Want to know if this job is worth applying to?
Praha, Česko
Hybrid
Not Available
FTMO is the prop firm traders trust. OANDA is a regulated global broker, now part of the FTMO group. Both run on software that has to be fast, correct and compliant, and we want AI-first delivery to be how that software gets built.
This role sits in the each entity AI Team, a small senior team reporting to the AI Lead, and serves either FTMO or OANDA, engineering and business alike. It works closely with the team leads, product teams, infrastructure and security to turn AI-native delivery from a set of experiments into the default way the we ship in our companies.
AI moves faster than any roadmap, so most of what this team takes on has no playbook. Nobody has solved it before, at least not here and often not anywhere. The problem defines the work, not the job title. You learn the domain fast, find the real problem, ship something that works, and hand it over so the team owns the result.
The role also supports non-engineering teams that build their own apps with AI, helping them ship safely within the group's governance policies.
The work moves between these four streams, often within the same week:
The AI development harness. Designs and owns the company's harness: initializer and coding-agent patterns, AGENTS.md and CLAUDE.md conventions, progress and checkpoint files, feature lists, and end-to-end test loops that let agents make reliable progress across long-running sessions and many context windows.
The AI software factory. Builds and runs the factory that teams across the company use to build with agents.
The AI-native SDLC. Maps each phase (plan, design, build, test, review, document, deploy) to a clear delegate / review / own model, and wires the tooling that makes it real: MCP integrations, eval loops and CI hooks.
Agent-first delivery. Ships production-grade software mainly by orchestrating coding agents, keeps the codebase clean and mergeable, and sets the example for other engineers.
AI infrastructure. Helps build and harden model access and routing, sandboxes, secrets and permissioning, observability, and cost controls for agent runs.
AI ambassador in new teams. Joins a team that has barely used AI, learns how it works, finds the highest-value problem, ships a first win with them and leaves them able to carry on alone.
Guardrails and fluency. Sets the conventions other engineers' agents follow, and raises AI fluency through patterns, internal documentation and hands-on pairing.
Citizen-built apps. Supports non-engineering teams in shipping their apps and governs them under the group's governance policies.
Unsolved problems anywhere in the company. Takes problems from any department (finance, accounting, compliance, support, operations), learns the domain from zero and automates or rebuilds the process with agents. Most of these have no existing playbook.
Fast proofs of concept. Tests new models, agent patterns and tools against real codebases and workflows, then turns the ones that hold up into production-ready patterns.
The frontier. Tracks model and tooling advances, evaluates them, and brings the ones that earn their place into the stack.
Staff-level scope. Picks the problems worth solving with the AI Lead, makes architecture calls that outlive the project, and raises the bar for engineers across the company without managing them.
Impact and reporting. With the team leads and the AI Lead, prioritizes where AI delivers the most business outcome, agrees the metrics, and reports adoption and impact honestly, including where it does not yet work.
You're a senior/staff-level engineer who has already made the switch to AI-first: you ship production-grade software mainly by directing agents and can show the harness that makes it work.
You get energy from problems nobody has solved yet, and are as happy learning an accounting workflow from zero as tuning an agent harness. You move between the two without needing a spec. We care about what you've shipped and owned, not titles or years on a CV. A candidate with the right certifications and a thin portfolio loses to one who can show three things they have shipped this way.
Proven record of shipping and owning production software across the full lifecycle.
Staff-level scope: you've led cross-team technical work with no playbook, set a direction others followed, and made decisions that held up.
You learn unfamiliar domains fast: has walked into a business area they knew nothing about and shipped something its people actually use.
You thrive on ambiguity and context switching: frames the problem and drives it to done without a spec or a product manager.
Strong general engineering: you read, design and review production code in at least one major stack, and reason about architecture, testing and security even when an agent writes the lines.
Solid knowledge of IT system architectures and principles, and current with AI architecture patterns.
Understanding of where AI fails and how agents can introduce security vulnerabilities into the SDLC.
Previous AI-first role, with recent work where you wrote little to no code by hand and still shipped top-grade results.
Deep, practical fluency with agentic coding tools and harnesses (Claude Code and the Claude Agent SDK, Codex, Cursor or equivalents): context management, compaction, long-running sessions and multi-agent patterns.
You've built agent scaffolding: AGENTS.md and CLAUDE.md conventions, init scripts, progress and checkpoint files, feature lists, and end-to-end test loops.
MCP and tool integration: you wire agents safely to internal systems, APIs, repositories, logging and CI.
Evals and quality: you design eval suites and verification loops so agent output is measured against tests, latency and style, not vibes.
AI infrastructure: model routing and access, sandboxes, permissioning and secrets, observability and cost control for agent runs.
Prompt and context engineering as an everyday discipline.
Ownership, teaching and influence: you take problems from proof of concept to production, and get other engineers to adopt new ways of working without a mandate.
Clear communication: you can explain a delegate / review / own boundary to a skeptical senior engineer and a team lead, and convince both.
Pragmatism: you know what to delegate to an agent, what to review and what to own.
English language at C1 level or above.
Experience in fintech, trading or another regulated, high-stakes environment.
Public contributions: open-source agent tooling, write-ups or talks on AI-native development.
Has built custom agents, MCP servers or automated evaluation of generated code.
Has supported or governed apps built by non-engineers.
Czech or Polish language.
Experience in Google Cloud Platform
We are a Czech fintech that, since 2015, has grown from an idea into a global project. 🚀
300+ amazing teammates. We’re a great team who learn from each other every day.🤜🤛
How do we work? We focus on meaningful work and open communication, while only adopting processes that make our lives easier.
Prague, Národní třída. Enjoy our modern offices at the Quadrio shopping center, offering beautiful views and excellent accessibility.
What if I don’t trade? No worries. We’ll show you what our product is all about and introduce you to the basics of trading.
Free fruit, snacks, and coffee are always within reach in the office.
How do we promote strong relationships and well-being? Company cottage, team building events, and running club.
Flexible hybrid model. We prefer collaborating in person to keep the team spirit high, but we offer the flexibility you need to stay balanced.
*The benefits mentioned above apply to FTMO on-site employees in our Prague office.
