Fuck Big Tech vs OpenCode Manual Routing

OpenCode is a harness. Manual routing is a workflow habit. Fuck Big Tech is the operating layer that makes routing measurable and repeatable.

If you already use OpenCode and manually decide which model gets which task, you are doing the right instinctive thing. The problem is that instinct does not create an audit trail.

Fast comparison

Capability	OpenCode / manual routing	Fuck Big Tech
Coding harness	Yes	Wraps it
Human decides model	Yes	Can record and enforce policy
Shared memory across harnesses	Depends on user discipline	Built into the workflow
Route telemetry	Usually missing	Required
Memory freshness checks	Usually missing	Built into degradation tests
Cost lane audit	Manual	Structured
Regression corpus	Manual or absent	First-class

Manual routing can work for one person on one task. It gets fragile when you need consistency.

The core difference

OpenCode helps you do the work.

Fuck Big Tech helps your whole AI workflow remember:

• what tool did the work
• what model or lane was chosen
• why that lane was chosen
• what memory was read
• what was written back
• whether the outcome was good
• whether future routing should change

That is the difference between a good session and a system that improves.

Why this matters for teams

Teams do not just need “use local when possible.” They need policy.

Examples:

• routine summarization can go local
• privileged legal documents cannot go to personal accounts
• final external copy needs human review
• browser actions need side-effect gates
• paid model usage should explain itself

FBT turns those rules into checks and records instead of vibes.

The benchmark angle

The benchmark is the proof layer.

Before changing skills, routing, qmd, memory, or hook behavior, run the degradation suite. That catches cases where the system still “works” in the obvious demo but silently loses source truth, stale-index detection, or handoff precedence.

That is what manual routing usually misses.