AI Engineer World's Fair 2026
Forward Deployed Engineering at Cursor — Pauline Brunet
Pauline Brunet· VP of Forward Deployed Engineering,
Cursor20:47
Read the talk
Forward Deployed Engineering at Cursor
Pauline Brunet explains how Cursor matches forward deployed engineering to customer maturity, scopes collaborative projects, and measures value that lasts beyond the engagement.
From a talk by Pauline Brunet
When does a customer need an FDE?
Should a customer get documentation, a conventional deployment team, or an engineer embedded in their organization? That decision sits behind the design of a forward deployed engineering (FDE) function. Pauline Brunet leads that function globally at Cursor, drawing on a decade of enterprise AI deployments across consulting and three technology companies. Cursor’s product direction—from autonomous coding through asynchronous and synchronous agents toward an AI software factory—provides the context for the team she is building.
Start with the customer’s transformation journey, then examine the product you are asking them to adopt. Brunet uses two axes: customer digital maturity and product customization. Does the customer have engineers and the technical capability to implement changes? Does the product work almost immediately after receiving a login, as in her example of Teams, or does it require substantial configuration and development? Those questions are more useful than treating FDE as another name for professional services or staff augmentation.
| Customer maturity | Product customization | Engagement model |
|---|---|---|
| High | Low | Self-service and documentation |
| Low | Low | Traditional SaaS deployment |
| High | High | Advisory and acceleration |
| Low | High | Embedded transformation |
A technically mature customer using a simple product should largely be able to serve themselves with good documentation. A less mature customer adopting the same kind of product may need a conventional, sequential rollout. Neither automatically warrants scarce FDE capacity. With a highly customizable product, mature customers can do substantial work themselves but benefit from advice and acceleration. Customers earlier in their transformation may need an embedded team because they cannot yet hire or staff the expertise required.
The blue FDE region on Brunet’s matrix emphasizes where deeper engineering involvement can help. Even mature customers with relatively simple products may benefit from a feature extension, and that work can create a useful feedback loop to product engineering. But the opportunity for FDE impact is smaller there than in engagements requiring substantial transformation.
For mature customers with customizable products, remember how much they can already do themselves. Brunet does not want FDE work to collapse into solution architecture, recording bugs, or updating documentation. Likewise, a traditional deployment might need a configuration extension without needing an FDE team to run product 101 and 201 workshops for the whole company. Define consequential work before hiring for it. Recruiting and compensating exceptional engineers only to assign routine work creates boredom and attrition; the role delivered no longer matches the role sold.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Engineering across organizational levels
Brunet’s founding FDE profile combines deep technical ability with high emotional intelligence. The same engineer may work with a CIO, CTO, or COO, then with developers, engineering managers, and leaders responsible for transformation or AI. They must quickly discover a useful problem, understand the existing process, and identify how technology could improve it. That discovery also includes organizational culture: who can make a change happen, and how will people adopt it?
Installing advanced technology does not ensure anyone will use it. The FDE has to accompany people through the change while remaining technically credible. With new releases arriving every week, curiosity and continuous learning are part of the job: customers expect the engineer in front of them to understand what is now possible.
Embedding engineers also gives the vendor an unusually direct view of customer needs. Recurring requests become evidence for what product and engineering should build next. The FDE must carry that feedback carefully between the customer and internal teams, translating repeated friction into product direction rather than merely collecting isolated requests.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Build something the customer will keep running
Cursor’s customers include mature organizations with deeply technical buyers and users, which shapes both its hiring and its project model. Brunet seeks an economic buyer or a senior champion who can connect the proposed work to a strategic company objective. The project needs meaningful business value, customer staff assigned to it, access to the systems and codebase where development will happen, and top-down support for changes to working practices. Sponsorship is useful because it secures the conditions for delivery, not simply because it gives the project visibility.
Brunet reports deployments of cloud agents, long-running agents, automations, and applications built on the Cursor SDK. The exit condition is practical: the result should create enough value that the customer continues using it after the FDE team leaves. A deployment that gets switched off at the end of the engagement has not achieved the durable change the team is aiming for.
The team also tests where Cursor can help beyond the software development life cycle. Brunet describes work with HR, finance, supply-chain, and e-commerce teams, including retailers and banks exploring asset management. These engagements let FDEs probe the platform’s boundaries with customers and discover whether its capabilities can improve work elsewhere in the company.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Ask who will build alongside you
Co-development requires a customer team. Cursor’s FDEs work in the customer’s codebase, but that does not make them replacement staff. A request justified primarily by being understaffed is a warning sign for Brunet. The customer should consider the project important enough to commit its own people.
Her screening question is direct: “Who will be the working team?” Suppose the proposed use case is long-running agents that automate a call-center ticketing system. Before agreeing to build, identify the people who will collaborate on it. That question turns an enthusiastic discussion about automation into a concrete test of customer commitment. The relationship also works in both directions: FDEs influence the product roadmap, then bring new capabilities back to customers as previously inaccessible use cases become feasible.
An explicit mission makes these boundaries easier to maintain. Cursor’s mission is to partner with customers to co-design and co-build their AI software factory, transforming how they design, develop, and maintain software across its life cycle. It tells customers what the engagement is intended to accomplish and gives engineers a way to assess assignments. When a project sounds like staff augmentation, the team can ask whether it advances that mission.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Start with experienced generalists, then specialize
At the time of the talk, Cursor’s founding FDE hires require five-plus years of software engineering experience, alongside customer-facing experience. Brunet is initially hiring people who can cover both sides of the role rather than graduates or early-career engineers. She expects that to change as the team grows. The eventual split could include people stronger in customer work who retain technical aptitude, and deeply technical engineers who have the aptitude to learn customer-facing work. These are plans for the team’s evolution, not a description of a split already in place.
Cursor uses a matrix organization, initially organized by geography. Brunet expects industry specialization to become more important: an engineer talking to a bank needs to understand payment systems and asset-management risk. Industry language is part of establishing credibility, because it demonstrates an understanding of the customer’s actual work.
Product expertise provides another dimension. The team has specialists in long-running cloud agents and the Cursor SDK whom colleagues can bring into projects. Brunet names Spotify, Rippling, and Palantir among the organizations her team members came from, while emphasizing their enthusiasm and aptitude for working with customers. The structure remains deliberately adaptable: as products change and customers advance in their transformation, the team’s roles and hiring profiles should change too.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Let customer learning change the offering
Brunet’s advice is to try a real customer project, learn quickly, and pivot—even when the project fails. Extended planning cannot substitute for discovering what customers actually need. She gives an example from her own offering: customers repeatedly asked how to reorganize after adopting powerful AI tools. They wanted help with hiring profiles, job descriptions, team structures, collaboration, and processes that would let them capture the tools’ value. She had not intended to offer that work, but was considering it and would hire the appropriate team to support it.
Partners can extend delivery without requiring the FDE team to do everything itself:
- Customer knowledge: Systems integrators and consultants often have long-standing relationships and already understand the organization.
- Complementary capabilities: Brunet identifies change management as work she would prefer partners to help deliver.
- Repeatable rollout: Once an established product has been deployed repeatedly, partners can broaden its reach. She points to rollout experience across telecommunications, healthcare, and life sciences.
This division preserves room for FDE work while making deployment possible at a larger scale.
The same discipline applies to refusing work. When Cursor is not the right tool, Brunet recommends saying so. Honest boundaries build credibility and can lead customers to bring a better-fitting use case later. Accepting unsuitable work also leaves the team accountable when it goes poorly. The offering, hiring profile, compensation, and assignments should fit together so that the people recruited to deliver it remain motivated.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Fix the outcome, keep implementation scope directional
An engagement begins by checking that the stated problem is the real problem. A request may describe a symptom, or come from someone who does not own the process. Ask who is responsible and speak with that person. Brunet wants access to the owner of the workflow, process, or department before deciding what to solve.
Next, agree on a success criterion. Brunet’s illustrative target is to reduce an end-to-end process from a three-hour baseline to twenty minutes, subject to the customer agreeing that this would count as success. This is a proposed outcome, not a reported deployment result. She contrasts a problem-based commitment with an open-ended allocation of two FDEs for six months: supplying people without defining what they should accomplish is, in her view, a recipe for failure.
Translate the intended outcome into a directional implementation scope:
- Name the process to automate. Define the work that should run from beginning to end.
- Describe the agent’s inputs and decisions. Specify which data it will obtain and what decisions it will make.
- Identify human participation. Establish who belongs in the feedback loop.
- Connect the workflow to a KPI. Examples include mean time to resolution or the time taken to answer a customer’s insurance claim.
This gives the project a technical shape without pretending that every detail of the customer’s systems is already understood.
Brunet illustrates directional scope with a six-week engagement covering initial phases, selected automations, and agent setup, while leaving the achievable amount of work flexible. The reason is uncertainty about the customer’s processes, data, and systems. Learning during implementation may reveal more work than expected or a better direction to pursue. Customers may also want to pivot once they see what is possible, so the scope should accommodate that learning while remaining tied to the intended outcome.
Customer participation must continue through scoping, solution design, building, implementation, human-in-the-loop validation, comparison with the baseline, and identification of ROI. The customer owns the outcome; the FDE remains hands-on. Engineers still configure systems and develop in the customer’s codebase, but they should not be doing it alone. Brunet’s warning image is an FDE isolated in a cubicle at the customer’s office. At the end, return to the agreed process-duration target and ask whether the work came close.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Measure the value, not just the agent bill
If delivery misses the target, investigate why and decide what can be changed. Then communicate the return on investment clearly. Brunet recounts an agent reported to cost $2,000 per day whose work helped send the right person to repair equipment. Her response was to ask whether the reduction in equipment-dispatch costs justified that expense. The person she spoke with agreed it could, but had not measured the benefit that way; the anecdote supplies no quantified savings or net ROI.
Brunet organizes business value into three categories:
- Increase revenue: Help the business earn more.
- Decrease costs: Reduce the expense of delivering an outcome.
- Mitigate risks: Reduce exposure to adverse outcomes.
An engagement should support at least one of these, and may support several. The operating cost of an agent becomes meaningful in relation to the business outcome it changes.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Leave artifacts and room to do good work
The engagement also needs a handoff: leave documentation artifacts that help the customer continue after the team departs. At the organizational level, build the FDE function that fits the company, then learn, pivot, and scale it. That requires a culture where people want to participate, keep learning, and have the opportunity to do their best work.
Brunet closes with the hiring maxim “A players hire A players, B players hire C players.” Her emphasis is on maintaining the quality and fit of the team as it grows: hire for the organization and its customers, and keep those people motivated with work worth doing.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Resources
From the talk
- Cursor's SDK announcement introduces programmatic agents running locally or in cloud VMs, with a TypeScript example and links to documentation.
- Explains Cursor's long-running agent harness, planning and review approach, and early coding-task results.
- Current setup and usage guidance for agents working in isolated cloud development environments.
Further reading
- A companion interview with Pauline Brunet about enterprise deployments, customer workflows, and building Cursor's FDE team.
- Scheduled and event-triggered agents, with templates for recurring engineering tasks.
Transcript & slides
Loading transcript and slides…