After enterprises integrate AI into existing workflows, someone still needs to be responsible for rollout, maintenance and handover. Atlassian’s publicly disclosed FDE model provides a researchable organizational example for these responsibilities. This article treats it as a delivery reference for post‑investment engineering teams; the event itself does not involve a controlling‑interest transaction.
Fact: 12‑week project co‑delivered with customer engineers
- October 7, 2026, Atlassian announced the FDE program: engineers enter the customer environment, with a typical collaboration period of 12 weeks, while the customer also deploys its own engineers. The service includes custom Rovo agents, system integration and workflow automation.
- The company disclosed that the Expedia project has archived 600,000 pages of Confluence content, saving approximately 3,200 hours per year; this is based on the company’s disclosure scope and has not been audited financially.
- The same day, the official blog disclosed that the team already has 46 engineers, has served more than 100 enterprise customers, and has deployed over 80 AI agents into production. The company states that on‑site code, connectors and product improvements will be fed back into the product ecosystem.
Assessment: post‑investment teams should incorporate handover capability into acceptance criteria
FDEPE assessment finds that the most valuable aspect of this model is embedding customer personnel into the build process. For transformation teams of controlling‑interest enterprises, the project’s endpoint should encompass both business usability and organizational takeover: business leaders must be able to explain the outcomes, internal engineers must be able to maintain the connectors, and someone must be able to disable automated actions and revert to manual processes when anomalies occur. Otherwise, once the on‑site team departs, a short‑term rollout can easily turn into a long‑term outsourcing dependency.
It is recommended to split a project into three acceptance questions. First, whether the process is truly replaced: record total task volume, proportion of manual intervention, rework rate, and processing time, avoiding counting only call frequency. Second, whether improvements cover the full cost: include internal engineers, external on‑site staff, model calls, reviews, and maintenance in project costs. Third, whether the client can operate independently: have the client team complete a version upgrade and fault recovery to test if the handover documentation is sufficient. The above evaluation methods are proposed in this article and do not represent metrics adopted by Atlassian.
Saving labor hours still requires financial conversion. If employees use the freed time to handle more backlog while wages and outsourcing expenses remain unchanged, the benefit first appears as increased capacity and service quality. Only after linking to outcomes such as new revenue, outsourcing elimination, reduced overtime, or avoided hiring does it become appropriate to include in profit‑improvement calculations. The saved hours should not be simply multiplied by salary and treated as realized cash benefit.
Platform reuse also needs separate acceptance. When the next client adopts the same capability, if data still needs to be re‑cleaned, permissions rewritten, and workflow approvals redone, delivery costs may not show a clear reduction. Post‑investment engineering teams can track person‑days for the second deployment, the proportion of components directly reusable, and client‑side maintenance hours to assess whether experience translates into replicable assets.
Counter‑evidence: the number of production deployments does not equal replicable profit.
- Public materials do not disclose each project's fees, full engineer costs, control groups, or cash benefits, making it impossible to calculate the project's net return based on that information.
- The content‑governance outcomes for a single client cannot be automatically extrapolated to procurement, risk control, or financial‑approval processes; each task type has different error costs and authorization boundaries.
- Reuse within a vendor product ecosystem may not be directly transferable to the conglomerate's heterogeneous systems. If the client co‑creates and occupies a large amount of internal talent, project speed may come at the expense of delays in other work.
Atlassian official blog: Meet the Forward Deployed Engineer, October 7, 2026.