# Flare / Constellation — Transcript Extraction Notes

> Source: "Leadership | Horizon & AI Transformation" — conversation between Elie Rodrigue and Alex Savard.
> The gamified task-marketplace discussion runs roughly from 1:02:00 to 1:33:00, with problem context starting ~54:00.
> Working names: **Flare** or **Constellation**. Currency candidates: **Stardust**, **Photons**, **Lumens**
> (in the meeting the placeholder was "Horizon points"; Green Stars = existing internal recognition currency).

---

## 1. The problem it solves

- **Invisible lost hours.** Yoann's framing: ~600 people in Customer, utilization ~70% — the missing 30% represents thousands of hours per week that nobody can see or direct ("Where are these hours going?"). (~1:23)
- **Closed-door work attribution.** SMs and directors invent work and hand it to their subordinates "à huis clos" (behind closed doors). Nobody checks whether someone else was free or better suited; the driver is "I want to please Sid", not efficiency. (~1:07)
- **Favoritism over merit.** Career progress currently flows through attaching yourself to 2–3 patrons. The person who networks well gets the good work; the equally capable person who doesn't is "stuck doing QA". Promotion stagnation (2 years without promos) has broken the old "do this and I'll promote you" lever. (~1:22–1:25)
- **Hidden costs / undersold projects.** Freed-up hours get absorbed by leaders using PRD work to hide extra costs on undersold projects. Yoann wants control and visibility on where liberated hours go. (~1:24)
- **Bench management is unmanageable top-down.** Yoann floated making Claudia "chief of staff" of the bench (coordinating 60–150 idle people at any time, a constantly rotating pool) — Alex's reaction: that's really complicated; a rotating pool needs fixed objectives, not micromanagement. (~58:00)
- **Junior-heavy teams don't work with classic management.** The York project is the internal counter-example: a team of superstars that runs badly because roles are unclear and everyone steps on each other. Nobody (e.g., Marc) can micromanage 30 juniors. (~1:03)
- **Bench churn + no quality gate.** On Nation Kit, staffing from the bench means 2 developers rotate out every 2 weeks — "incontrôlable". On Premières Nations, a bench person proudly delivered an unrequested Amazon Connect integration and the SM wanted to merge it — uncontrolled contributions with no quality gate. (~59:30)
- **Meeting/ceremony overhead is the same disease.** Earlier in the call (Polytechnique estimate cut from 12 to 9, ~3,000 hours): "Scrums have a weekly sync-up, and the weekly sync-ups have a sync-up with design… the hours pile up." Fully-specified postings with no meetings attack the same overhead. (~17:37)
- **Small tasks carry absurd overhead.** Today, getting a developer for 1–2 hours means WBS codes, discussions, meetings — so the work simply doesn't happen. Elie: "If I had one place to make a small posting — I need X — and in 2 days it's done, that's it." (~1:13)

## 2. The core concept

A **gamified internal task marketplace / job board** ("we're literally describing an RPG" — Alex, ~1:12):

- Anyone with a need posts a **task** publicly, with clear criteria, template, and expected result — "here's the template, here are my objectives, go for it". No meetings, no WBS.
- People on the bench / in PRD time pick tasks up, **submit solutions**, and the poster **accepts a winner** — points transfer on acceptance.
- **Democratize access to work; centralize/open up the decision of what gets worked on.** Everything visible to everyone — the opposite of closed-door staffing.
- Leaderboards + rewards make it a game people *want* to play; the community self-organizes.

## 3. Mechanics discussed

| Mechanic | Detail | Where |
|---|---|---|
| Public posting | Task with criteria, template, objectives, deadline. Open to all of Customer or a sub-offering. | 1:07, 1:29 |
| Submission & review | Multiple people can submit; poster reviews and picks the winner; points transfer on accept ("I pay on the button"). | 1:11, 1:27 |
| Anti-spam | If you spam 500 low-quality submissions, the poster filters you out and ignores your answers. Review burden grows with granularity — needs a management/review layer. | 1:11, 1:30 |
| First-come-first-serve / competition | Top-down structure defines the project, deadlines, deliverable order; then it's open competition for points. | 1:05 |
| Bottom cut | Alex's aggressive version: cut the bottom 20% of the leaderboard each cycle. | 1:05 |
| Onboarding module | An AI module (in Polaris) generates a ~3h onboarding course from primary project material; passing it is the gate to start earning points on that project. | 1:05 |
| Priority boosting | Pay more to raise your task's priority (karaoke tipping analogy: tip more, get picked sooner). | 1:09 |
| Posting costs | Posting isn't free — it costs units/tokens, scaled to the points offered. Prevents someone from AI-splitting a whole project into 1,000 tasks and flooding the market. Counterweight: an attractive 10h job could burn 500 bench-hours across 50 people. | 1:15–1:16 |
| Private postings | Rent-a-coder memory (Elie): private bids/projects offered to a single chosen developer; escrowed funds released on delivery. Kept as a later-phase idea. | 1:26 |
| Auctions (V2) | Reverse auction: post a task with no fixed points; people bid "I'd do it for N points, here's my skill profile". | 1:27 |
| Signaling / self-organization | Mario Maker analogy: community boards showing who's working on what, so people don't duplicate effort; top runners share tips on hard problems in Discord-style chat; hard tasks attract collaboration. | 1:16–1:18 |
| Moderated, not free-for-all | Leadership watches the game: "5 people tried this problem with 3 different solution paths" is itself valuable insight. | 1:17 |
| Skill profiles | Ties into the earlier CV/skills idea: everyone maintains a profile of skills and preferences; track record on tasks validates claimed skills. | 54:00, 1:01, 1:27 |

## 4. Economy & currency

- **Points** earned per accepted task. Placeholder name in the meeting: **"Horizon point"** — deliberately an invented unit worth ≈ $1,000, so project budgets can be denominated in it ("Polytechnique's consultant layer becomes AI: $20–40k of tokens = N Horizon points; I budget 3 points for this task"). (~1:20)
- **Lottery draws**: every point = a raffle ticket. Weekly $1,000 draw, monthly $10,000 draw. "People will throw themselves at it." Hub-seeding example: 1 point for posting a use case, 1 point per page visit — Nav would post 300 in a day. (~1:08, 1:10)
- **Green Stars** to leaderboard tops each period (existing recognition currency). Ask Yoann for **$25k loaded in Green Stars** to seed the system. (~1:08, 1:15)
- **Util relief** as an alternative reward: completing tasks counts toward utilization at review time. (~1:13)
- **Performance-review integration**: Yoann declares at mid-year/year-end that points count. 50% util + many points = "your PRD was extremely productive". 50% util + 0 points + on no opportunities = "you're wasting my time", managed out. Hard edge: 3–6 months on the bench with zero points and no staffing = out. Cash + mid-year promos for top scorers outbid the dead patronage promise ("do what I say, I'll promote you" — empty for 2 years). (~1:06, 1:14, 1:25)
- **Why the ~$1,000 unit is natural**: AI is replacing the junior-consultant layer anyway, so projects already carry a real token cost to charge (e.g., $20–40k on Polytechnique) — that measurable cost is what gives the invented unit its anchor. Requirement: posting a task must hit the poster's project budget, or the marketplace is free internal labor. (~1:20)
- **The wage-theft trap** (Alex): if points map to real hours/money, paying 5 billable-hours-worth for a task that took 20 hours is codified wage theft — lawyers would object that people paid 40h/week demonstrably work 80. Mitigations discussed: keep salary fixed and frame points as *bonus* on top of the expected 40h, pay in vacation, or keep currency symbolic (points → draws → recognition). (~1:21)
- **The rake**: resource owners (Ali) charge a rake on hours, so converting bench→billable via the marketplace keeps them happy; Horizon could take a cote/commission on hours flowing through the tool while HR keeps owning the people. (~1:12, 1:18)

## 5. Rollout plan

1. **V1 — "bonne franquette" (keep it dead simple).** Public posting + submissions + accept button + point transfer + a small leaderboard. No auctions, no private bids. (~1:27)
2. **Pilot on Polaris.** "We struggle to get Polaris tasks executed — make a Polaris contribution leaderboard." Then AI Transformation tasks ("I have transcripts, I need a two-pager — I post it and see what happens"). (~1:27, 1:32)
3. **Seed the players.** People who publicly state AI ambitions in their year-end aspirations (QR code / snapshot machinery Alex is building) get plugged into the leaderboard. (~53:00, 1:29)
4. **Widen slowly** across Customer instead of a big-bang rollout. (~1:28)
5. **Chief-of-staff positioning.** If Claudia becomes chief of staff of the bench, the first act is directing bench people into this funnel; later, business development tasks too. (~1:28)

## 6. Strategic value

- **Talent discovery pipeline.** People who rack up points self-identify as strong; take them off bench into billable Horizon work ("Swift billable"). No staffing fight — "I didn't even request anyone; he picked himself via a bounty." Ali never complains because bench→billable keeps his rake flowing. (~1:18)
- **Performance filter.** Premium on efficiency and AI skill — good people "rise to the top immediately". People with no points and no billables become visibly unstaffable; the firm gets a defensible basis to manage out and replace via contract-to-hire. (~1:18, 1:00)
- **Pod-lead emergence.** Top scorers who show vision ("I want all the blue cards, I understand this") get promoted to pod leads — the leadership panel forms itself. Replaces the Carson group. (~1:30)
- **Influence.** Whoever runs the machinery runs the bench. "It puts us in the position of a useful version of ARM"; within 1–2 years the majority works "the Horizon method" — consulting's future. (~1:12, 1:19)
- **Delicate politics**: Livia, Davin and Ali own those people/bandwidth today — the tool must use their idle time *to their benefit* (rake preserved, HR ownership untouched). (~1:12)
- **Near-term answer to Yoann's agentic-pods fixation**: Yoann (after Martin Jeffers' agentic-harness demo) wants "teams that are one manager delivering a project end-to-end". Real implementations are months out — the job board is the realistic near-term structure that discovers and legitimizes pod leads instead of appointing them top-down. (~55:00–57:30, 1:31)

## 7. Open questions / risks

- **Currency realism**: how symbolic vs. real-money should points be? (wage-theft/legal exposure, market distortion, "I pay 5 for what takes 20")
- **Market health**: pricing tasks right (crowd may ignore underpriced deadline-driven tasks); avoiding flooding (posting costs); avoiding wasted parallel effort (50 × 10h on one job) vs. valuing parallel exploration.
- **Review burden**: more granularity → more submissions → more review effort; needs moderation capacity.
- **Favoritism doesn't vanish**: a poster can still always pick their friend as winner; preferences are partly legitimate (project context/knowledge).
- **Quality gate for what gets posted**: things should pass through a curating team (like the building team) rather than pure free-for-all.
- **Funding**: who loads the prize pool (ask: $25k in Green Stars from Yoann); how posting budgets map to project budgets.

## 8. Analogy bank (used in the meeting)

- **Rent-a-coder / vWorker-style bid sites** — bidding, private projects, escrow, deadlines (Elie's youth).
- **Bug Crowd / bug bounties** — post systems/tasks with bounties; people build a CV from solved bounties.
- **Mario Maker server-shutdown community** — self-organized boards of who's-working-on-what, tip-sharing on hard levels, no duplicated effort.
- **Karaoke tipping** — pay more to jump the queue.
- **Guide/job boards & the hub** — post a need, crowd responds, points per contribution/visit.
- **RPG** — "we're literally describing an RPG right now."