More companies, especially in AI, are building Forward Deployed Engineering (FDE) teams: engineers who work inside customer environments, ship bespoke solutions fast, and bring what they learn back to the product.
Most teams I see set up Atlassian for only half of that job.
FDE runs two loops:
- Engagement loop. It’s fast, customer-specific work, measured by time to value.
- Harvest loop. It turns field learnings into reusable assets and product, measured by reuse.
A Jira board per customer handles the first loop. The second loop is where the value is, and it usually lives in someone’s head.
A simple ladder for the harvest loop:
- One-off: built for one customer
- Pattern: solved the same way twice, then documented
- Product: promoted into the core platform
The third time something shows up in the field, it should become visible as a candidate for promotion. Here’s how I’d structure the tools to make that happen.
Jira: One FDE space with a custom hierarchy: Engagement → Workstream → Task. Make Customer an Assets object rather than a text field, so you can report across engagements. Create separate spaces only where confidentiality requires them.
Jira Product Discovery: This is the bridge. FDEs capture “field signals” (gaps, friction, workarounds) as insights linked to ideas. Product can then see which ideas have evidence from one customer and which have evidence from seven.
Confluence: Each engagement gets a light page tree: brief, architecture, decision log, handoff runbook. A central Field Playbook space holds everything that has reached Pattern stage.
Compass: Catalog field-built accelerators. Track owner, adopting customers and health, and use scorecards for “productization readiness.”
Bitbucket: Keep customer repos separate from shared accelerator libraries. When customer code imports an accelerator, that counts as reuse.
Goals, Loom and Rovo: Give each engagement an outcome goal with weekly updates. Use Loom for field-to-core handoffs. Use Rovo to draft updates and spot repeated signals across customers.
Three lessons:
- FDEs push back on process, so signal capture has to take seconds.
- Measure the harvest loop: reuse rate, and how many shipped features started as field signals.
- Design for confidentiality from day one: per-customer permissions, anonymized signals in JPD.
How are others structuring this? I’m especially curious whether you use one space or a space per customer, and how you get field insights into product prioritization.