The Lab

Field notes from production AI.

Long-form essays and short engineering notes from our forward-deployed work. Cross-posted from ForwardDeployment.Engineer.

Featured · Operating model

Why forward deployment is the only way to hire AI engineers

The org chart that worked for cloud doesn't work for AI. A short argument for shipping engineers into the customer's stack.

May 20266 min read
// runtime.partners/fdev0.4.2
$ runtime deploy --pod=fde-01
→ embedding 3 engineers
→ proxy: online
→ mcp.git: attached
→ tracker: recording
// Shipping intelligence to production...
Infrastructure

MCP: the hidden key to connecting LLMs to your private data

Why every serious AI program ends up building Model Context Protocol servers — and how to scope yours so it doesn't sprawl.

May 20269 min
Strategy

ROI math for AI: stop counting tokens, start counting workflows

A board-ready model for ranking AI initiatives by margin impact instead of model spend.

Apr 20267 min
Safety

Your first AI investment should be an LLM proxy

Before agents, before fine-tunes, before vector databases. The case for centralizing every model call behind a proxy you control.

Apr 20265 min
Migration

Read-only HITL is the cheat code for enterprise AI

How to ship your first production AI flow without a single autonomous action — and why your CRO will thank you.

Mar 20264 min
Evals

Write the eval before the agent

Agents are easy. Knowing whether they're good is the actual work. A practical eval-driven build process.

Mar 20268 min
More on ForwardDeployment.Engineer ↗
Have a topic you'd want us to write up? Tell us. →