upgraded it into a hybrid of scripted and generative logic. Today it's a multi-agent system that reads procedures written in natural language, calls APIs, and resolves customer issues end to end for Netflix, Vimeo, Levi's, Reddit, and thousands of others. The whole arc took two and a half years.

01 · Deterministic

Dialog flows

Hand-built flows. Every path designed in advance.

02 · Generative

Zero-training bots

LLMs inside the dialog logic. Faster setup, same deterministic core.

03 · Agentic

Multi-agent systems

Reasons over procedures, calls APIs, resolves on messaging, email, and voice. Judged by evals.

02

The commercial product

Resolution-based pricing

When customers buy AI they buy an outcome, and per-seat pricing puts the vendor's incentive against theirs. At one point we were losing money on our best customers. We tried various pricing models in production before we found the one that worked. Zendesk became the first CX vendor to implement outcome-based pricing: charging for verified automated resolutions instead of capacity.

Per seat
pay for capacity
Per resolution
pay when it resolves
03

The org as product

The agentic product org

The org is a product too. Building AI agents splits product work into two speeds. One side is AI-native: research cycles, evals, fast prototyping, planning around model releases. The other ships the rest of the product: features, bug fixes, customer commitments, the actual revenue motion. The two need different cadences and different career paths, and most companies let the split break them. I design for it instead:

  • Prototypes over PRDs: PMs build with coding agents
  • Evals are infrastructure with owners: decomposed metrics for retrieval, generation, and execution
  • Deletion is rewarded: deprecation goals live in annual objectives
  • Career mobility between AI-native and traditional teams, by design
  • Internal processes run on agents
04

M&A integrations

Two acquisitions, from the inside

Solvemate into Dixa, then Ultimate.ai into Zendesk. Product due diligence, merging product orgs, scaling the acquired product, and consolidating platforms and infrastructure without stalling the roadmap.

Solvemate → Dixa · 2022 / Ultimate.ai → Zendesk · 2024

Advisory

Strategy that gets built

I work with founders and product leaders on the AI decisions that are expensive to get wrong: product strategy, org design, and M&A.

01

AI product strategy

Decide where agents are worth building, define the product and commercial model, and turn an AI ambition into a sequence of decisions.

02

Agentic organizations

Redesign product work around prototypes, context architecture, evals, and teams that can operate at two speeds.

03

M&A integration

Plan the product side of a deal, integrate the product orgs, and keep the acquired business shipping through the transition.

From decision to prototype

I help shape the decision and, where useful, build a focused prototype that gives the team something concrete to test.

Mirza is a one man wrecking crew. An absolute machine and the guy you want on your team if you want to win.
John Fontein · Co-founder, Sesame
Work with me→

Speaking

What AI agents change about the job

Keynotes, fireside chats, panels, and podcasts, drawn from decisions I've had to make building enterprise AI products.

What I speak about

  1. AI transformation. Why most AI pilots stall and what taking one to $100M actually required. The demo is the easy part; everything after the demo is the job.
  2. AI-native leadership. How leaders restructure their own work, product organizations, and operating models around AI.
  3. The Three Tensions of building AI agents at scale. The Control Paradox, the Orchestration Trap, the Evolution Crisis.
  4. The agentic era of product management. Prototypes over PRDs, context architecture, spec-driven development, and evals as a core product skill.

Watch a recent talk

Human in the Loop: PMs in the agentic era — ProductLab Berlin →
His talk was in-depth and hands-on, and our audience ranked him top 3 speakers of the conference. A year later, people are still talking about it.
Daniele RoncaProductLab Conf Berlin · September 2025
The conversation around moving from pilots to measurable business value was one of the most important themes across the summit.
Till SchmidATS Accelerate Tomorrow AI Summit · June 2026

More talks

Selected recordings on AI product leadership, careers, and customer-service automation.

How to Train Agentic PMsThe skills, working practices, and organizational support product managers need to move into AI product work.Product People livestream · 2025→ AI Agents and the Future of Product ManagementHow agentic systems change what's expected of product managers.Alps GenAI Summit · 2025→ Finetuning a Career in AIA career conversation on getting into AI product work, and what the job is actually like.ADPList BeMore · 2024→ Delivering Human + AI-Powered Customer ServiceAn early take on splitting customer service work between automation and humans, from 2023.PAKCON · Berlin · 2023→
Book me to speak→

Selected writing

I write You Are the Product

Essays for product leaders building AI products and the teams behind them.

01The Three AI Tensions Breaking Product ManagementA practical framework for the control, orchestration, and evolution problems that make production AI agents fail.→ 02How to Train AI PMs and Keep Your JobWhat an AI-capable PM does differently, and how to retrain a team without pausing the roadmap.→ 03Finetuning Your Career: AI Product LeadershipA field note on how building AI products changes the work of PMs, product leaders, and research teams.→
Subscribe→ Browse the archive→

About

Product leadership is translation

I run product for Zendesk AI Agents, the company's flagship AI automation platform. My teams have scaled it past $100M ARR, automating 80%+ of customer interactions for Netflix, Vimeo, Levi's, Reddit, and thousands of others. That includes the commercial side: Zendesk was the first CX vendor to price AI per automated resolution.

The route here was not a straight line. Before AI, my product career ran through travel tech at Omio, insurance at wefox, streaming at Zattoo, and a social network for scientists at ResearchGate. In 2022 I moved into conversational AI and never left. Four companies and two acquisitions later, I'm building agentic bots.

I trained as a linguist and spent six years as a translator and interpreter. Now I build chatbots for a living. That's still how I work: making complex things intelligible, bridging what one side knows with what the other needs. Good AI products translate between what machines can do and what people actually need, without taking human judgment out of the loop.

Mirza joined my company as our product lead. This had a huge impact on the acquisition of the company thereafter.
Jürgen Vogel · Co-founder, Solvemate

The journey

  1. 2024–nowZendeskDirector of Product, AI AgentsCustomer service software used by 100K+ businesses
  2. 2024Ultimate.aiHead of Product; acquired by ZendeskAI agents for customer support
  3. 2022–24DixaGroup PM, then Director of ProductOmnichannel customer service platform
  4. 2022SolvemateProduct Lead; acquired by DixaChatbot automation startup, Berlin
  5. 2012–21European scale-upsResearchGate · Omio · wefox · Zattoo
  6. 2005–11TranslationTranslator and interpreter
See my full experience on LinkedIn→

Contact

Get in touch

Speaking, advising, or comparing notes on building with agents. The fastest way to a yes is a specific message: what you want from me, and when.

Emailmirza@mirza.tech NewsletterYou Are the Product LinkedInin/mirzabesirovic GitHubgithub.com/mrzbsr