Key Takeaways

  • Modal grew its forward-deployed engineering group from 4 to 24 engineers in a single year while deliberately hiring zero dedicated product managers.
  • Bernhardsson directs engineers to catch client accounts over-provisioning CPU or memory, sitting down with those teams to cut their cloud bills.
  • Hiring product managers too early at an infrastructure startup risks shipping superficial widgets instead of the slow, hard foundational building blocks required to survive.
  • Field engineers at Modal are not a permanent professional services arm; their explicit mandate is to find missing platform abstractions so core engineering can automate them away.
  • Modal manages this balance using The 3-Part Forward Deployed Engineering (FDE) Framework.

The 3-Part Forward Deployed Engineering (FDE) Framework

Objective A: Deal Closing and Sales Support

Participate directly in sales and deal closing by assisting prospective clients with migrations, validation, and hands-on technical support.

Objective B: Proactive Customer Optimization

Serve active customers by tuning model performance (e.g., draft models, inference parameters in vLLM/SGLang) and proactively discovering over-provisioned CPU and memory to lower client costs.

Objective C: Bridging Gaps into Productized Features

Act as an incubator for platform functionality, identifying repetitive custom solutions and codifying them into self-serve platform primitives: “Anything an FDE does, my question is always: long-term, how do we turn this into product so we don't need an FDE for this?”

When This Works (and When It Doesn't)

This framework works when you build complex developer infrastructure on top of fast-moving, unstable open-source tech. When clients struggle with container startup times, model serving libraries, or memory leaks, sending an experienced systems engineer into their repository closes deals faster than any sales deck. As Bernhardsson points out, “Doing product management at an infra company really well is very, very hard because it takes a very deep level of strategic thinking about foundational building blocks.” FDEs surface actual operational pain from real production systems, preventing product managers from inventing roadmap items that sound good on slides but miss technical realities.

It breaks down the moment you treat forward-deployed engineering as custom software consulting. If your sales compensation rewards raw deployment revenue without penalizing bespoke code, FDEs become an expensive services shop. They will spend their weeks maintaining one-off client glue code instead of passing blueprints back to the core platform team. If your core engineers do not view FDE tickets as missing platform primitives, your engineering organization fragments into two disconnected teams that resent each other.

What to Do With This

If you run an early technical startup and are considering your first product manager hire, pause that search this week. Take your next complex enterprise prospect, assign a senior software engineer to embed directly in their Slack channel, and give that engineer one clear goal: run the migration by hand. Instruct them to track every bespoke configuration they write into a single document. Once they run the same manual optimization for three distinct accounts, assign a core backend engineer to turn that manual script into a self-serve platform primitive.