Productive Capacity Design
Menu

Toolkit

AI Deployment Filter

AI amplifies the system it enters. Filter the work system before filtering vendors.

AI Deployment Filter

If you're not sure, the system isn't ready to be sure either.

Core screen

Is the workflow stable enough to automate?
Is ownership clear?
Are decision rights clear?
Is the data trusted?
Are exceptions understood?
Is there a human verification loop?
Could AI increase harm, bias, confusion, or surveillance anxiety?
Would the tool reduce friction or just move work somewhere less visible?
Who benefits from the capacity released?
How will employees experience the change?
What happens to the work that gets displaced?

Semantic readiness screen

Are the core business terms defined?
Are definitions owned by named stewards?
Are metrics calculated consistently?
Are source systems identified?
Is lineage available?
Are freshness and quality signals visible?
Are validation rules documented?
Are permissions and sensitivity rules enforced?
Are exceptions understood?
Are approved AI skills/actions defined?
Can the AI explain which definition, source, or rule it used?
Can a human challenge or override the output?
Does the workflow have a feedback loop to improve the semantic layer over time?

Verdict

Stay out until redesigned

The proposal may move work somewhere less visible instead of reducing friction.

Full autonomy is never issued by this filter; human accountability remains part of the design.

No-go list

  • Is the workflow stable enough to automate? (AI-readiness debt)
  • Is ownership clear? (Role debt)
  • Are decision rights clear? (Decision debt)
  • Is the data trusted? (Data debt)
  • Are exceptions understood? (Process debt)
  • Is there a human verification loop? (Role debt)
  • Would the tool reduce friction or just move work somewhere less visible? (Trust debt)
  • Are the core business terms defined? (Data debt)
  • Are definitions owned by named stewards? (Data debt)
  • Are metrics calculated consistently? (Data debt)
  • Are source systems identified? (AI-readiness debt)
  • Is lineage available? (AI-readiness debt)
  • Are freshness and quality signals visible? (AI-readiness debt)
  • Are validation rules documented? (AI-readiness debt)
  • Are permissions and sensitivity rules enforced? (Data debt)
  • Are exceptions understood? (Data debt)
  • Are approved AI skills/actions defined? (Data debt)
  • Can the AI explain which definition, source, or rule it used? (AI-readiness debt)
  • Can a human challenge or override the output? (AI-readiness debt)
  • Does the workflow have a feedback loop to improve the semantic layer over time? (AI-readiness debt)