Airwallex Innovation & Technology Culture

What types of products or services does your engineering team build? What problem are you solving for customers?

We’re creating an AI CFO for founders and small companies with AI automation for several services ranging from bookkeeping to payroll and eventually financial planning and analysis. One of the aspects of the CFO function we’re tackling is managing a company’s day-to-day finances, also known as bookkeeping. Traditionally, founders would either handle this themselves or hire someone part-time. It involves tracking all the money flowing in (for example, customer revenue through online checkout, invoices or wires) and all the money flowing out (such as expenses for salaries, rent, utilities) to ensure the company’s finances are accurate and up to date. We’re building intelligent algorithms that can understand these financial patterns and handle the nuances of accounting much like a human accountant.

Tell us about a recent project where your team used AI as a tool. What was it meant to accomplish? How did you use AI to assist?

My team uses AI to supercharge our smaller team with Cursor and Claude Code able to generate quick prototypes in the frontend, along with making our integrations with platforms easier. Also, MCP servers have been significantly useful in speeding up integrations with different tools. I used Ramp MCP servers to pull data from live customers and then tested a variety of machine learning models with HuggingFace MCP. After creating multiple pull requests, I used Graphite agents to review and evaluate code style and examine edge cases that I may have missed.

Lastly, when building out our platform, we relied heavily on AI to help us flesh out possible algorithms with Claude research. Using AI to iterate through these algorithms allowed us to evaluate trade-offs much faster. I went through several conversations with Claude to understand edge cases of the algorithms and with each changing requirement, the assistant could recommend reasonable suggestions. The research mode also provides citations for the papers, allowing me to read each one and draw my own conclusions.

What would that project have looked like if you didn't have AI as a tool to use? 

Finance and accounting are complex and nuanced fields with lots of rules, guidelines and judgment calls that companies have to make. There’s some flexibility in how you approach things, but you still need to stay within the boundaries of what’s allowed. Building AI-powered workflows for areas like this takes a lot of research to really understand which technologies and algorithms are best suited to handle all that complexity while making sure the finances are accurate.

One of the ways that AI is unique in empowering our workflow is that it can quickly assess what algorithms and technology are out there and break it down into categorization.This saves me hours of manual research and summarization. Also, all of the context that AI can synthesize and summarize means that I can quickly assess the pros and cons of different algorithm ideas that I have proposed. As we gather more requirements after talking to our design partners, we can then use the saved context to continue to iterate on these algorithms.

Although AI has some limitations, I primarily believe it changes my workflow by offloading menial testing and condenses the research I need to do to iterate through different designs.

How does innovation show up in your company culture?

At Airwallex, innovation isn’t a department; it’s our operating system. We are redefining global banking with AI by discarding the traditional playbook. Every employee here is an AI practitioner. It’s not just engineers using coding agents. Our finance team is “vibe coding” their own applications, our talent team uses AI to streamline hiring, and our operations team is driving massive efficiency gains through automation.

In product and engineering, we’ve embraced a high-velocity “rebuild” mindset. In the AI era, building for the foreseeable future is a myth. We build knowing that as underlying models improve every few months, we may need to rebuild aspects of our product to unlock even more value for customers. This shift even extends to how we grow our team: Our engineering interviews now test for proficiency in working with AI rather than banning it. We aren’t looking for people who can memorize algorithms; we’re looking for “AI pilots” who can leverage these tools to build faster and smarter.

What’s one recent innovation that improved user or employee experience?

One of our most impactful innovations is our AI assistant embedded in the Airwallex product, designed to make opportunities obvious and finance nearly invisible. Airwallex offers a broad, powerful product surface that can be complex for new users moving from onboarding to real value.

The assistant acts as an intelligent guide. It understands your context, draws on information you share and relevant public signals, and creates a tailored journey that gets you to outcomes like funding accounts, issuing cards or scheduling international payments much faster. It doesn’t just answer questions. Behind the scenes, specialized agents execute complex tasks on your behalf, from configuring accounts to setting up cards and payments.

It sits on a context layer with access to structured and unstructured data about Airwallex, our products, the user’s business and the user’s Airwallex data. This real-time layer powers experiences across our product and the AI assistant. This is only possible because of advances in agentic architectures and foundation models, an early example of how those breakthroughs create a smoother customer experience and a foundation for the finance agents we’re building now.

How do you balance experimentation with stability?

We balance experimentation and stability by being very deliberate about how and where we take risks. Experiments start with clear hypotheses, success metrics and a bias toward running many small, fast tests rather than a few large ones. We learn quickly, kill what doesn’t work, and harden the ideas that do into our core product.

Because we’re a financial platform, stability is non‑negotiable, especially in our core money‑movement and risk systems. Most new ideas — especially AI‑driven ones — start behind feature flags and in constrained surfaces like onboarding or decision support, with offline and online evaluations, strong guardrails and close monitoring. Before promoting an AI experiment into production, we run automated evaluations, red‑team style testing, and monitor for regressions on key metrics so we can catch issues early and roll back safely. This lets teams move fast at the edges of the product while intentionally keeping the underlying financial rails stable.

Agentic coding has rapidly evolved from experimental novelty to essential infrastructure. At Airwallex, we’ve been building AirDev — our internal platform for AI-powered software development — and we’re ready to share what we’ve learned.

This post covers what AirDev is, why we built it, and how we think about AI-assisted development. A follow-up post will dive deeper into the technical architecture.

Why build custom coding agents?

Airwallex powers global payments for businesses worldwide. Our engineering teams maintain a large number of services spanning payment processing, treasury management, compliance systems, and global infrastructure.

Like many fast-growing engineering organizations we faced the friction of high-stakes monotony from tasks that demand perfect execution but offer no creative problem-solving.

Examples include propagation of configuration changes across multiple services, creation of new API endpoints based on existing conventions and replication of infrastructure updates across environments.

While these tasks are critical, they are also repetitive which makes them perfect candidates for automation.

Off-the-shelf coding assistants help with individual code generation, but they don’t understand our specific patterns, deployment pipelines, or the full context of why a change is needed. We wanted agents that could operate end-to-end: receive a task, explore the codebase, implement a solution, write tests, and open a merge request. All without human intervention.

What is AirDev?

AirDev is our platform for running fully autonomous coding agents. Engineers create a task describing what they need, and AirDev agents handle the rest.

The core workflow:

1. Task ingestion: Tasks flow in from our project management tools with full business context
2. Repository discovery: Agents search for and identify repositories relevant to the task
3. Codebase exploration: Agents clone repositories, analyze existing patterns, and identify relevant files
4. Task decomposition: For complex tasks, agents break down the work into multiple steps — planning first, then execution — to ensure each step addresses the core problem
5. Implementation: Agents write code, tests, and configuration following repository conventions
6. Review pipeline: Changes go through our standard merge request process with human review
7. Status tracking: Full audit trail of what was changed and why

The key insight: agents operate within the same workflows as human engineers. They follow our commit conventions, branch naming standards, and code review requirements. The merge request from an agent looks like any other MR, because it goes through the same process.

The impact

The value of AirDev extends far beyond code output. It’s changing how we operate as an engineering organization.

For engineers, it means reclaiming time. Tasks that would typically take several hours — understanding a codebase, implementing changes across multiple services, writing tests, opening merge requests — now complete autonomously. Engineers can redirect that time toward higher-leverage work: system design, complex problem-solving, and mentoring. The cumulative effect adds up to thousands of engineering hours returned to work that genuinely requires human creativity and judgment.

For the business, the implications are transformative. Features that once required weeks of engineering bandwidth can now move from idea to production in days. Requirements that would have queued behind other priorities get addressed immediately. Infrastructure improvements that teams “never had time for” are now happening continuously.

This isn’t about doing the same things faster, it’s about unlocking capacity that didn’t exist before. Teams can pursue initiatives that previously would have been deprioritized. Product velocity increases without increasing headcount. Technical debt gets addressed in parallel with feature development rather than waiting for dedicated cleanup sprints.

Perhaps most importantly, AirDev changes the economics of software development. The marginal cost of well-defined engineering tasks approaches zero. This fundamentally shifts what’s possible: improvements that weren’t worth the engineering investment suddenly become viable.

We’re still early, but the trajectory is clear. AI agents won’t replace engineers, they’ll amplify what engineering teams can accomplish.

What kinds of tasks work well?

We’ve found that agents excel at certain categories of work:

• Configuration changes: Updating settings across multiple services consistently
• API development: Adding new endpoints that follow established patterns
• Infrastructure automation: Provisioning resources and updating deployment configurations
• Test coverage: Adding unit tests for existing functionality

The common thread: well-scoped tasks with clear patterns to follow. Agents learn from the existing codebase and replicate what works.

How agents work

Each AirDev agent runs in an isolated environment with access to:

• Repository tooling: Git operations, file exploration, code search, and language-specific tooling
• Integration APIs: Project management, documentation, version control systems, and CI/CD pipelines — agents monitor pipeline results and fix issues based on feedback

Agents are powered by large language models and operate through a structured prompt system that encodes our engineering standards. When an agent encounters a new repository, it explores the existing patterns before making changes — examining how similar features are implemented, what testing conventions are used, and how configuration is structured.

The goal is consistency: an agent’s code should be indistinguishable from code written by a human engineer familiar with the repository.

What we’ve learned

Context matters more than instructions. Agents that understand why a change is needed produce better solutions than agents following detailed specifications. Business context leads to better implementation choices.

Patterns beat documentation. Our most successful agents learn from examples in the codebase, not from written guidelines. When adding a new configuration, agents examine how existing configurations are structured and replicate the pattern.

Scope determines success. Agents excel at well-defined tasks: adding fields to data models, implementing API endpoints following existing patterns, enabling configuration options. Vague requests produce vague results.

Human review remains essential. Every merge request goes through human review. Engineers verify the approach, check test coverage, and ensure edge cases are handled. This isn’t a limitation, it’s the design. Agents are collaborators, not replacements.

What’s next

AirDev is under active development. We’re exploring multi-agent coordination for features spanning multiple services, improved context sharing across related tasks, and better feedback loops from code review.

We’re also working on the hard problems: handling ambiguous requirements, recovering gracefully from errors, and knowing when to ask for clarification rather than guessing.

Airwallex has evolved significantly over the past two years, bolstered by acquisitions, innovative new solutions — and ambitious leaders. 

Adrian, a data science director, is one such leader who has been busy driving the global payments and financial platform’s growth. Since joining the company about two years ago, he has helped build its product data science and growth data science teams from the ground up, setting the organization up for continued success. 

Adrian has witnessed the company reach major milestones, such as its recent acquisition of OpenPay, which added billing capabilities to the company’s core infrastructure and accelerated revenue workflows; a $300-million Series F investment; and an annual recurring revenue milestone of $900 million in August.

Amid all of these positive changes, Adrian has been focused on optimizing product performance, accelerating customer acquisition and more, building out strong and engaged teams rooted in a culture of ownership, autonomy, and trust.  

Adrian isn’t just committed to the business’s success; he’s committed to the people, too. 

“At Airwallex, you’ll be joining a team where you will have opportunities to drive direct business outcomes and accelerate your professional growth,” he said. 

Learn more about Adrian’s work at Airwallex, some of the company’s greatest accomplishments so far, and what the team has planned for the future. 

About Airwallex

Airwallex offers global businesses fully integrated solutions to manage everything from business accounts and payments to spend management and embedded finance. 

How Adrian Built Data Science at Airwallex

I’m a data science director based in Singapore, leading the product data science and growth data science teams, which are rapidly growing teams of 19 data scientists in both Singapore and Asia-Pacific, and we’re expanding the team in San Francisco. Over the past two years at Airwallex, I’ve built out both teams from scratch. 

The product data science team has two main focuses. The first is driving product growth by producing science-driven insights and analysis that directly impact product decisions, leading to improvements in activation, expansion, cross-sell and retention; and optimizing product performance to ensure that our products are best-in-class. A key example of the latter is Optimize 360, Airwallex’s payments optimization initiative, where we deployed contextual bandits to optimize payments acceptance success rates for our customers. This directly led to sustained uplift in payments success rates and a ten-times increase in payments volume.

Meanwhile, the growth data science team focuses on accelerating customer acquisition via A/B testing and customer insights; optimizing our marketing investments through data-driven multi-touch attribution; media mix modeling and incrementality testing; and building an efficient lead conversion funnel with ML-driven lead routing and accurate enrichment models.

I’m also passionate about building a strong, engaged team at Airwallex and supercharging the professional growth of every team member. So, I invest a lot of effort into interviewing every candidate while maintaining a high bar for talent. Across the team, I’m also continually fostering a strong team culture of ownership, autonomy, and psychological safety. 

What Business Milestones Mark the Latest Growth Phase

Airwallex’s mission is to build the future of global banking.

Legacy systems are slow, fragmented, and costly. We’re changing that. With our own infrastructure, software, and AI, we’re creating a borderless, intelligent, real-time financial OS. We’re not riding the old rails — we’re building new ones — so entrepreneurs and modern businesses can launch and scale globally from day one. This is the future of global banking, and we’re building it now. 

Earlier this year, Airwallex completed a $300 million Series F funding round at a $6.2 billion valuation, bringing total funding to more than $1.2 billion. The company announced it’s on track to hit $1 billion annualized revenue in 2025 alongside expanded investor participation, including Visa Ventures.

In Brazil, the company received a payment institution license; meanwhile, in Mexico, the company received approval to close its MexPago acquisition, positioning it to launch in Latin America’s largest markets. The company officially launched in New Zealand earlier this year and signed an agreement to acquire CTIN Pay in Vietnam, accelerating the company’s expansion in the Asia-Pacific region. Additionally, the company opened new offices in Paris, New York, and Toronto, established first hires in the United Arab Emirates, and moved to its permanent U.S. headquarters in San Francisco.

Earlier this year, the company became Arsenal FC’s exclusive finance software partner, supporting supplier payments, spend cards, expense management and gateway services for hospitality. The company also won a hat-trick at the Asia FinTech Awards 2025, earning accolades for “Banking Tech of the Year,” “Best Employer of the Year,” and “Director of the Year.”

In July 2025, the company launched Purchase Orders to unify procurement and bill pay workflow. The company also introduced Airwallex Pay, which offers instant B2B transfers within the network, scheduled conversions that lock foreign exchange rates for up to 90 days, and expanded local payment methods to include Venmo, Cash App, Pix and TWINT.

Airwallex teammates tackling a challenge

Additionally, the company enabled Discover’s Diners Club acceptance across multiple countries and launched Apple Pay/Google Pay and physical cards in Canada and New Zealand. It also scaled its Yield investment product beyond Australia, launching in Hong Kong and Singapore with more than $350 million funds under management, and prepared for rollouts in Europe, the United Kingdom and the United States.

How the AI-First Vision Shapes Products and Careers

We’re further scaling and improving contextual bandits as a part of Optimize 360 to increase customer coverage with a higher level of performance and are developing AI agents that autonomously discover and propose optimization strategies for each product. We’re also further optimizing our marketing investments with a full suite of causal inference and machine learning approaches. We plan to develop accurate lifetime value estimation models to enable smart targeting of high-potential prospects.

Proof of Scale: Revenue and Client Growth

• $720 million in annualized revenue and $130 billion in annualized payments volume as of March 2025, with expectations to reach $1 billion annual recurring revenue in 2025
• $900 million in ARR and $200 billion in annualized transaction volume as of August 2025, with 13,372 new transacting clients in the second quarter and ~30 percent lower customer acquisition cost year over year
• Delivered strong APAC performance in 2024, with 83% revenue increase YoY by the third quarter and global revenue up by 73% YoY 

We’ve experienced dramatic growth in the past two years, which has translated into several growth opportunities for the team. Individuals’ scope has been quickly expanding as they deliver results, and team members are becoming more “single-threaded,” driving company-level projects from end to end, which is relatively uncommon for most data scientists.

Why Jobseekers Can Build Long-Term Careers Here

Deep cross‑functional collaboration with product and engineering is now the norm, so all data scientists have the full business context and leverage to directly drive company-level outcomes. High performers are getting promoted as a reflection of their sustained impact, and new roles are opening up across the entire team, so every team member has the opportunity to own green field problem areas from day one.

We employ roughly 1,900 people across 26 offices worldwide, supporting resilience and local execution, and we support more than 160 local payment methods and wallets. 

Airwallex obtained a rare Chinese payments license, becoming only the second foreign company to do so, underscoring our licensing strength in complex markets. Also, a Forrester study found that our embedded finance solution can deliver 237% ROI, faster market expansion and multi‑million savings for customers over three years. 

The Airwallex team

We operate from a clear, shared data science model that defines our mission and first principles. It keeps us aligned as we scale. We review and evolve it together, so everyone has ownership. Our operating principles map directly to the model, ensuring our day-to-day decisions reflect company values.

We’re an AI‑first company, pushing the frontiers of product optimization by combining ML and AI solutions intelligently, with Optimize 360 being our flagship example. Being on the forefront of AI, we’re evolving the data science role to maximize our effectiveness in this new paradigm.

At Airwallex, we have a wide suite of products with a lot of synergy, so there are unique opportunities to globally optimize that horizontally across products instead of locally optimizing each product independently. We’re doubling down on data scientists’ unique roles in powering product-led growth and directly influencing product decisions. 

What’s your rule for fast, safe releases — and what KPI proves it works?

Our rule is simple: Every change requires a merge request with human review. We make no exceptions, even for code by AI agents. We combine this with automated CI gates and progressive rollouts to ensure speed doesn’t compromise safety.

Our internal platform, AirDev, runs AI agents that autonomously implement well-scoped tasks. These agents submit merge requests that follow the same review and CI pipeline as human-authored code. 

Throughput of merged changes proved the system works. Since launching AirDev in late 2025, our agents have contributed over 11,000 merged MRs across 200+ repositories. Monthly volume grew from 60 MRs in the first two months to over 3,700 in a single month by March 2026. That’s roughly 120 agent-authored MRs reaching production every day, all passing the same quality gates as human code. 

We now ship previously deprioritized work continuously alongside new features. We track deployment frequency and change failure rate to ensure we ship faster without regressions. Our internal survey gave AirDev a 4.2/5 satisfaction rating, with 91 percent rating it four or five stars.

Which standard or metric defines “quality” in your stack?

Our quality principle is simple: AI-generated code must meet the exact same bar as human-written code. Every merge request follows the same path. Whether from an engineer or an AirDev agent, it must pass automated CI checks, static analysis, SonarQube quality gates and human review. If a gate fails, the change doesn’t ship.

This rigor matters because our AI agents have now contributed over 11,000 merged MRs. At that scale, you can’t afford a separate quality bar. Our engineers review every agent-authored MR to verify it follows existing patterns, handles edge cases and avoids unnecessary complexity. 

We’ve learned that context beats instructions. Agents that understand why a change is needed produce better code than those following rigid specs. That insight shapes our quality standard. We want solutions that are idiomatic and easy to maintain. 

Finally, we prioritize small, focused changes. We’d rather ship 120 focused MRs a day than a few sprawling ones. Small changes are easier to review, safer to deploy and faster to roll back.

Name one recent AI/automation shipped and its impact on the team or business.

We built AirDev to handle software tasks from start to finish. Our AU agents manage everything from reading a Jira ticket to submitting a merge request. These agents work in isolated environments with access to our full toolchain, including Git, GitLab, Jira, Confluence and CI/CD pipelines. They clone repos, analyze existing patterns, implement changes, and open MRs for human review.

Since launching in late 2025, AirDev has produced over 11,000 merged MRs across 200+ repositories, contributing roughly 440,000 lines of code. Monthly output grew 60 times, jumping from 60 merged MRs at launch to over 3,700 in March 2026. 

The impact goes beyond volume. Engineers have reclaimed about 20 engineering-years previously spent on repetitive tasks like configuration updates, boilerplate endpoints and test coverage. This frees them to focus on design and mentoring. 

The biggest surprise was the shift in priority. When the cost of a well-scoped task approaches zero, we can finally tackle work we used to ignore. We now fix technical debt alongside new features instead of pushing it off. AI boosts our capacity without lowering our standards.