diff --git a/docs/pitch/PITCH.md b/docs/pitch/PITCH.md new file mode 100644 index 0000000..8d00618 --- /dev/null +++ b/docs/pitch/PITCH.md @@ -0,0 +1,76 @@ +# Sky Lattice — AI Hackathon Pitch (Narrative) + +Speaker notes and copy you can trim. Slides: open [`index.html`](index.html) in a browser (`←` / `→`). + +--- + +## 30-second pitch (elevator) + +Snowflake gives you a blank cloud. Every customer, we rebuild isolation, RBAC, and data-product layout by hand — lately with AI chats that forget *why* we made exceptions. **Sky Lattice** is the layer under the copilots: a versioned blueprint, customer intent, durable decisions, and a deterministic planner. AI fills the forms; Terraform still ships the platform. + +--- + +## 2-minute pitch (spoken narrative) + +**Hook.** +We’ve all felt this: Snowflake looks empty, best practices live in people’s heads, and “let’s ask the LLM” turns into a week of clever SQL that nobody can re-run with confidence. + +**Problem.** +Setting up a real platform — environments, zones, roles, warehouses, service users — is repetitive but never identical. Today’s AI-assisted workflow accelerates the typing. It does **not** create institutional memory. Terraform state tells you what exists. The next chat session still re-litigates judgment. + +**Insight.** +Don’t fire the LLM. Put a lattice under it. Codify the recipe. Capture deviations as decisions. Make the agent drive files and a CLI — not invent production IAM from scratch every time. + +**Product.** +Sky Lattice is that lattice: +- a **blueprint** of how *we* set up Snowflake +- per-customer **intent** and **decisions** +- **`platformctl`** to validate, plan, drift, and apply +- optional **Cursor/Claude skills** as the natural-language front door + +Same path for greenfield and brownfield: interview → durable artifacts → deterministic plan. + +**Why us / why now.** +Platform teams are already using AI for infra. The winners won’t be who prompts harder — they’ll be who productizes the delivery system. We’re turning consulting tribal knowledge into software. + +**Status.** +Working v0.1 scaffold: wizard, decision store, planner, inventory drift, skills, demo customers. Next: live Terraform→Snowflake vertical slice. + +**Close.** +Sky Lattice — structure under the copilots. Help us harden it. + +--- + +## Slide-by-slide talking points + +| # | Slide | Say this | +|---|---|---| +| 1 | Title | Name + one-liner. Don’t over-explain yet. | +| 2 | Problem | Blank slate + amnesiac AI. Point at cost: time, inconsistency, audit. | +| 3 | Insight | Quote beat. Pause. | +| 4 | Product | Show the flow: NL → skill/intent → plan → TF. | +| 5 | Demo arc | Walk Acme story; emphasize *decision* as the magic. | +| 6 | Differentiation | Four cards — AI placement, memory, deviation, friction. | +| 7 | Market | Consultancies + internal platform teams. | +| 8 | Traction | Honest v0. What works / what’s next. | +| 9 | Ask | Design partners / judges who feel the pain. | + +--- + +## Objection handling + +| Objection | Response | +|---|---| +| “Just use Terraform modules.” | Modules are the engine. We add intent, decisions, wizard, policy, and an AI contract so modules get used the same way every time. | +| “Just use the LLM.” | We do — as the driver. Without durable intent/decisions, you get speed without repeatability. | +| “SnowDDL / DataOps.live already exist.” | Engines and platforms exist. Our IP is *our* blueprint + engagement memory model — and it’s agent-native. | +| “v0 doesn’t apply to Snowflake yet.” | True and intentional. We proved the control plane first; next slice is live apply. Judges should score the architecture + working demo path. | + +--- + +## Taglines (pick one) + +- Structure under the copilots. +- Snowflake platforms with memory. +- Codify. Tweak. Maintain. +- AI drives the forms. The lattice holds the truth. diff --git a/docs/pitch/index.html b/docs/pitch/index.html new file mode 100644 index 0000000..f58c0cc --- /dev/null +++ b/docs/pitch/index.html @@ -0,0 +1,409 @@ + + + + + + Sky Lattice — AI Hackathon Pitch + + + + + + +
+ +
+

AI Hackathon · Platform Engineering

+

Sky Lattice

+

The missing layer between AI copilots and Snowflake — so platform setup is repeatable, auditable, and still human-fast.

+
+ Intent + Decisions + Blueprint + Terraform + AI Front Door +
+
+ +
+

The problem

+

Snowflake is a blank slate.
AI chats are amnesiac.

+
+
+

Today

+
    +
  • Every customer rebuild starts near zero
    • +
  • RBAC, envs, zones = tribal knowledge
    • +
  • LLM + Terraform: works once, forgets why
    • +
  • Follow-ups reopen a new chat with TF state, not judgment
    • +
+
+
+

Cost

+
    +
  • Weeks of senior time per engagement
    • +
  • Inconsistent security posture
    • +
  • Hard handoffs, harder audits
    • +
  • AI accelerates typing — not institutional memory
    • +
+
+
+
+ +
+

Insight

+

“Don’t replace the LLM. Put a lattice under it.”

+

Codify the recipe. Capture the why. Let AI drive the forms — never invent production IAM from scratch each session.

+
+ +
+

What we built

+

Sky Lattice

+

An internal toolkit that turns “how we set up Snowflake” into a versioned product: blueprint + customer intent + durable decisions + deterministic plan — with Cursor/Claude as the optional front door.

+
+ Natural language + → + Skill + intent/decisions + → + platformctl plan + → + Terraform / Snowflake +
+
+ +
+

Live story

+

Greenfield in minutes.
Brownfield without amnesia.

+
    +
  1. +
    + Init Acme + platformctl init + wizard → domains, envs, restricted finance +
    +
    2. +
  2. +
    + Plan the lattice + Deterministic object graph: DBs, zones, roles, warehouses +
    +
    3. +
  3. +
    + Record the why + Decision file: temporary exception, expiry date, rationale +
    +
    4. +
  4. +
    + Come back later + New chat opens the customer repo — not a blank brain +
    +
    5. +
+
+ +
+

Why this wins

+

Not another chatbot.
Not another Terraform module dump.

+
+
+

AI where it belongs

+

Skills interview and edit intent. Policy blocks unsafe shortcuts. Apply stays deterministic.

+
+
+

Memory that survives sessions

+

TF state = what. Decision store = why. Intent = what we mean to run.

+
+
+

Repeatable with deviation

+

Same blueprint, different customer params and approved overrides — not copy-paste HCL.

+
+
+

Low friction for Snowflake people

+

Five verbs. Snowflake vocabulary. Familiar plan output. Optional agent IDE.

+
+
+
+ +
+

Who cares

+

Anyone shipping Snowflake platforms more than once.

+
+
+

Primary

+
    +
  • Data platform consultancies
    • +
  • Internal platform / enablement teams
    • +
  • Multi-account Snowflake orgs
    • +
+
+
+

Pain they already feel

+
    +
  • “We reinvent RBAC every client”
    • +
  • “AI helped, then we lost the thread”
    • +
  • “Brownfield is a black box”
    • +
+
+
+
+ +
+

Hackathon status

+

Working scaffold. Honest v0.

+
+
+
v0.1
+

Blueprint · platformctl · wizard · decisions · drift inventory · Cursor skill · demo customers

+
+
+

Next vertical slice

+
    +
  • Live Terraform → Snowflake apply
    • +
  • Overrides that change the plan graph
    • +
  • Live account discover
    • +
+
+
+
+ +
+

The ask

+

Help us harden the lattice.

+

We’re building the control plane for AI-assisted data platform delivery. Join as design partners, Snowflake SMEs, or judges who’ve felt this pain.

+
+ Open source trajectory + Consulting IP today + Product tomorrow +
+

Sky Lattice — structure under the copilots.

+
+ +
+ +
+ Sky Lattice + 1 / 9 + ← → or click +
+ + + +