Builder Capstone: Design an Agent for Your Life

No code. Just design. Pick a real task you do every week and draft a complete agent spec — goal, tools, loop, stop, approvals, and what success looks like.

40 min · Reviewed 2026

The exercise

Pick a task you do every week that you'd rather not. Not coding. Something from your real life — sorting receipts, clipping news, replying to the same kind of email, filing school forms, checking prices on a thing you want to buy. Your job is to design (not build) an agent that could do it.

The spec template

# Agent Spec: [Name your agent] ## 1. Goal In one sentence, what does success look like? ## 2. Trigger What starts the agent? (A button, a schedule, an email, my voice?) ## 3. Inputs What does the agent see to begin? (URL, file path, prompt, calendar.) ## 4. Tools needed - [ ] Filesystem (which folders?) - [ ] Browser (which sites?) - [ ] Email (send, read, or both?) - [ ] Calendar - [ ] Shell / code - [ ] MCP server (which?) - [ ] Other ## 5. Steps (the loop) Sketch the think/act/observe cycle for a typical run. ## 6. Stop condition What makes the agent say 'done'? What makes it give up? ## 7. Approval gates Which actions need a human yes before running? ## 8. Failure modes What could go wrong? How would you notice? ## 9. Sandbox Where will it run? What can it NOT touch? ## 10. Success metric How will you measure if it's actually helping?Fill this in on paper, in a doc, or in a notes app. Short answers only.

A worked example

# Agent Spec: Weekly Receipt Sorter 1. Goal: At the end of each week, every receipt email is saved as a PDF in ~/Finance/2026/ and logged in a running CSV. 2. Trigger: Sunday 6pm, automatic. 3. Inputs: Gmail inbox. Label 'Receipts'. 4. Tools: Gmail MCP, Filesystem (write only to ~/Finance), Code execution (for PDF generation + CSV updates). 5. Steps: - Fetch new 'Receipts' emails since last run. - For each: extract merchant, amount, date. Save as PDF. - Append one row to receipts-2026.csv. - Mark email as 'Processed'. 6. Stop: No more unprocessed receipt emails, OR 50 emails processed (cap), OR an error on 3 consecutive emails. 7. Approvals: None for read/write inside ~/Finance. Approval required if any email suggests fraud (flag amounts > $500). 8. Failure modes: - Email without a parseable amount → skip + notify. - Filesystem full → stop + notify. - Gmail API throttle → back off, retry next run. 9. Sandbox: Runs in OpenClaw. Filesystem scope: ~/Finance only. Network: Gmail API only. 10. Success metric: Weekly ritual of 'where did I put that receipt' takes zero minutes instead of thirty.A real spec you could hand to a developer (or your future self).

Common design mistakes

Goal too vague ('manage my life') — break it down until each agent does one thing.
No stop condition — always set a cap on steps, time, or cost.
Too many tools — if your agent needs 10 tools, it's probably 3 agents.
No failure plan — what if a tool is down? What if the input is weird?
Skipping approvals on risky steps — speed is not worth surprise bills.

When you're ready to actually build — wire up tools, pick a framework, write code — head to the Creators track. You now have the design muscle to ship something safe and useful.

End-of-lesson check

8 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-agentic-builder-capstone-builders

What is the main idea of "Builder Capstone: Design an Agent for Your Life"?
1. Pick a real task you do every week and draft a complete agent spec — goal, tools, loop, stop, approvals, and what success looks like.
2. Use AI as the final authority for the whole decision
3. Avoid checking the answer once it sounds polished
4. Focus only on speed instead of judgment
Which concept is most central to "Builder Capstone: Design an Agent for Your Life"?
1. specification
2. agent design
3. capstone
4. use case
Which use of AI fits this topic best?
1. Let the AI decide what matters without your review
2. Use the answer before checking whether it fits the situation
3. Goal too vague ('manage my life') — break it down until each agent does one thing.
4. Use the first answer without checking it
What should a careful learner remember about "Why design-only"?
1. Use AI to draft or organize ideas about agent design, then verify before acting.
2. Skip the context so the tool can guess faster
3. Treat the output as private even after sharing it online
4. Use the answer without checking the source
You want to use AI after this lesson. What is the safest next step?
1. Act immediately because the AI answer is written clearly
2. Use the AI answer as a draft, then check it against a reliable source.
3. Hide uncertainty so the final answer looks cleaner
4. Use private or sensitive details before checking permission
How should AI output about agent design be treated?
1. As proof that no other source is needed
2. As a replacement for context, consent, or expert review
3. As a draft or helper output that still needs human judgment and verification
4. As something that becomes correct when it sounds confident
Name one way to verify an AI answer about agent design.
Which action would help you apply "Builder Capstone: Design an Agent for Your Life" responsibly?
1. Use the tool to avoid thinking through the tradeoff
2. Keep going even if the output conflicts with a trusted source
3. Use the first answer without checking it
4. No stop condition — always set a cap on steps, time, or cost.

← Back to interactive lesson

Tendril · Builders · Agentic AI

Builder Capstone: Design an Agent for Your Life

No code. Just design. Pick a real task you do every week and draft a complete agent spec — goal, tools, loop, stop, approvals, and what success looks like.

40 min · Reviewed 2026

The exercise

The spec template

# Agent Spec: [Name your agent] ## 1. Goal In one sentence, what does success look like? ## 2. Trigger What starts the agent? (A button, a schedule, an email, my voice?) ## 3. Inputs What does the agent see to begin? (URL, file path, prompt, calendar.) ## 4. Tools needed - [ ] Filesystem (which folders?) - [ ] Browser (which sites?) - [ ] Email (send, read, or both?) - [ ] Calendar - [ ] Shell / code - [ ] MCP server (which?) - [ ] Other ## 5. Steps (the loop) Sketch the think/act/observe cycle for a typical run. ## 6. Stop condition What makes the agent say 'done'? What makes it give up? ## 7. Approval gates Which actions need a human yes before running? ## 8. Failure modes What could go wrong? How would you notice? ## 9. Sandbox Where will it run? What can it NOT touch? ## 10. Success metric How will you measure if it's actually helping?Fill this in on paper, in a doc, or in a notes app. Short answers only.

A worked example

# Agent Spec: Weekly Receipt Sorter 1. Goal: At the end of each week, every receipt email is saved as a PDF in ~/Finance/2026/ and logged in a running CSV. 2. Trigger: Sunday 6pm, automatic. 3. Inputs: Gmail inbox. Label 'Receipts'. 4. Tools: Gmail MCP, Filesystem (write only to ~/Finance), Code execution (for PDF generation + CSV updates). 5. Steps: - Fetch new 'Receipts' emails since last run. - For each: extract merchant, amount, date. Save as PDF. - Append one row to receipts-2026.csv. - Mark email as 'Processed'. 6. Stop: No more unprocessed receipt emails, OR 50 emails processed (cap), OR an error on 3 consecutive emails. 7. Approvals: None for read/write inside ~/Finance. Approval required if any email suggests fraud (flag amounts > $500). 8. Failure modes: - Email without a parseable amount → skip + notify. - Filesystem full → stop + notify. - Gmail API throttle → back off, retry next run. 9. Sandbox: Runs in OpenClaw. Filesystem scope: ~/Finance only. Network: Gmail API only. 10. Success metric: Weekly ritual of 'where did I put that receipt' takes zero minutes instead of thirty.A real spec you could hand to a developer (or your future self).

Common design mistakes

Goal too vague ('manage my life') — break it down until each agent does one thing.
No stop condition — always set a cap on steps, time, or cost.
Too many tools — if your agent needs 10 tools, it's probably 3 agents.
No failure plan — what if a tool is down? What if the input is weird?
Skipping approvals on risky steps — speed is not worth surprise bills.

When you're ready to actually build — wire up tools, pick a framework, write code — head to the Creators track. You now have the design muscle to ship something safe and useful.

End-of-lesson check

8 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-agentic-builder-capstone-builders

What is the main idea of "Builder Capstone: Design an Agent for Your Life"?
1. Pick a real task you do every week and draft a complete agent spec — goal, tools, loop, stop, approvals, and what success looks like.
2. Use AI as the final authority for the whole decision
3. Avoid checking the answer once it sounds polished
4. Focus only on speed instead of judgment
Which concept is most central to "Builder Capstone: Design an Agent for Your Life"?
1. specification
2. agent design
3. capstone
4. use case
Which use of AI fits this topic best?
1. Let the AI decide what matters without your review
2. Use the answer before checking whether it fits the situation
3. Goal too vague ('manage my life') — break it down until each agent does one thing.
4. Use the first answer without checking it
What should a careful learner remember about "Why design-only"?
1. Use AI to draft or organize ideas about agent design, then verify before acting.
2. Skip the context so the tool can guess faster
3. Treat the output as private even after sharing it online
4. Use the answer without checking the source
You want to use AI after this lesson. What is the safest next step?
1. Act immediately because the AI answer is written clearly
2. Use the AI answer as a draft, then check it against a reliable source.
3. Hide uncertainty so the final answer looks cleaner
4. Use private or sensitive details before checking permission
How should AI output about agent design be treated?
1. As proof that no other source is needed
2. As a replacement for context, consent, or expert review
3. As a draft or helper output that still needs human judgment and verification
4. As something that becomes correct when it sounds confident
Name one way to verify an AI answer about agent design.
Which action would help you apply "Builder Capstone: Design an Agent for Your Life" responsibly?
1. Use the tool to avoid thinking through the tradeoff
2. Keep going even if the output conflicts with a trusted source
3. Use the first answer without checking it
4. No stop condition — always set a cap on steps, time, or cost.

← Back to interactive lesson