Cursor's global Forward Deployment Engineering (FDE) team addresses the challenge of helping enterprises adopt and maximize value from their AI coding platform. The presentation outlines a strategic framework for building FDE teams based on customer digital maturity and product customization levels, emphasizing hiring senior engineers with both technical depth and business acumen. The solution involves project-based engagements focused on co-developing high-impact applications like long-running agents and automation systems directly within customer codebases, while maintaining tight feedback loops to product teams. Results include successful deployments across multiple industries including finance, healthcare, retail, and telecommunications, with measurable ROI through increased revenue, decreased costs, or mitigated risks, while simultaneously informing product roadmap decisions and expanding platform use cases beyond traditional software development.
This case study presents Cursor’s approach to building and scaling a Forward Deployment Engineering function to support enterprise adoption of their AI coding platform. Pauline Brane, who leads the global FDE team at Cursor, shares insights from 10 years of AI enterprise deployments about how to structure, staff, and operate an FDE organization that bridges the gap between cutting-edge AI technology and production enterprise environments. The presentation is particularly valuable for understanding how to operationalize LLM-based products in diverse enterprise contexts while maintaining product velocity and customer satisfaction.
Cursor is an AI coding platform that offers capabilities ranging from autonomous coding to asynchronous and synchronous agents, ultimately aiming to enable what they call an “AI software factory.” The FDE team serves as the critical interface between Cursor’s rapidly evolving product capabilities and enterprise customers who are at varying stages of digital transformation and technical maturity.
The core strategic insight presented involves a two-dimensional framework for deciding when and how to deploy FDE resources. The first dimension assesses customer digital maturity—essentially where organizations are in their transformation journey and how technically advanced their teams are. The second dimension evaluates product customization requirements—whether the product can be deployed self-service or requires significant configuration and integration work.
This framework yields four quadrants with different deployment strategies. For high-maturity customers with low-customization needs, the recommendation is purely self-service delivery with strong documentation, as these customers can effectively adopt the technology independently. For low-maturity customers with low-customization products, traditional SaaS deployment models suffice, following more waterfall-style implementation patterns. The FDE sweet spot emerges in two scenarios: first, with high-maturity customers requiring highly customizable implementations, where FDEs act as advisors to accelerate adoption and push platform boundaries; second, with lower-maturity customers needing highly customized solutions, where FDEs provide embedded transformation support.
This strategic clarity helps avoid common pitfalls where highly skilled engineers are misallocated to activities like basic training sessions, bug documentation, or generic solution architecture that don’t leverage their unique capabilities. The framework explicitly warns against using FDEs for product orientation workshops or standard professional services that could be delivered through other channels.
The FDE role at Cursor is defined as requiring both exceptional technical depth and high emotional intelligence. These professionals must operate across organizational hierarchies, engaging effectively with C-suite executives like CIOs, CTOs, and COOs while simultaneously working hands-on with developers and engineering managers. This breadth of engagement requires the ability to rapidly lead discovery sessions, identify appropriate use cases, understand existing processes and organizational culture, and guide customers through transformation journeys that involve significant change management.
A critical aspect emphasized is that FDEs must remain at the cutting edge of technological advancement. Given the rapid pace of innovation in the AI space, with new releases and capabilities emerging weekly, FDEs need inherent curiosity and genuine enthusiasm for continuous learning. Customers expect FDEs to be domain experts, making this continuous upskilling non-negotiable.
The FDE also serves as a crucial feedback conduit between customers and internal product and engineering teams. Because FDEs are embedded within customer organizations, they develop the deepest understanding of actual usage patterns, pain points, and feature requests. This positioning makes them invaluable for informing product roadmap decisions and identifying which capabilities should be prioritized for development.
Cursor’s FDE organization is tailored to their specific customer profile—highly technical, mature organizations with sophisticated buyers and users. The team operates on a project-based model focused on high-impact initiatives that are strategic objectives for customer organizations and promise meaningful ROI.
The hiring profile is deliberately senior, targeting engineers with five or more years of software engineering experience. The initial team consists of what the presentation calls “unicorns”—individuals combining deep technical expertise with customer-facing experience. Team members have been recruited from organizations like Spotify, Rippling, and Palantir, bringing proven ability to work in high-performance technical environments.
The organizational structure uses a matrix approach designed for flexibility and rapid pivoting. Currently organized geographically, the team anticipates evolving toward industry-based organization as they mature. This evolution is driven by the recognition that industry-specific knowledge and terminology are critical for credibility. When working with banks, for example, fluency in payment systems, asset management, and risk concepts is essential for effective engagement.
The team also develops product area specialization, with individual members becoming subject matter experts in specific Cursor capabilities like long-running cloud agents or the Cursor SDK. This specialization allows team members to be pulled into projects where their expertise is particularly relevant, creating a dynamic resource allocation model.
An important philosophical point emphasized is that the team structure, roles, and hiring profiles will continuously evolve. The team operates with the explicit understanding that what they’re doing currently will be different in six months, as products evolve and customer needs shift.
Cursor’s FDE engagements follow a deliberate model designed to maximize impact while avoiding common pitfalls. Projects are scoped at the strategic level, typically engaging with economic buyers or senior champions who can ensure proper resourcing and organizational support. This top-down sponsorship is essential because FDE work requires significant customer commitment—providing system access, allocating internal team members to work alongside FDEs, and supporting organizational change.
The team explicitly avoids staff augmentation scenarios, which are viewed as misuse of FDE talent. When customers indicate they’re simply understaffed and need bodies to fill gaps, this raises red flags. Instead, the focus is on collaborative development where customer teams are actively engaged partners. A practical validation technique is simply asking who from the customer organization will be on the working team—if there’s no clear answer, it’s likely not an appropriate FDE engagement.
Project scoping is kept deliberately directional rather than rigidly defined. Rather than committing to deliver precisely specified features within fixed timelines, engagements are framed around solving specific problems using particular approaches over estimated timeframes. For example, an engagement might be framed as automating a process from start to finish using long-running agents over approximately six weeks, doing as much as possible within that timeframe. This flexibility is pragmatic—FDEs haven’t yet seen the customer’s actual processes, data, or systems in detail, so rigid commitments risk being unrealistic. Additionally, customers often want to pivot based on learnings that emerge during the engagement, and directional scoping accommodates this adaptive approach.
The FDE team delivers several categories of implementations. Core work involves deploying cloud agents and long-running agents that can operate autonomously over extended periods. These agents are designed to automate substantial processes, often reducing time requirements dramatically—the presentation mentions examples of reducing processes from three hours to twenty minutes.
The team also builds custom applications on top of the Cursor SDK, extending the platform’s capabilities to meet specific customer needs. This development work happens directly in customer codebases, requiring deep integration with existing systems and workflows.
An exciting aspect of the FDE work is pushing edge cases beyond the software development lifecycle. While Cursor’s primary value proposition centers on AI-assisted coding, FDE projects explore applications across HR, finance, supply chain, and e-commerce functions. Work with retailers, financial institutions, and other sectors investigates how the platform can enable better asset management, call center automation, and other non-traditional use cases. These explorations both serve immediate customer needs and inform Cursor’s understanding of platform extensibility.
Specific examples mentioned include automating call center ticketing systems, claims management processes, and equipment maintenance dispatch decisions. In each case, the focus is on agents that can ingest data, make decisions, involve humans in appropriate feedback loops, and drive measurable business outcomes like reduced mean time to resolution or faster customer response times.
A consistent theme is the importance of defining and measuring success explicitly. Engagements begin with clear success criteria—if a specific outcome is achieved, will the customer consider the project successful? This upfront alignment prevents misunderstandings and provides objective evaluation criteria.
The ROI framework is deliberately simple, organized around three fundamental business drivers: increasing revenue, decreasing costs, or mitigating risks. Every engagement should clearly target at least one of these outcomes, with many addressing multiple categories. This simplicity cuts through organizational complexity to focus on what actually matters to businesses.
Measurement happens throughout the engagement lifecycle. Baselines are established for current process performance, results are validated with human-in-the-loop feedback, and final outcomes are compared against initial baselines. The presentation emphasizes over-communicating these results, as customers may not naturally frame outcomes in ROI terms even when substantial value is being created. An example illustrates this: an agent costing $2,000 per day initially seemed expensive until reframed against the value of sending the correct person to fix equipment, which far exceeded the daily cost.
The FDE strategy explicitly incorporates partnerships with system integrators and consulting firms. These partnerships serve multiple purposes. First, partners typically have existing customer relationships and deep organizational knowledge that accelerates adoption. Second, partners can handle work that doesn’t require FDE-level capabilities—the presentation specifically mentions change management as an example of valuable work that partners may be better positioned to deliver.
Partners also enable scale. Once Cursor’s FDE team has established proven patterns for particular industries or use cases—telecom deployments or healthcare life sciences implementations, for example—partners can replicate these patterns across broader customer bases without requiring direct FDE involvement. This leverage effect allows Cursor to expand their reach while focusing FDE resources on high-complexity, high-value engagements.
Several operational principles guide the FDE function. A core principle is rapid learning and pivoting—rather than extensive upfront planning, the team favors trying projects with customers, learning from failures, and quickly adjusting. This experimental mindset is presented as more valuable than prolonged planning cycles.
Active customer listening drives service evolution. The presentation describes how repeated customer requests about organizational transformation—questions about hiring profiles, job descriptions, team structures, and process changes needed to capture AI value—led to developing new service offerings that weren’t initially planned. This responsiveness ensures the FDE function remains aligned with actual customer needs rather than predetermined assumptions.
The team maintains disciplined boundaries about appropriate work. Being willing to say no—to acknowledge when Cursor isn’t the right tool for a particular problem—builds credibility as a trusted advisor. This honesty leads to more productive long-term relationships where customers bring additional use cases because they trust the FDE team’s judgment.
Customer involvement is mandated throughout every engagement phase: scoping, solution design, implementation, validation, and ROI identification. The principle is that customers should own the journey with FDE support, not that FDEs should work in isolation. If an FDE finds themselves working alone in a cubicle without customer interaction, something has gone fundamentally wrong.
Documentation and artifact creation are essential at engagement conclusion. Leaving comprehensive documentation enables customers to maintain and extend work after FDE departure, supporting sustainability and ongoing value realization.
Significant emphasis is placed on attracting, compensating, and retaining top talent. The presentation invokes the principle that A players hire A players while B players hire C players, stressing the importance of maintaining hiring standards. Given the FDE role’s demands—deep technical expertise, customer skills, continuous learning, and high-impact work—retention requires ensuring team members are actually doing FDE work, not being misallocated to less challenging activities.
The team mission statement serves as both external communication and internal alignment: “We partner with your organization to co-design and co-build your AI software factory. We transform how you design, develop, and maintain software across your entire life cycle.” This clarity helps team members identify when proposed work doesn’t align with their core mission, such as staff augmentation requests.
Creating a culture where people want to do their best work requires giving them autonomy, challenging problems, and recognition. The presentation acknowledges that the FDE function is imperfect and evolving, emphasizing learning from mistakes as part of the cultural fabric.
From an LLMOps perspective, this case study illuminates several critical dimensions of operationalizing LLM-based products in enterprise environments. The FDE function essentially serves as a production deployment and optimization layer, bridging the gap between platform capabilities and real-world enterprise constraints.
The emphasis on customer digital maturity and product customization as determining factors for deployment strategy reflects the reality that LLM applications often require significant contextual adaptation. Unlike traditional SaaS products, AI coding platforms interact with unique codebases, processes, and organizational cultures, making one-size-fits-all deployment impractical for many high-value use cases.
The focus on measurable ROI and business outcomes rather than technical metrics represents mature LLMOps thinking. While technical performance matters, production success ultimately depends on driving revenue, reducing costs, or mitigating risks. The FDE function ensures this business alignment throughout deployment.
The feedback loop from FDEs to product teams represents a critical LLMOps pattern. FDEs encounter edge cases, integration challenges, and feature gaps before they become widespread issues, allowing proactive product evolution. This tight coupling between deployment experience and product development accelerates the platform’s ability to meet real-world requirements.
The recognition that organizational change management is inseparable from technical deployment reflects sophisticated understanding of LLM adoption barriers. Technology alone doesn’t drive adoption—people must understand, trust, and integrate new capabilities into their workflows. The FDE role explicitly encompasses this change management dimension.
The adaptive, experimental approach to engagements—directional scoping, willingness to pivot, rapid learning from failures—reflects appropriate uncertainty management given LLM behavior complexity and organizational variability. Rather than pretending deployment outcomes are perfectly predictable, the model embraces learning and adjustment as intrinsic to the process.
Finally, the push toward industry specialization and the exploration of use cases beyond core software development illustrate how production LLM systems evolve through deployment experience. Understanding which patterns work in which contexts, developing industry-specific expertise, and identifying novel applications all emerge from sustained engagement with diverse production environments—precisely what the FDE function enables.
Cursor, a developer tool company, shares their journey of building what they call a "software factory" where AI agents handle increasingly autonomous software development tasks. The presentation outlines how they progressed through levels of autonomy from basic autocomplete to spawning hundreds of agents working asynchronously across their codebase. Their solution involves establishing guardrails through rules that emerge dynamically, creating verifiable systems with automated testing, and building skills and integrations that enable agents to work independently. Results include engineers managing fleets of agents rather than writing code directly, with some features being developed entirely by agents from feature flagging through testing to deployment, though significant work remains in observability, orchestration, and preventing agents from going off-track.
Cursor's Forward Deployed Engineering (FDE) team helps enterprise organizations implement AI coding agents across the entire software development lifecycle, creating what they call "AI software factories." The problem they address is that while individual early adopters (10-20% of employees) successfully use AI coding assistants, organizations struggle to scale agent adoption across teams and processes. Cursor's solution involves deploying software engineers with 5+ years of experience directly into customer environments to configure and customize AI agents that work across planning, design, development, testing, review, and deployment stages. The team is experiencing rapid growth, planning to expand tenfold by December 2026, and works across industries including financial services, telecommunications, software development, technology, and semiconductors to transform how entire organizations build and maintain software.
Cursor replaced a complex git worktrees feature consisting of approximately 15,000 lines of code with a markdown-based skill implementation of roughly 40 lines. The original feature enabled parallel agent work across isolated git checkouts with sophisticated management, judging, and cleanup systems. By leveraging two existing primitives—agent skills and sub-agents—the team reimplemented both the worktree and best-of-n features using primarily prompt engineering. While the new approach significantly reduced maintenance burden and enabled new capabilities like multi-repo support and mid-chat switching, it introduced challenges around model reliability in staying within designated worktrees, particularly for smaller models and longer sessions. The team is addressing these limitations through evaluation frameworks, reinforcement learning improvements, and continued prompt refinement.
