The FDE Feedback Loop

I keep coming back to one diagram about forward-deployed engineering. It's a loop with four steps, and the tightness of the loop is what determines whether you build a consulting firm or something more durable.

The loop

Engagement → Pattern → Tool → Faster engagement
     ↑                                  │
     └──────────────────────────────────┘

1. Engagement. A client pays you to ship something inside their environment.
2. Pattern. While shipping, you notice a generalizable pattern. (Auth setup for Stripe Connect. Postgres schema for multi-tenant SaaS. Whatever.)
3. Tool. You codify the pattern as something repeatable — a CLI command, a Claude Code skill, a template, a reusable migration.
4. Faster engagement. Next engagement that hits the same pattern uses the tool. You ship in days instead of weeks. Margin compounds.

Why most consulting firms can't do this

The trap: most consulting firms bill hourly. Every hour spent extracting a pattern and writing a tool is unbillable. The incentive is to redo the engagement from scratch every time, because that maximizes hours.

Fixed-price engagements with the FDE compounds reverses the incentive. Now every hour you save is your margin. The toolkit is the asset.

How Benmore implements it

Three things make the loop tight at Benmore:

• Foundry CLI — every engagement contributes commands back to a shared CLI used by all engineers.
• Claude Code skills — generalizable workflows become skills the next engineer invokes by name (/stripe-integration, /django-celery-expert, etc.).
• docs/solutions/ — every gnarly debugging session gets written up as a solution so the next person hits the bug for two minutes instead of two days.

The win condition

You know the loop is working when a junior engineer onboards in week one and ships production code in week two — using tools that didn't exist twelve months ago, that were extracted from work no one is doing again.

That's the entire game.