Day 1 Close & Day 2 Preview

What you built · why it matters · what's next

Day 1 · Session 8 — 15 minutes · laptops closed

The close is as important as the open.
You leave knowing exactly what you built, exactly why it matters,
and exactly what to have ready for 9 AM tomorrow.

This Morning vs. Now

You came in this morning as developers who use AI tools.
You leave today as developers who build AI tools.

Take a breath. Let that land before we look at tomorrow.

Part 1 — What You Built

Five artifacts. One system.

The Five Artifacts

# Artifact What it is Its role in the full StackLog system
1 stacklog-mcp/src/index.ts Your MCP server — create_entry + search_entries The data layer. Tomorrow's React app reads this exact store.
2 stacklog-entries.json The persistent JSON store (external memory) The bridge between Day 1 and Day 2. The app will display every entry in it.
3 lab3-system-prompt.md A reusable agent prompt — patterns + five guards The design document for autonomous agents. Reused for Day 2's Architect persona.
4 lab3-retro.md Break scenarios: failure mode, root cause, guard Evidence of engineering thinking. Feeds Day 2's spec as a source of requirements.
5 Antigravity MCP config mcp_config.json — three servers connected The agent runtime. Hosts the Architect/Developer/Reviewer personas all day tomorrow.

Look at Row 2

stacklog-entries.json doesn't exist because of a pre-written demo.

It exists because you built the MCP server that creates it —
and because an AI called your code to write it.

And row 5: this morning your MCP config was empty.
Now it connects three servers — two pre-built, one you wrote from scratch.

That's an agentic development environment you own.

The Two Days Are One System

Tomorrow, a React app reads row 2 and displays it in a browser.

Day 1 and Day 2 aren't separate.
They're one system — and you've already built half of it.

Part 2 — Three-Question Retro

Say it out loud — it consolidates the day

Retro — Question 1

"What's the single most important thing you built today — and why?"

The MCP server? (the one artifact used again tomorrow)
The system-prompt process? (it generalises far beyond this room)

Both are right. The insight: the artifact matters less than the thinking behind it.

Retro — Question 2

"What would you do differently for a real production system?"

Likely answers — and note them, they preview tomorrow's spec:

  • Authentication on the server · SQLite instead of JSON
  • Better tool error handling · rate-limiting on search
  • Human-in-the-loop approval before create_entry

Every one of these is a requirement — exactly what a spec captures.

Retro — Question 3

"Finish the sentence: this morning I thought agents were ___,
now I think they are ___."

Common shifts: magic → designed · automatic → explicit · smart → structured · hard to build → totally buildable.

Whatever you say — that gap is the mindset shift this workshop exists to create.

Part 3 — Day 2 Preview

From "How" to "What"

The Pivot

Today you mastered the How —
how agents work, how tools are called, how prompts control behaviour,
how failures happen and how to guard against them.

Tomorrow is the What —
given all that, how do you build software well with AI?

The answer is spec-driven development.
And the spec you write tomorrow describes an app
that reads the file you built today.

What Tomorrow Looks Like

Session What you'll do Connection to today
S9 — Why specs beat prompts What a spec is, why it beats ad-hoc prompting your lab3-retro.md is an informal spec
S10 — BMAD lifecycle The 6-stage pipeline + 3 personas Architect persona mirrors your Session-6 prompt design
Lab 4 — Spec sprint Write the PRD + Tech Spec for StackLog web the spec describes the app that reads your JSON store
Lab 5 — Stories & tasks Spec → prioritised backlog (≥12 tasks) stories map to the tool schemas you wrote in Lab 2
Lab 6 — Codegen sprint Build React + Express + SQLite the backend serves the same data your MCP server writes
S11 — Spec drift & review Reviewer persona: diffs vs. spec same diagnostic skill as the break exercise, on code
S12 — Showcase Live: MCP creates entry → app displays it both days connect in one demo

The Connection Moment

Tomorrow at the showcase, every team will do exactly this:

  1. Open the agent. Run a prompt.
  2. Watch the MCP server from Lab 2 create a new entry.
  3. Open the browser. Refresh the web app from Lab 6.
  4. The entry you just created appears on screen.

One prompt. Two days of work. One live system.

It starts with writing a spec that describes exactly how that app works.

Part 4 — Evening Prep

Three things to do tonight

Tonight's Three Actions

1 — Install / verify your codegen agent
Make sure your Antigravity agent is ready for code generation (Labs 5–6).
(If your track also uses Copilot Chat in VS Code, install it now: search "GitHub Copilot Chat" in Extensions.)

Why: Lab 6 generates the app — you need the agent working before you can follow along.

2 — Clone the BMAD Method templates

🪟  cd $env:USERPROFILE\stacklog-workshop
🍎🐧 cd ~/stacklog-workshop

git clone https://github.com/bmadcode/BMAD-METHOD docs/bmad

Why: Lab 4 opens the BMAD PRD template. Cloning tonight = no download wait tomorrow.
If the URL changed: search "BMAD METHOD github" — it's the top result.

3 — Skim the BMAD README (15 min max)
Read only the overview of docs/bmad/README.md.

Why: you just need to recognise the four document types — PRD, Tech Spec, Story Map, Task List — so Session 9 lands immediately.

Optional, but Worth It

Open stacklog-entries.json and read the entries you created today.
These are the records the StackLog app will display tomorrow.

Think of one feature you'd want to add —
you may get to spec and build it in the Lab 4 extension challenge.

Day 1 Completion Check

Confirm before you leave (or by 9 AM tomorrow):

  • [ ] stacklog-mcp/ pushed to GitHub — both tools implemented
  • [ ] lab3-system-prompt.md committed — four sections, five guards
  • [ ] lab3-retro.md committed — ≥ 2 break scenarios documented
  • [ ] stacklog-entries.json has ≥ 2 entries (one from Lab 1, one from Lab 3)
  • [ ] MCP config has 3 servers: filesystem, search (ddg-search/mock-search), stacklog

Missing a GitHub push? You have until 9 AM. We won't hold the class.

Close — On Energy, Not Admin

You came in this morning as developers who use AI tools.
You leave as developers who build AI tools.

What you built today — an MCP server, a multi-tool agent pipeline,
a documented break-and-harden process — is not a workshop exercise.
It's a portfolio artifact. Put it on GitHub. Talk about it in interviews.

See you at 9 AM. Laptop charged. Codegen agent ready.