s2-B1-1 cover Session 2
Run · Design · Prioritise · Build — 90 minutes, four moves
Backlog Intelligence — Run, Agree, Build (M1 workshop)
s02 · Betterworks · 90m · PMs ·
released
s02 → 5 blocks, 38 slides, 17 trainer notes, 7 exercises, 4 reflects, 2 quizzes, 1 asides
Pre-read
- Confirm Pace Aha! sandbox login works — fix in `the cohort chat` if blocked
- Have **Claude.ai** AND **Claude Code** open in tabs — we'll switch between harnesses live
- Be signed in to the cohort chat — we'll use the pinned `the cohort chat` thread from Block 3 onward
- (Optional) skim 10 ideas in the Pace backlog — note one "huh, that's messy" reaction
No reading required. The workshop is in the room.
- Have **Claude.ai** AND **Claude Code** open in tabs — we'll switch between harnesses live
- Be signed in to the cohort chat — we'll use the pinned `the cohort chat` thread from Block 3 onward
- (Optional) skim 10 ideas in the Pace backlog — note one "huh, that's messy" reaction
No reading required. The workshop is in the room.
Session goal
By end-of-session each PM has:
1. **Run** the `/m1-cleanup` Skill on Pace data themselves — own receipt, own artifact
2. **Discussed** a shared ongoing approach to keep Aha! clean — based on what they just saw
3. **Co-authored** the v1 spec in the shared gdoc — name · inputs · output · trigger · owner
4. **Heard** what the S03 clinic looks like — each PM builds the agreed spec on their machine
5. **Asked** what they needed to ask — the open-Q&A block clears blockers before Thursday
1. **Run** the `/m1-cleanup` Skill on Pace data themselves — own receipt, own artifact
2. **Discussed** a shared ongoing approach to keep Aha! clean — based on what they just saw
3. **Co-authored** the v1 spec in the shared gdoc — name · inputs · output · trigger · owner
4. **Heard** what the S03 clinic looks like — each PM builds the agreed spec on their machine
5. **Asked** what they needed to ask — the open-Q&A block clears blockers before Thursday
Recap · Aha! adoption · idea→spec · chaining prompts · harness · M1 frame
29 itemsSet the cadence: this is a workshop, not a teach. Recap fast, frame the M1 pattern fast, land harness selection (the missing piece most PM curricula skip), and pass the mic to a cohort spotlight before the hands-on.
s2-B1-2 list Last time
- Learner > knower — show up curious; reps over answers
- Picture your AI-native team — vision before tactics
- Pick the model on purpose — not on habit
- Structure beats model — compound prompting is the lever
- … 2 more
s2-B1-3 list Today
Session 02 · 90 min · PMs
- Frame the capstone arc · ground in your Aha! reality
- Run the M1 chain on Pace data — see it in action
- Discuss the shared approach to keep Aha! clean
- Co-author the v1 spec · Q&A · close
- … 1 more
s2-B1-4 section M1 → M2 → M3, one story
The programme arc. Tonight's M1 is where it starts.
s2-B1-5 list The three milestones we're connecting
One arc · your capstone weaves these together
- M1 · Aha! Administrative Intelligence Agent — MCP-driven cleanup of the Ideas/enhancement ecosystem
- M2 · Mobile Experience Prototype — AI-native mobile UX rethink (2–3 user stories)
- M3 · AI/API Workflow & Integration — wire the M2 prototype to real data via MCP/APIs
- Tonight — we make a start on M1 together
s2-B1-6 activation Tell us about your current Aha! adoption
Quick walk-thru from anyone who loves Aha! (or doesn't)
s2-B1-7 activation Your current idea → spec workflow
How does an idea become a written spec on your team today?
s2-B1-8 section Chaining prompts
s2-B1-9 list Chaining prompts — why we break it up
One job per step. Output of one feeds the next.
- One concern per prompt — easier to debug, easier to swap
- Output is structured — JSON, table, list — next step can consume it
- Checkpoint between steps — restart at last good output, not from scratch
- Cost is sized — each step small and predictable
- … 1 more
s2-B1-10 section Where does the agent live?
Same chain. Different harness, different shape.
s2-B1-11 list What is a "harness"?
The environment that holds the agent — and shapes how it ships
- Input / output — chat · CLI · IDE · embedded app · how people talk to it
- Context — projects · memory · files · codebase · what it knows
- Tools — file system · web · MCP servers · what it can reach out to
- Distribution — who can run it, on what surface, with whose data
- … 1 more
s2-B1-12 table Three ready-made harnesses
Same Claude · different shape · pick on purpose.
col 1
col 2 · Claude.ai
col 3 · Claude Code
col 4 · Cursor
s2-B1-13 list Custom App — when you build the harness yourself
Any provider's API, embedded in your product code
- Pick your model — Claude, GPT, Gemini, open-source · per route
- Embedded in your product — your users invoke it without opening a chat
- Highest ceiling — full control of context, tools, fallbacks, evals
- Highest cost — engineering, infra, model spend, ongoing maintenance
- … 1 more
s2-B1-14 table Pick your harness — what each one gives you
col 1
col 2 · Claude.ai
col 3 · Claude Code
col 4 · Cursor
col 5 · Custom App
s2-B1-15 activation Which harness do you default to?
Drop one word in chat — claude.ai, code, cursor, custom, or none yet
s2-B1-16 list The M1 5-step chain
Each step's output feeds the next. Checkpoint between.
- EXTRACT — normalise ideas (titles, tags, signal score)
- DEDUPE — cluster on intent, not surface wording
- MAP — link clusters to Features (orphans + loop-closers)
- CLUSTER — group unlinked into 5–7 demand themes
- … 1 more
Hands-on — run /m1-cleanup on Pace data
20 itemsThe cohort runs the Skill on Pace data. PG circulates the cohort chat. The receipt lands at the end of the block — but this time *they* watched it land in their own tab.
s2-B2-1 section Let's give it a spin
Land your own receipt. Questions in the cohort chat, not meeting chat.
s2-B2-2 list What's in the package
One Skill + the 5 prompt files it wraps · Drive link in the cohort chat
- SKILL.md — the manifest Claude reads to invoke
/m1-cleanup - Step 1 EXTRACT — normalise raw Ideas
- Step 2 DEDUPE — cluster by intent
- Step 3 MAP — link to Features · surface orphans
- … 2 more
s2-B2-3 code The Skill manifest — what Claude reads to invoke /m1-cleanup
s2-B2-4 code Step 1 — EXTRACT
s2-B2-5 code Step 2 — DEDUPE
s2-B2-6 code Step 3 — MAP TO FEATURES
s2-B2-7 code Step 4 — CLUSTER INTO THEMES
s2-B2-8 code Step 5 — REPORT (Notion-ready)
s2-B2-9 statement Pit-stop discipline isn't a constraint —
it's the craft.
Pit-stop discipline isn't a constraint — it's the craft.
s2-B2-10 code How you'll run it
s2-B2-11 activation Run the chain on the Pace slice
Skill + 48-Idea slice in the package · drop your receipt in the cohort chat when it lands
Discuss · shared M1 capstone — leading questions
6 itemsThe cohort just ran the M1 chain on Pace data. Now we make it theirs. Leading questions surface what to keep, what to change, what to maintain together. The discussion converges on a shared artifact spec by end of block.
s2-B3-1 section Keep Aha! clean — together
One skill we agree on. One artifact we maintain.
s2-B3-2 list Leading questions — discuss + take a stance
What you just saw in your run · what your team needs ongoing
- What stood out? The one finding from your run worth a leadership conversation
- The mess we keep killing — orphans? dupes? stale items? something else?
- Cadence — weekly run, monthly hygiene, per-release sync?
- Owner — rotating PM-of-the-week? dedicated owner? co-maintained?
- … 1 more
Scope the v1 — commit the spec
7 itemsThe convergence block — turn the discussion into a build-able skill spec, captured in a shared gdoc. By end of block, the cohort agrees on the v1 spec. We *build* it on each PM's machine in the S03 clinic on Thursday — tonight we lock the brief.
s2-B4-1 section From discussion to a build-able spec
Co-author the spec live · we build it Thursday in the clinic.
s2-B4-2 list The shared M1 capstone skill — v1 spec
Five sections in the gdoc · cohort fills these in live
- Name — slash command · e.g.
/aha-hygiene - Inputs — Aha! Ideas export · features list · …?
- Output — the one artifact (digest · dashboard · report)
- Trigger — manual, scheduled, on-demand?
- … 1 more
s2-B4-3 activation Co-author the spec in the gdoc
Five headings already there — drop your stance under each
Build it live · takeaways · S03 setup · close
12 itemsThe session winds down from agreement to practical close. PG outlines what S03 looks like — each PM builds the agreed spec on their own machine, with PG floating to answer. Then open Q&A (the highest-leverage stretch of the session), takeaways, and close.
s2-B5-1 section S03 — your machine, your build
We agreed the spec. Thursday we build it together — on each of your machines.
s2-B5-2 list S03 Build Clinic — what to expect
Thursday · 60 min · same room · same gdoc
- Open clinic shape — each PM builds the v1 spec on their own machine
- PG floats — questions in the cohort chat, real-time answers
- Run on your slice — tonight's Pace slice + the v1 you just built
- Walk out with — your own working
/[agreed-name]Skill + a findings page
s2-B5-3 activation Open Q&A — what do you need before Thursday?
Anything · setup · scope · real-backlog · cohort chat or unmute
s2-B5-4 list Today's takeaways
- You ran the chain — your own receipt, your own artifact
- We agreed — one shared M1 capstone spec · named owner · in the gdoc
- Thursday you build it — same spec, your machine, PG floating
s2-B5-5 recap Next time
S03 — Build Clinic · Thursday 19:00 IST
- Bring the gdoc spec — you'll build it on your machine
- Re-run on your Pace slice with the v1 you built
- Bring one extension idea for your real Betterworks backlog
s2-B5-6 close Thank you
One thing that surprised you tonight — drop it in the reflection below.
After this session
- **Re-read the gdoc spec** before S03 — make sure the Owner field has at least one name (yours or someone else's). Add any field you didn't get to tonight.
- **For S03 (Build Clinic, Thursday 19:00 IST):** open Claude Code in a fresh project, bring tonight's Pace slice (`sandbox/aha/slice/`), and the gdoc spec. You'll build the agreed `/[spec-name]` Skill on your machine, then run it on your slice. PG floats — questions in the cohort chat.
- **Optional pre-Thursday warm-up:** install tonight's reference `/m1-cleanup` Skill into your project (one-line from the cohort chat). Running it once on your own slice will surface any setup gaps before the clinic.
- **Cost cap:** keep total spend under $5 per PM across S02 + S03. If you're going over, ping the cohort chat before continuing.
---
## Open items (for polish pass / PG review)
1. **B1 + B2 timing** — Block 1 carries 16 slides in a 15-min slot (capstone framing + M1 frame moved here from B5). Block 2 has the 7-slide package walkthrough before the 15-min race. Both run ~5 min long; B3-B4 reshape (now 15 min each, no meta-prompt or vote) gives back the float. Feel free to speed-walk the 5 step-prompt slides in B2 if B1 over-runs.
2. **Cohort chat thread (B3 stances)** — before B3 starts, post one message in the cohort chat (*"📌 S02 shared-skill discussion — drop your stances here"*) and **pin it**. B3 stances post here; B4 spec convergence happens in the gdoc.
3. **Spec gdoc (B4)** — live at https://docs.google.com/document/d/1XAsTWaVucoIxFrnPpSVzHa_DHp5DqalfoOWfbzl4M7s · edit access open to anyone with the link. Five headings (Name · Inputs · Output · Trigger · Owner). Pre-seed each section with a one-line *"what we saw on Pace"* prompt before B4 starts so PMs land on a primed page, not a blank one.
4. **[?CONFIRM] Pod-X spec** — `curriculum/pod-x-spec.md` referenced. Draft in polish pass per plan v2 if not yet created (members: Anurag, Lalit, Varnika, Harshini — opt-in only).
5. **[?CONFIRM] Reference receipt numbers** — verify against `sandbox/aha/m1_dryrun.py` on the slice (~14 clusters / 12 orphans / 4-6 themes / $0.05-0.15 / 1-2 min) before delivery.
6. **Race-track frame** — moved into B2 cold-open (was a standalone block in v1). 30-sec frame, not a section. If PG wants it bigger, easy to lift back into B1.
7. **Activation count** — 7 exercises across 5 blocks (target ≥5 for 90 min per `_activations.md`). Types: Aha!-adoption walk, idea→spec share, race, shared-skill discussion, gdoc spec co-author, open-Q&A, harness-poll. Passes the variety floor. (Live build moved from S02 → S03 clinic 2026-06-17 per direction — gives each PM real build time on their own machine.)
8. **S03 alignment** — `curriculum/s03.mdx` needs a rework before Thursday to match the new arc: reframe from *"run the chain on your slice"* → *"build the agreed gdoc spec on your machine, then run it on your slice."* Stale recap copy + same `layout=agenda` / `title:` bugs S02 had pre-2026-06-17. ETA ~30-45 min of edits.
9. **Unused asset** — `prompts/s02/meta-prompt-agent-ideation.md` drafted earlier, no longer used in S02. Keep for future sessions (S03 side-pod for advanced PMs, or a later session).
- **For S03 (Build Clinic, Thursday 19:00 IST):** open Claude Code in a fresh project, bring tonight's Pace slice (`sandbox/aha/slice/`), and the gdoc spec. You'll build the agreed `/[spec-name]` Skill on your machine, then run it on your slice. PG floats — questions in the cohort chat.
- **Optional pre-Thursday warm-up:** install tonight's reference `/m1-cleanup` Skill into your project (one-line from the cohort chat). Running it once on your own slice will surface any setup gaps before the clinic.
- **Cost cap:** keep total spend under $5 per PM across S02 + S03. If you're going over, ping the cohort chat before continuing.
---
## Open items (for polish pass / PG review)
1. **B1 + B2 timing** — Block 1 carries 16 slides in a 15-min slot (capstone framing + M1 frame moved here from B5). Block 2 has the 7-slide package walkthrough before the 15-min race. Both run ~5 min long; B3-B4 reshape (now 15 min each, no meta-prompt or vote) gives back the float. Feel free to speed-walk the 5 step-prompt slides in B2 if B1 over-runs.
2. **Cohort chat thread (B3 stances)** — before B3 starts, post one message in the cohort chat (*"📌 S02 shared-skill discussion — drop your stances here"*) and **pin it**. B3 stances post here; B4 spec convergence happens in the gdoc.
3. **Spec gdoc (B4)** — live at https://docs.google.com/document/d/1XAsTWaVucoIxFrnPpSVzHa_DHp5DqalfoOWfbzl4M7s · edit access open to anyone with the link. Five headings (Name · Inputs · Output · Trigger · Owner). Pre-seed each section with a one-line *"what we saw on Pace"* prompt before B4 starts so PMs land on a primed page, not a blank one.
4. **[?CONFIRM] Pod-X spec** — `curriculum/pod-x-spec.md` referenced. Draft in polish pass per plan v2 if not yet created (members: Anurag, Lalit, Varnika, Harshini — opt-in only).
5. **[?CONFIRM] Reference receipt numbers** — verify against `sandbox/aha/m1_dryrun.py` on the slice (~14 clusters / 12 orphans / 4-6 themes / $0.05-0.15 / 1-2 min) before delivery.
6. **Race-track frame** — moved into B2 cold-open (was a standalone block in v1). 30-sec frame, not a section. If PG wants it bigger, easy to lift back into B1.
7. **Activation count** — 7 exercises across 5 blocks (target ≥5 for 90 min per `_activations.md`). Types: Aha!-adoption walk, idea→spec share, race, shared-skill discussion, gdoc spec co-author, open-Q&A, harness-poll. Passes the variety floor. (Live build moved from S02 → S03 clinic 2026-06-17 per direction — gives each PM real build time on their own machine.)
8. **S03 alignment** — `curriculum/s03.mdx` needs a rework before Thursday to match the new arc: reframe from *"run the chain on your slice"* → *"build the agreed gdoc spec on your machine, then run it on your slice."* Stale recap copy + same `layout=agenda` / `title:` bugs S02 had pre-2026-06-17. ETA ~30-45 min of edits.
9. **Unused asset** — `prompts/s02/meta-prompt-agent-ideation.md` drafted earlier, no longer used in S02. Keep for future sessions (S03 side-pod for advanced PMs, or a later session).
Add ?token=… to the URL.