Lucy coaches the people closest to the problem — domain experts, people with lived experience — through the design discipline that turns their context into useful AI-supported solutions.
Lucy is a cloud-backed desktop application that lives in the flow of your work. Always available, never another tab to remember, contextually aware of what is actually happening on your screen so coaching is specific instead of generic.
On trust, Lucy observes Signal and Context, never Content without explicit invocation. See the V·C·C posture in full →
The Coaching Loop: Observe → Spot the AI opportunity → Personalize a plan → Coach → Iterate. Lucy's repeating user-facing engagement, running in flow with you. Most coaching moments start proactively (Lucy surfaces an opportunity) or reactively (you bring her a task) — both use the same loop. The loop runs inside every Evolutionary Framework stage; it is most visible during Adoption because that is where habits form, so its canonical home lives here.
Each pass produces either a small win — the agent for this opportunity — or a learning: this opportunity wasn't right, here's why. The loop runs continuously while Lucy is active. Lucy is quiet during deep work, present when you ask, conservative by default — coaching intensity is planned as a user-side control.
Users progress through Trust → Personal value → Adoption → Capability → Scale. The five branches below organize Lucy's mechanism content along that spine, plus the research grounding and what Lucy teaches.
Lucy starts with your reality, not a template. A short first-session interview maps your work; a two-week observation period learns how it actually flows. The Signal · Context · Content data taxonomy keeps coaching personal without crossing the trust line.
On Day 15 Lucy delivers a Personal Coaching Plan built from the interview, the observation, and your current capability baseline. The first useful win comes from one high-impact, low-effort task you already do frequently — proof you can feel in the same week.
The Coaching Loop runs in flow with you: Observe → Spot the AI opportunity → Personalize a plan → Coach → Iterate. Lucy adapts the kind and frequency of coaching to your pace — step-by-step when you're new to a pattern, lighter-touch when you're moving fast, quiet when you ask her to sit back.
Lucy teaches the pattern, not just the answer — the structure that made the prompt work, the heuristic for the next variant, the principle behind the iteration. You become a solution builder for your own real work. What you learn and build with Lucy is yours; it travels with you to your next role and the next model generation.
Successful patterns built by the people closest to the work become visible to others — attributed to their original builder, reviewable before broad adoption, and forkable by adjacent teams. Lucy coaches the pattern as it spreads so depth carries with it. Governance, approved tools, and data residency travel along, not around.
For the per-stage commitments and roadmap labels, see the trust architecture — it carries the stage-honest labels per item.
When you first install Lucy, the first session is a structured conversation about your work — what you do, what slows you down, what's repetitive, where the meaningful judgment lives, what you're already trying to use AI for. The interview is conversational, not a form. Lucy uses what you say to shape what it observes next.
The 2-week observation period — after the interview, Lucy observes how you actually work (workflow patterns, tool usage, where execution time goes), with low-friction coaching capped at one nudge per day. You can engage her any time; she just does not push. What she learns shapes the Personal Coaching Plan delivered on Day 15.
Lucy distinguishes three kinds of data. The boundary is enforced at the architecture level, not as a setting toggle.
The first-session interview, the observation period, and the Signal·Context·Content boundary are available. Additional user-side controls — disable retention, redact, delete profile, export profile, coaching intensity — are planned. Manager-side admin controls are planned for GA.
Inside the first session, Lucy proposes one specific moment in your real work where AI could save measurable time today. Lucy picks the smallest moment that gives you back at least 15 minutes a week, and walks you through implementing it once.
Useful is operational: time saved, repeated decisions made faster, output quality lifted on a task you do regularly. Useful is not "you used the product." Useful is "your work is measurably easier in this one specific way, starting now."
Around Day 15, after Lucy has observed enough of your real work, Lucy delivers a Personal Coaching Plan — a personalized roadmap of the AI-leverage opportunities in your specific work, prioritized by time-savings and feasibility. The PCP is the artifact you keep; the Coaching Loop above is the mechanism that moves you through it.
The biggest predictor of whether someone keeps using an AI tool is whether they got value from it on day one. Lucy is engineered for the day-one win, not the day-thirty epiphany.
Lucy watches the Signal and Context layers of your work — application activity, patterns of repetition, time-on-task. The observation is read-only.
Lucy identifies a specific AI-shaped task — meeting prep, recurring report, categorization, draft. The proposal is concrete: "the meeting prep agent we'd build would save 35 minutes per Tuesday."
Lucy proposes a plan tailored to your specific tools, data sources, and risk posture. The plan is editable; you say what you want different before any agent is built.
Lucy walks you through the build — what tool to use, what to test, what to verify, when to ship the agent into production. You do the building; Lucy coaches the design discipline.
Once the agent runs, Lucy observes whether it actually saves the time it promised. If it does, the loop moves on. If it doesn't, you debug it together — the failed attempt is part of the learning, not a hidden statistic.
See the trust architecture for what Lucy observes (Signal · Context · Content boundary) and what it refuses to observe.
When the Coaching Loop walks you through building an agent, the coaching is structured around the underlying design pattern — not just "do this, then this." The pattern is the part that transfers to the next agent you build. After ten loops, you're using Lucy to refine work you can already shape on your own.
The knowledge and skills everyone needs: the patterns Lucy coaches inside real work so you can use AI well across products, providers, and model generations.
Tool examples: ChatGPT, Claude, Gemini; Microsoft 365 Copilot; Perplexity and NotebookLM.
Once the foundation is working, Lucy adapts the coaching to the job in front of you: the domain, workflow, deliverable, risk, and standard of quality your role actually requires.
Tool examples: Cursor, GitHub Copilot, Claude Code; Elicit, Consensus; Gamma, Beautiful.ai; Julius, Hex; Midjourney, Runway; n8n, Make.
This is where Lucy turns AI use into durable capability: you learn how to shape a real problem, choose the smallest useful solution, compose approved tools, and save the pattern so it can be reused and improved.
Know when the work needs a prompt, a coached workflow, a small agent, or custom software.
Wire approved tools through APIs, connectors, and reviewable steps where orchestration creates the value.
Define scope, permissions, human checkpoints, data boundaries, and approval rules before the solution runs.
Tool examples: LangGraph, OpenAI Agents SDK, Claude Agent SDK, CrewAI; n8n, Zapier AI, Make; approved business-stack connectors and MCP services.
What you learn and build with Lucy is yours. The capability, the prompts, the workflows, the agents — they go with you to your next role, your next employer, and the next model generation.
When agents built in one team work measurably well, the pattern (not the agent itself, the design pattern) becomes visible at the company level — surfaced as the kind of work other teams might be doing similarly. Visibility is opt-in by the builder and reviewable by the manager and the works-council representative before any cross-team rollout.
The people who built the agents that worked become the champions of the next wave — not because they're appointed to that role, but because they have the lived experience of building. Champion emergence is bottom-up; Lucy makes it visible to the company without forcing it.
As patterns prove out, they enter an opt-in Best Practices Library — searchable by other teams, with clear attribution to the original builder, with the data context that made the pattern work in its original setting.
The 80/20 reality: general tools solve the shared 20%. Lucy helps with your 80%.
Your business stack is becoming agent infrastructure. Salesforce, Microsoft, OpenAI, Anthropic, and MCP-connected services are turning the AI layer into a stack of tools that compound when you route work correctly.
Lucy coaches you on routing literacy: how to read the shape of a task and the shape of a tool, when to keep your default where it works, when to layer in a specialist, and when to stack agents on top of systems you already trust.
The expensive thing is not having multiple tools. The expensive thing is routing the work incorrectly.
See the compressed five-card differentiator grid on the homepage.
Vertical applications — how the framework lands across specific industries and roles. These are combinations of UA + SD applied to a vertical's work, not standalone domains.
Six representative domains where Lucy's design partners are building. The mechanism is the same; the surface area changes.
Prospect research, account-plan generation, RFP responses, opportunity hygiene.
NLP / Whisper / TTS / RAG-backed voice agents that support hands-on work (mechanic / nurse / field tech) without replacing it.
Triage, runbook-grounded answers, resolution coaching for the responder.
Visit-summary drafts, ICD-coded suggestions, structured discharge instructions; clinician in control of the final document.
Role-fit screening prompts, candidate interview prep, structured candidate notes.
The same Coaching Loop applied to non-work: language learning, fitness habit, household administration.
The list is not exhaustive — Lucy applies wherever a domain expert is doing repetitive, judgment-heavy work that would benefit from being agent-supported.
