Contents (8)
The 20/80 Worksheet
Architecting Agentic Workflows · Boulder Startup Week 2026
Workshop: Architecting Agentic Workflows for the Lean 2026 Startup Speaker: Sophia Stein · AI Architect Resources: agenticarchitect.ai/blog · github.com/sudosoph/bsw26-agentic-workflows
The 3 axes
| Axis | High value (automate) | Low value (don't) |
|---|---|---|
| Volume | Daily / weekly recurrence | Once a quarter or less |
| Determinism | Same steps every time | Every case is different |
| Reversibility | If the agent's wrong, you fix it | If the agent's wrong, real damage |
GREEN = high vol × deterministic × reversible → automate now
YELLOW = high vol × ambiguous × reversible → automate WITH HITL
RED = low vol × ambiguous × irreversible → don't automate
Step 1 — List 5 tasks you do every week
Anything recurring counts. Outbound. Support. Research. Operations. Content. Reporting. Hiring funnel. Pricing decisions.
| # | Task | Volume (H/L) | Determinism (D/A) | Reversibility (Low/High stakes) | Color (G/Y/R) |
|---|---|---|---|---|---|
| 1 | |||||
| 2 | |||||
| 3 | |||||
| 4 | |||||
| 5 |
Step 2 — Pick your first agent
Look at your Greens. Pick the one that would free up the most time per week. Write it here:
My first agent will do:
________________________________________________________________
________________________________________________________________
It will run every: ☐ Day ☐ Week ☐ Triggered by event: __________
I'll approve / send / accept its output via:
☐ Gmail Drafts ☐ Slack ☐ Linear ☐ Notion ☐ Other: __________
Step 3 — Sketch the architecture
Every agent has the same 5 boxes. Fill them in for YOUR agent.
┌──────────────┐ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ TRIGGER │ → │ READ CONFIG │ → │ SEARCH / │ → │ PROCESS │ → │ OUTPUT GATE │
│ │ │ (the files) │ │ EXTRACT │ │ (cascade) │ │ (HITL) │
│ │ │ │ │ │ │ │ │ │
│ │ │ │ │ │ │ │ │ │
└──────────────┘ └──────────────┘ └──────────────┘ └──────────────┘ └──────────────┘
Examples to spark you:
Inbound triage:
Webhook (form fill) → ICP file → enrich lead → Sonnet draft → Gmail draft
Competitor watch:
Cron (daily) → competitors.csv → scrape changelogs → Haiku diff → email digest
Content repurposer:
Webhook (new blog) → voice.md → break into 3 channels → Sonnet rewrite each → drafts in Drive
Step 4 — Estimate the impact
Be specific. Vague numbers don't motivate.
Today, I spend ___ hours / week on this task.
With the agent + HITL gate, I'll spend ___ hours / week.
Hours back per month: ___
What I'll do with those hours:
________________________________________________________________
Step 5 — Take it home
Tonight:
- ☐ Sign up for n8n.cloud (or self-host on a $5 Hetzner VPS)
- ☐ Get an Anthropic API key with $50 budget
- ☐ Get a Firecrawl trial key (free tier, 500 credits)
- ☐ Write your
voice.mdandicp.mdfiles (start with the templates from the repo)
This weekend:
- ☐ Fork github.com/sudosoph/bsw26-agentic-workflows
- ☐ Import the n8n JSON
- ☐ Wire up your credentials
- ☐ Do 3 dry runs
Next week:
- ☐ Let it run daily for 7 days
- ☐ Review every draft before sending — track approval rate
- ☐ When approval rate hits 95%+, you're ready for Stage 3 (conditional auto)
Two ways to keep going
Architecture Audit — 5 free 30-min audits for BSW attendees. First-come. Agentic Architect Blog — agenticarchitect.ai/blog · weekly deep-dives.
Worksheet by Sophia Stein · MIT licensed · feel free to fork & adapt
Local-First AI
If this template is useful, weekly notes go deeper on patterns like this.
n8n templates, cost teardowns, and what is actually working in 2026. No drip sequences, no upsells. Reply to opt out.