Architect's Field Kit 03
The 9 Skills
The durable habits behind capable AI and agent architects
- Audience
- Builders ages 13-18 and families
- Format
- 18 minute field read
- Use
- Read the overview now. Open the matching card when a build calls for it.
Field purpose
Nine skill cards adapted from the ReInvent U knowledge base, each with a real-world rep, a visual model, and evidence that can appear on the capability record.
Field panel 01
grid
The skill tree at a glance
These skills are earned by use. A builder does not need all nine before starting.
- 01
Resourceful start
Use AI to find the next move without giving away the goal.
- 02
Skeptical checking
Assume fluent is not the same as true.
- 03
Context loading
Give the material and situation the tool cannot guess.
- 04
Prompt iteration
Treat the first result as the start of a conversation.
- 05
Tool chaining
Move work between tools with a purpose and a checkpoint.
- 06
Delegation thinking
Define what AI can do and what the human must own.
- 07
Agent task design
Set a goal, steps, limits, and stop conditions.
- 08
Quality control
Know what good looks like before approving the output.
- 09
Learning velocity
Pick up a new tool quickly and transfer the method.
Field panel 02
skill
Skill 01 - Resourceful start
When stuck, use AI to reveal routes forward without asking it to own the destination.
- What it is
Move from fog to options
Name what you are trying to do, what you have tried, and the exact place you are stuck.
- Builder rep
Ask for three next moves
Request three small options with tradeoffs, then choose one and say why.
- Evidence
The choice is yours
Keep the prompt, the options, and your reason for picking one.
Field panel 03
skill
Skill 02 - Skeptical checking
AI can sound certain when it is wrong. Strong builders design a check before they trust the answer.
- What it is
Trust evidence, not tone
Separate claims, calculations, links, and assumptions so each can be tested.
- Builder rep
Catch one error
Compare a claim with a primary source, run a calculation another way, or test the output with a known case.
- Evidence
The check record
Record what you checked, how you checked it, what failed, and what changed.
Field panel 04
skill
Skill 03 - Context loading
A tool cannot see the goal, audience, files, tone, or limits unless the builder provides them.
- What it is
Give the missing world
Supply the real material, intended user, useful examples, constraints, and privacy boundaries.
- Builder rep
Run before and after
Try a vague prompt, then a context-rich prompt. Compare what became more useful.
- Evidence
A context pack
Keep the brief and a note naming which context changed the result most.
Field panel 05
skill
Skill 04 - Prompt iteration
The first prompt is a draft. Architects question, refine, push back, and narrow.
- What it is
Direct through dialogue
Inspect the first result and name the strongest part, weakest part, and next change.
- Builder rep
Three-pass improvement
Run a shape pass, a stress-test pass, and a final fit pass. Change one variable at a time.
- Evidence
A visible trail
Keep the original, the prompts that changed it, and the final version.
Field panel 06
skill
Skill 05 - Tool chaining
One tool rarely does every job best. A chain makes handoffs deliberate instead of messy.
- What it is
Give each tool one job
For example: collect input, organise it, draft an output, design it, then publish it.
- Builder rep
Map a two-tool chain
Name the input and output for each tool and place a human check between them.
- Evidence
The handoff map
Show the chain, the checkpoint, and one problem the checkpoint caught.
Field panel 07
skill
Skill 06 - Delegation thinking
Good delegation is not 'do everything.' It is a clear transfer with ownership left in the right place.
- What it is
Split the work honestly
AI may explore, sort, draft, or transform. The builder owns the goal, facts, judgment, permission, and final approval.
- Builder rep
Draw the ownership line
List tasks under AI can do, human must do, and human must check.
- Evidence
The disclosure block
State what the builder directed and what AI produced in plain language.
Field panel 08
skill
Skill 07 - Agent task design
An agent needs more than a prompt. It needs a goal, tools, a loop, limits, and a safe way to stop.
- What it is
Design bounded autonomy
Define allowed actions, forbidden actions, required approval, retry limits, and a clear finish condition.
- Builder rep
Run a small agent
Use a low-risk job such as organising public information or proposing a family plan that an adult approves.
- Evidence
The agent card
Keep the goal, tool list, boundaries, test cases, run log, and stop control.
Field panel 09
skill
Skill 08 - Output quality control
Checking truth is essential. Quality control also asks whether the result is useful, clear, safe, and right for its user.
- What it is
Define good before review
Write three to five observable success checks before asking AI to produce or improve anything.
- Builder rep
Review with a rubric
Use no numerical rating. Mark each check as passes, needs work, or cannot yet verify, then fix the highest-impact gap.
- Evidence
The approval note
Keep the checks, result, change made, and reason it is now ready.
Field panel 10
skill
Skill 09 - Learning velocity
Tools will change. The durable skill is becoming useful with the next one without losing the method.
- What it is
Transfer the pattern
Find the new tool's input, output, limits, sharing controls, and fastest safe practice task.
- Builder rep
One-hour tool switch
Use an unfamiliar approved tool to complete a small job, keeping notes on surprises and equivalents.
- Evidence
The transfer map
Show old move, new move, what stayed the same, and what needed relearning.
Keep it in the field
Print it, mark it, return to the build.
The PDF carries the same source edition as this web guide and is designed with A4 and US Letter-safe print margins.