AI Department · Unicity · internal
What we're building, why, and what it's producing.
A Forward-Deployed-Engineering function for non-engineers across Unicity. It teaches AI fluency, governs how non-devs build with AI, builds the data foundations that make AI work, and sets Unicity's AI posture — on one ethos: AI unlocks people, it does not replace them.
Own the context. Swap the AI. Show your work.
not measured rather than guessing.Purpose
The mission
Make Unicity the most AI-fluent, AI-secure, AI-native wellness and e-commerce company — by enabling every non-engineer to learn, build, and work with AI productively; governing how they build to protect company data; building the operational foundations that make AI actually work; and shaping Unicity's AI posture across markets and channels.
source: brain/charter/mission-and-pillars.md
How it's organized
The four pillars
P1 builds the people, P2 contains the risk, P4 lays the data substrate, P3 aims all of it at the right bets. No single pillar produces Unicity's AI capability — the four together do.
Fluency & Culture
Make AI fluency a deliberate capability across non-engineering, not something that depends on who happens to be curious. Delivered by leverage: high-performers first, then department groups, then baseline literacy. (Three-tier path: Awareness → Power User → Builder.)
Vibe-Coded App Governance
Govern, don't ban, how non-engineers build with AI. A registry of every vibe-coded app touching company data, a sanctioned tools list, and graduation gates that trigger Security/Legal review at defined thresholds. The validated risk: a vibe-coded app already leaked credentials. Co-governed with Security/IT and Legal.
Strategic Positioning & Bet Selection
Of everything AI could do for Unicity, what is worth doing, and in what order. Identifies and prioritizes the opportunities; owns how Unicity talks about AI externally (grounded outcomes first). Ships no code; points the other three pillars at the right targets.
AI-Native Foundations — Make Unicity Queryable
Make everything legible to an intelligence layer. This is the brain vault. PCC is the working prototype; P4 scales the pattern: own the raw context, organized so any AI can read it, with every answer citing its source. This workspace is P4 in motion.
source: brain/charter/mission-and-pillars.md
The model in one picture
Own the context. Swap the AI. Show your work.
Every company is racing to build an AI that synthesizes their data. That's the wrong thing to own — models keep changing and it locks you to a vendor. The durable asset is the raw context itself, organized so any AI can read it, every answer showing exactly where it came from.
Raw signals
meetings, docs, tools — never copied out of their source
The brain
curated pages that point down to the raw
Whatever model is best
reads the owned context — not locked to a vendor
Cited answer
every answer links back to the raw
source: brain/vision/north-star.md
How we're judged
The KPI framework
The CTO set six results for the department. The cards below are the working measurement design. One distinction is load-bearing: the six statements are approved direction; the IDs, formulas, and thresholds are v0.1 working definitions until their named owners confirm them — they are not CTO-approved semantics.
No authoritative actuals ledger exists yet, so every KPI's current value is not measured. This site measures what it honestly can (starting with the HR Assistant) and leaves the rest empty rather than fabricating a number.
Sustained department adoption
AI adoption in most departments
Working definition
Propose ≥60% of a frozen eligible-department roster, with ≥1 approved workflow active on real work for ≥60 days or 3 normal cycles and evidence of repeat use, value, quality, and governance. Licenses, training, demos, and one-off use do not count.
Verified capacity released
10,000+ annual hours saved
Working definition
Show trailing-12-month realized hours and verified annualized run-rate separately. Count net human time after review, rework, exceptions, implementation, and maintenance. Hours are capacity—not automatically dollars.
Verified annual business impact
At least $500K–$1M in quantifiable annual business impact
Working definition
Propose $500K as floor and $1M as stretch; keep realized impact separate from annualized run-rate; require Finance validation for material claims.
Active production automations
20+ production automations
Working definition
Propose counting distinct active end-to-end workflows—not prompts, scripts, agents, steps, revisions, or model swaps—with owners, controls, production history, and measured quality/value.
Company-wide AI governance
Company-wide AI governance framework
Working definition
Treat framework delivery as the first outcome; then track control coverage. Proposed coverage thresholds are 100% of high-risk and ≥95% overall.
Executive AI strategy stewardship
Executive AI strategy ownership
Working definition
AJ stewards the enterprise AI strategy, evidence, and portfolio-review process and makes fund/pilot/scale/stop/retire recommendations. Approval rights remain with the executives named for each decision.
Accepted outcome rate
Preserve or improve the pre-AI quality baseline
Working definition
Outputs accepted without material correction divided by completed outputs; each workflow sets its threshold at or above its pre-AI baseline.
Material control failures
No value claim overrides a failed production gate
Working definition
Track Sev-1/Sev-2 incidents, knowingly bypassed controls, open critical exceptions, and time to remediate.
source: content/brain/goals/metrics.md · registry v1 · updated 2026-07-09 · confidence medium
Operating principles
How we work
- 01
AI unlocks people; it does not replace them.
Dan's commitment: no AI-driven role elimination. Every deliverable is judged 'did this unlock people or threaten them?' Klarna (replacement → quality collapse → reversal) is the anti-pattern; Moderna's grounded-outcomes pattern is the model.
- 02
Support ideas and move fast — under safe, sustainable architecture.
Keep enabling fast tests and deploys; when an idea is big and touches critical data/systems, build it properly (still at AI speed) by people who can make it safe to scale.
- 03
Find and multiply high-performers first.
Concentrate on the high-leverage people; broad literacy runs in parallel. Multiply the ones who'll run with it; let the signal cascade. (Dan's principle, from the view across 50+ markets.)
- 04
Demonstrate before mandating.
Prove a capability on ourselves first, then extend only where there's appetite. No mandatory enterprise rollouts in year 1. (PCC, and now the brain vault, are the demonstrations.)
- 05
Graduate departments, don't create dependency.
FDEs embed, leave a trained champion, and exit. The exception is software we built that a department runs: that gets a permanent, dedicated support resource.
- 06
Discretion on people signals.
Identifying high-performers is done with managers and HR, criteria documented. People whose roles AI could threaten are approached by showing what AI takes off their plate, not by lecturing them.
source: brain/charter/operating-principles.md
Scope
Who this is for
Everyone at Unicity outside the engineering org: Marketing, CS, Finance, Legal, HR, Supply Chain, Distributor Ops, Sales, IT ops, exec teams. Engineers already use Claude/Codex/Cursor under their own standards.
- Not for Engineering — they run their own AI house.
- Not a research lab.
- Not the team that ships customer-facing AI features.
The primary archetype is the FDE (Palantir/Anthropic/OpenAI lineage): technically capable people who embed with non-engineer teams, build the automations, teach in context, and leave trained champions behind. Not consultants, not pure engineers, not pure trainers — they build while teaching.
source: brain/charter/scope.md