Trezor · Product Vision 2026
"AI isn't just changing our tools —
it's changing how teams are organized."
Daniel Valčík
Head of Design / Trezor / March 2026
of enterprise apps will integrate AI agents by end of 2026
Gartnerof product professionals already use AI regularly
Reforgeof large engineering teams will reorganize by 2030
Gartnersaved through deep AI integration into workflows
ReforgeWhere the industry stands, and where Trezor fits in
Most product organizations today operate on the model described by Marty Cagan at SVPG. An empowered product team consists of a product manager, a product designer, and engineers led by a tech lead.
The team is focused on outcomes, not output, and has the authority to find the best solutions to the problems it's been entrusted with.
This model worked brilliantly in the cloud era.
Trezor currently operates in domain-based teams: Growth, Wallet, Trade, Earn, Foundation, and others. Each team has a PM, designer, and tech lead with engineers.
It's a classic empowered model. The CEO wants to go the AI route, which opens a fundamental question:
How must this model adapt?
AI isn't just a new tool. Like cloud computing before it, AI represents a paradigm shift that will completely redefine how we build products. Cloud didn't just add servers. It catalyzed cross-functional pods, continuous deployment, product-led growth. AI will do the same for team structure and decision-making.
Specialized departments hand off work sequentially through documentation gates. Requirements → Design → Code → Test. Upfront planning was rational when hardware cost millions.
Brooks argues for small, sharp teams around a chief programmer. "Adding manpower to a late project makes it later." Communication overhead grows quadratically. Doubling the team more than doubles the coordination cost.
Cross-functional teams deliver working software in short iterations. 17 practitioners write the Agile Manifesto at Snowbird, Utah. Customer collaboration replaces contract negotiation.
"You build it, you run it." Development and operations merge into one team. 10+ deploys per day at Flickr. Infrastructure becomes code.
Autonomous squads own features end-to-end. Tribes, chapters, guilds for cross-cutting alignment. The model everyone copied, including Spotify eventually moving on from it.
PM + Designer + Engineers solve problems, not build features. Accountable for outcomes, not output. Cagan's SVPG model becomes the gold standard for product organizations.
AI coding assistants dramatically increase scope per person. GitHub reports nearly half of new code written with Copilot among adopters. Team sizes start compressing as individual leverage skyrockets.
The trend is clear: smaller teams, bigger scope, every decade. But AI doesn't just shrink teams. It changes how they're structured entirely.
Research from the world's leading product thinkers and organizations
Thanks to AI-powered tools, the cognitive load on teams decreases, which means the scope of responsibility for a single team can increase dramatically. Fewer teams with smaller sizes, but greater autonomy and impact.
"The majority of people that believe all these roles are about to be automated are just telling me that they are not yet aware of what they don't know."
In early startups, something magical happens: teams ship more with fewer people. As the company grows, that magic disappears. AI can maintain the magic of early-stage startups even at scale.
"In the age of AI, taste will become more important than skills as much of skill-based work is offloaded to compute." — Scott Belsky, former CPO of Adobe
PwC introduces the "AI Studio", a centralized hub with reusable tech components, evaluation frameworks, a sandbox for testing, and deployment protocols.
"The grassroots approach may generate impressive adoption numbers but seldom produces meaningful business outcomes."
Designers, marketers, and PMs can contribute directly to code through AI. Three essential skills: clear product intent, effective AI coaching, and quick problem-solving.
"Deep understanding of software systems remains valuable, but output quality depends on the taste of the person directing the AI."
Hybrid units where human members collaborate with AI copilots and embedded agents. Liquid Workflows dynamically redistribute work. Cognitive Mesh shares intelligence across squads.
CEO aggressively promoted AI as a replacement: workforce dropped from 5,500 to 3,400. An AI chatbot reportedly replaced 700 customer service agents.
Six months later: customer satisfaction dropped dramatically. Siemiatkowski admitted: "We went too far."
Lesson for Trezor: Change how people work with AI. Augmentation, not substitution.
An AI-first design system must support adaptability, context, and memory. We're moving from fixed layouts to dynamic spaces.
AI is moving from individual usage to team and workflow orchestration. From passive assistant to active collaborator.
The experimental phase is over. Discovery and delivery have merged into a single continuous loop.
Hub-based architecture for AI-native teams
A hybrid model that preserves empowered teams, adds shared AI hubs, and enables role convergence toward the Orchestrator.
Repeated in every team. Siloed knowledge.
Today, each domain team has its own PM, designer, and tech lead with engineers. Knowledge and patterns stay siloed. Each team reinvents the wheel.
Design systems exist, but they're passive libraries. Research is fragmented across teams. Copy is written ad hoc, per feature. There's no shared intelligence, and when someone leaves, their knowledge walks out the door.
The Orchestrator doesn't build from scratch. They query the hubs. Not all six hubs on every project. Security is a mandatory checkpoint for everything. Experimentation kicks in during testing. Localisation when scaling to new markets.
Each Orchestrator leads a smaller team (+ Tech Lead + 1–2 engineers) but has far greater scope and impact because the hubs eliminate the grunt work.
Hubs aren't self-running. Each is owned and maintained by a domain expert:
Domain teams don't disappear. They become smaller, faster, and more autonomous. And because each team needs fewer people, we can have more teams covering more domains than we do today.
AI makes everything possible. Hubs make sure we only build what's validated.
Without centralized design governance, consistent look & feel is impossible as teams scale. We cannot allow each team to use whatever they want. The hub guarantees quality and coherence across every product surface.
One communication style, everywhere. When a flow changes, the hub flags every place that needs updating: FAQ, tooltips, release notes, help center. Nothing gets forgotten. Standardization across all channels and languages.
Just because AI makes something easy doesn't mean we should do it. The Orchestrator must validate whether an effort is worth pursuing before they start. Focus on what matters and know why. Every learning feeds back into the hub for the next team.
Trezor builds a security product. Security isn't nice-to-have, it's core. One UX error can cost users money. This hub knows all security standards, regulations (MiCA, crypto legislation), threat models and best practices.
What growth teams do manually today, orchestrated across all teams. Knows what experiments are running, where they overlap, and when statistical significance is reached. Results automatically enrich the Insights Hub.
Global users, inherently education-heavy product. Not just translation, but cultural adaptation and contextual intelligence. A Japanese user needs different onboarding than a German one, beyond just language.
Could be a standalone hub, but makes more sense as an extension of Design Hub across the full stack: UI components + API patterns + coding standards readable by AI.
Overlap between Design Hub (visual identity) and Content Hub (tone of voice). Brand guidelines are a shared layer both hubs respect.
Analytical infrastructure, not a standalone hub for Orchestrators. The Insights Hub consumes data infrastructure under the hood.
Each domain team has an Orchestrator who queries the hubs. Security is a mandatory checkpoint for every project.
A new kind of product leader
The Orchestrator combines PM, designer, and technical leader. Not a solopreneur, but a leader of a smaller team who can:
"In the age of AI, taste will become more important than skills as much of skill-based work is offloaded to compute." — Scott Belsky & Guillermo Rauch
When looking for Orchestrators, prioritize product sense and design taste, not the strongest technical profiles.
The Orchestrator isn't a theoretical construct. Companies are hiring for this role right now. They just don't agree on what to call it yet.
From domain teams to hub-augmented teams
"Your keys. Your crypto. Let's get you set up."
"These 12 words are the only way to recover your wallet. Write them down. We can't do it for you."
"You're all set. Your Trezor is ready to protect your assets."
"Something didn't work. Don't worry, your funds are safe. Let's try again."
All results fed back into Insights Hub. Available to every team for the next project.
Product strategy, design sense, technical literacy, prompt engineering, outcome ownership
Primary user of all three hubs.
Architecture, code review, AI-augmented development, system integration
Consumes Design Hub for implementation.
DS, content pipelines, data infrastructure
Own hubs as a product.
Design governance, brand integrity, AI tool training, systems thinking
Owns the Design Hub.
User research methodology, data analysis, synthetic research
Owns the Insights Hub.
A phased approach: gradual, measured, reversible
AI may produce generic outputs.
Design Lead as quality guardian. Human-in-the-loop. Guardrails in Design Hub.
Orchestrators may become "mile wide, inch deep."
Preserve Hub Leads. Continuous learning. The tool is not the skill.
People may feel threatened by changing roles.
Transparent communication. Training as investment. Gradual pilot rollout.
Hubs can't keep up with demand.
Dedicated hub teams. Self-service access. SLAs on responses.
Overly aggressive AI replacement.
Augmentation, not substitution. Measure quality, not just speed.
Hallucinations, bias, inconsistency.
Automated validation. Red teaming. Feedback loops into hubs.
Consensus across all research sources
Gartner, SVPG, and Reforge all agree.
PwC, Gartner, and Reforge confirm from different angles.
The Orchestrator emerges, but deep expertise remains critical.
Skills will be commoditized by AI. Product sense defines winners.
From style guides to living, AI-enforced systems.
Klarna showed the opposite. Change how people work with AI.
Trezor has the opportunity to be an early mover in this transformation.
The hub-based architecture model is in direct alignment with trends documented by PwC, Gartner, SVPG, Reforge, and others.
Daniel Valčík · March 2026