Exploring Human Agency In The AI Age: Decoding

Ed Reif

Ed's AI Mission Log

A field guide for working with AI systems

Coaching The Machine

Stop treating AI like a search engine. Start treating it like a junior intelligence analyst you're training. The difference: clear operational context, iterative refinement, and mission-specific standards. Most operators get mediocre results because they give mediocre tactical guidance.

AI can't read your mind. If a human couldn't complete the task with what you've provided, neither can AI.

Bad prompt
"Write a fleet readiness report"
Good prompt
You're writing a weekly fleet readiness brief for the Naval Operations Center.

Audience: Flag officers who need operational decisions, not technical details
Format: 3 bullet points max, each under 30 words
Tone: Direct, mission-focused language

Current status: 7 of 12 destroyers mission-ready, 2 in maintenance, 3 awaiting parts delivery
Key decision needed: Deploy available assets now or wait 72 hours for full squadron?

Draft the brief now.

Context checklist:

  • Who is the audience?
  • What's the purpose?
  • What constraints exist?
  • What does success look like?

The first draft is rarely the answer. Build iteratively, like you would with any team member.

Multi-step process
Step 1: "Draft 3 different approaches for securing this shipping lane"
Step 2: "Take approach #2, but optimize for smaller patrol boat deployment"
Step 3: "Now add specific threat assessment metrics for each checkpoint"
Step 4: "Red-team this: what would hostile forces exploit in this plan?"

Iteration techniques:

  • Ask for multiple options before choosing one
  • Request reasoning before conclusions
  • Have it critique its own work
  • Specify what to improve in the next version

AI performs better when it knows what role to play. Be specific about expertise level and perspective.

Effective role assignment
"You're a senior maritime intelligence analyst with 8 years of naval operations experience. You're reviewing this port security assessment for operational feasibility and threat mitigation. Be constructively critical — identify vulnerabilities and recommend countermeasures."

Role clarity checklist:

  • What's their expertise level?
  • What's their perspective or bias?
  • What are they optimizing for?
  • How should they communicate?

Constraints unlock better solutions. Set limits to force creative thinking.

Constraint examples
- "Explain this threat assessment in under 75 words"
- "No classified info — brief for coalition partners"
- "Budget is $2M, not $20M, for this operation"
- "Only use assets currently in theater"
- "Must work in a contested communications environment"

Useful constraint types:

  • Word / time limits
  • Audience restrictions
  • Resource constraints
  • Technical limitations
  • Format requirements

Build quality checks into your process. Don't accept first drafts.

Quality gate process
Before finalizing any operational plan, ask:

1. "Is this actionable with current force structure?"
2. "What intelligence supports these threat assumptions?"
3. "What are the top 3 enemy courses of action?"
4. "How would a maritime domain expert critique this?"
5. "Does this actually achieve mission objectives?"

Standard quality checks:

  • Evidence backing claims
  • Clear next actions
  • Appropriate detail level
  • Logical consistency
  • Audience appropriateness

Copy-paste templates

Initial briefing template

  • Role: You are a [specific role] with [experience level]
  • Task: [Specific deliverable]
  • Audience: [Who will use this, and why]
  • Constraints: [Time, format, style, resource limits]
  • Success criteria: [How you'll judge quality]

Iteration commands

  • "Show me 3 different tactical approaches before we proceed"
  • "What threat assumptions are you making? List them."
  • "Now critique your own plan — what vulnerabilities exist?"
  • "Make this 50% more concise for a command briefing"
  • "What would enemy intelligence object to in this assessment?"

Quality gates

  • Is this operationally feasible with current assets?
  • What intelligence supports these assessments?
  • Have we considered enemy countermeasures?
  • Would this survive operational review?

The goal isn't to use AI faster — it's to produce better operational intelligence. These techniques take discipline up front, but they compound over time as you and the machine build better tactical collaboration.

Archive

Show more