Native to your work surface
Lucy lives on the desktop where the work actually happens — across Salesforce, Gmail, calendars, docs, design tools, terminals — not locked inside a single web app you have to remember to open.
Product
Lucy makes you a builder, not a tip-list reader.
Lucy turns you into a solution builder for your own real work, not a passive consumer of generic AI tips. She spots the moments where AI could actually help, coaches you through building the smallest useful solution yourself, and leaves the capability with you.
AI does not transform businesses. People transform businesses with AI. Lucy is the missing link between the people doing the work and the AI tools that could revolutionize it.
How you experience Lucy
A quiet desktop coaching suite, not another notification source.
Lucy is a cloud-backed desktop application that sits alongside the tools you already use. A small, quiet icon in the menu bar. A floating chat that opens when you ask, never barging in. She is non-pushy by design — "Maybe later" and "No" carry exactly the same weight as "Yes."
Visible only when you want her
A small icon in the menu bar. No popups, no streaks, no engagement metrics pulling at your attention. You open the chat when you have a question; otherwise she stays out of the way.
"Maybe later" and "No" carry equal weight
Lucy proposes; you decide. Saying no does not move you down a leaderboard. Saying "not now" saves the idea for a better moment instead of nagging. Lucy honors quiet hours, daily limits, snooze, and dismissal feedback.
Observation is permissioned and boundary-controlled. The point is personalization, not employee monitoring. No private prompt feeds, no individual scoreboards, no productivity ranking. Aggregate patterns flow to leaders; individual recognition is opt-in.
The first 15 days
Before Lucy teaches, she listens.
Nobody cares what an AI knows until it understands what they care about. Lucy starts by learning your goals, tools, constraints, preferences, and the moments where AI could actually help.
The first goal is not to drop you into a training path. It is to earn permission, understand your work, establish a useful baseline, and find AI-supported quick wins that prove Lucy can help.
Day 15 roadmappersonalized
Day 15
Priority workflowCustomer meeting prep
Matched to your goals, available approved tools, and onboarding preferences.
Next 3-5 actions
Draft better questions, review assumptions, save the reusable prep pattern, and decide what to try next.
Saved for later
Skills, workflows, tools, agents, automations, prompts, or solution patterns Lucy can bring back when timing improves.
Lucy takes time to understand you before she asks you to change how you work.
Activate Lucy
Account activation, desktop app install or MDM deployment, and permission confirmation.
Motivation interview
Lucy asks about role, goals, pain points, tools, preferences, and what would make work better.
Persona confirmation
Lucy reflects the likely role pattern and lets you confirm or correct it.
Welcome and privacy
Boundaries, data handling, and sharing preferences are visible from the start.
Baseline assessment
A low-stakes task checks prompting, iteration, evaluation, tool selection, and risk awareness.
Quick-win candidates
Lucy identifies practical tasks or learning goals that can create early Personal value.
What Lucy learns, and how.
SourceWhat Lucy learnsBoundary
User-provided contextGoals, pain points, motivations, learning preferences, tools, role, and starting capability.You can skip, correct, or update it.
Approved environmentApproved tools, restricted tools, trusted context categories, policies, and review boundaries.Policy and permissions shape recommendations.
User-granted work contextCalendar details, documents, notes, tickets, CRM views, manuals, files, screen-visible context, or other context you allow, provide, or make visible.Use is visible and tied to a clear purpose.
Observed workflow patternsRepeated work, bottlenecks, copy/paste loops, reports, switching, and AI-help moments.Pattern-finding, not surveillance.
User feedbackWhat helped, what was annoying, and where you want more or less guidance.Lucy adapts instead of forcing a fixed program.
You can start with one task, one frustration, or one thing you want to learn.
You name or confirm a task, decision, project, frustration, or learning goal.
Lucy asks enough follow-up to understand how the work happens today and what help would be worthwhile.
Lucy identifies the opportunity type: better thinking, context, tool use, workflow, automation, agent, script, or app.
Lucy proposes the simplest safe solution path and coaches the first build.
For customer-call preparation, Lucy can start with a coached meeting-prep workflow using materials available to you, then later help turn the pattern into a reusable workflow or agent.
Adaptive coaching modes
Coaching that meets you where you are.
Lucy reads how comfortable you are with the pattern in front of you and shifts her style accordingly. Same person, same Lucy, four different modes depending on the moment.
01
New to the pattern
Step-by-step coach. Where to click, what to enter, why each step matters, what to check, and what to do if something looks off. Lucy walks beside you on the first build.
02
Comfortable with the tools
Lighter-touch guide. A recommended path, examples, review points, and help if you get stuck — but the keyboard stays in your hands and Lucy does not over-explain.
03
Experienced or moving fast
Design partner. Pressure-tests tradeoffs, surfaces failure modes, names approval gates, flags context risks, and challenges architecture choices. The conversation gets sharper.
04
Asking Lucy to sit back
Quiet coach. Honors pause, quiet hours, daily limits, snooze, and "not right now." Saves the idea for a better moment and re-engages only when invited.
Adaptive does not mean noisier. Even in design-partner mode, Lucy stays out of the way until you ask.
Solution range
Solutions are continuous states, not one-off tasks.
Solutions are desired continuous states (I want to always be prepared for meetings) — not one-off tasks (I need to prepare for this meeting). The task is one instance. The solution is the persistent pattern that makes the desired state easier to reach every time.
One-time taskOngoing desired stateSolution path
Prepare for this customer meeting.Always be prepared for customer meetings.Start with a coached workflow; save the pattern; consider an agent only when it repeats.
Solutions take many shapes.
Solutions could be bounded, trustworthy agents or agentic workflows (preferred), simple prompts, coached workflows, AI-assisted research, AI as a brainstorming partner, automatons, scripts, or custom software. Lucy helps you pick the smallest useful shape and grow only when the pattern earns it.
Bounded agents and agentic workflows
Persistent patterns with memory, tools, state, repeated steps, and human checkpoints. Best when the work repeats, has enough complexity to justify a loop, and benefits from an approval gate.
Simple prompts and prompt patterns
Reusable thinking, writing, critique, or planning moves. The right shape when the pattern is small, you provide the context each time, and human judgment is the main work.
Coached workflows and AI-assisted research
An approved tool can do the work if you know the right sequence and context. Source-grounded synthesis, comparison, and decision support belong here too.
AI as a brainstorming partner
Tradeoffs, options, dissent, counterarguments, and quality checks. Lucy helps you turn AI from an answer machine into a thinking partner that sharpens your own decisions.
Automatons, scripts, and small software
Stable repeatable steps with low ambiguity become automations or scripts. When a small utility or specific work surface fits better than a large platform, Lucy coaches that build too.
AI is still a tool. Human judgment decides when to use it, how much structure it needs, and whether a heavier system is worth building. Start with the smallest useful solution that can be trusted, reused, and improved.
Worked example
How an agent gets built — Tom builds Jerry.
This is what coaching toward an agent actually looks like, end to end, in about two hours.
Tom is an account executive. Most Mondays disappear into prep for the week's external customer calls — pulling notes from Gmail, last-meeting summaries from Salesforce, fresh news from LinkedIn, and an outline of what to cover. Lucy notices the pattern over a few weeks: same prep flow, every Monday, copy-pasting between four windows.
One morning Lucy asks if Tom wants to build an agent for it, prep just this one meeting together, save the idea for later, or pass entirely. Tom picks "build the agent." Lucy starts by diagnosing how Tom prepares today — what he reads, what he writes, what he keeps, what he throws away — and reflects the pattern back. "It looks like prep is: calendar context, prior emails, account status, recent news, then a one-page brief. Does that match?"
It does. Lucy proposes a reusable agent — call it Jerry — that runs before each external customer meeting and produces a private draft brief. Approved tools only: read-only MCP connections to Calendar, Gmail, Salesforce, and a news source. No writing back into customer systems. No sending. Just a brief that Tom reviews.
Tom opens Codex. Lucy coaches the prompt pattern while Tom directs the build — what to summarize, what to skip, what tone, what length, what to flag. Codex assembles Jerry with the read-only MCPs, dry-run delivery turned on, and stop-conditions Lucy and Tom agree on. The first dry-run delivers the brief into a private Slack channel only Tom can see. He reads it, edits the prompt twice, runs it again, and the third version is the one he wants.
Tom saves Jerry with scheduling off. Next time he wants a brief, he triggers Jerry himself. When he is ready to let it run automatically before each external meeting, that is one toggle — but only after he has trusted the output enough times to flip it. Lucy remembers the coaching pattern so the next agent Tom builds takes less time.
Lucy spotted the opportunity. Tom owned the build. Codex assembled the agent. The capability — the prompts, the workflow, the agent — stays with Tom. See the live demo on the homepage →
Have Lucy show you what she'd notice in your week.
Talk to Lucy about a real task on your plate right now. She'll ask a few questions and propose the smallest useful solution shape — no install required to try the conversation.
Best-practice patterns Lucy coaches
HITL, dry-run, and scoping — built into every build.
Three patterns show up in every solution Lucy coaches because they are how AI work stays trustworthy. They are not bolted on after; they are part of the design from the first step.
Human-in-the-loop
Humans approve consequential action. Lucy never writes to customer systems on her own. Send, schedule, post, change, or delete are decisions a person makes, not the agent. The approval gate is visible in the build, not buried.
Dry-run before live
Test the agent with no side effects until you have seen the output enough times to trust it. Drafts go to a private channel, not a customer inbox. Scheduling stays off until you flip it. The third version is usually the one you keep.
Scope of access
Read-only by default. Named tools, named sources, explicit permissions — all visible in the build. If an agent needs broader access later, that is a deliberate, reviewable change, not a quiet one.
Capability in the wild
Six archetypes of what Lucy coaches you to build.
Concrete shapes for what "a Lucy solution" looks like in real work. Each one started as a one-off frustration and grew into a persistent pattern only when the repeat earned it.
Meeting Prep Agent
Lucy noticed prep burden across recurring customer calls. Pattern type: bounded agent. Sources: read-only Calendar, Gmail, CRM, news. Output: private draft brief. Boundary: human reviews before send; scheduling opt-in. Capability gained: a reusable prep flow that survives the next role and the next CRM.
Client Outreach Agent
Lucy noticed a draft queue building up across follow-ups. Pattern type: coached workflow with optional agent step. Sources: user-provided notes, examples, prior threads. Output: drafts queued for human tone-review and send approval. Boundary: Lucy never sends; the human approves. Capability gained: outreach quality without inbox triage every Monday.
Proposal Manager
Lucy noticed reusable proposal sections being rebuilt from scratch. Pattern type: pattern library plus review gate. Sources: approved templates, prior winning sections, source material. Output: assembled draft with margin checks and review gates. Boundary: pricing and commitments stay human-owned. Capability gained: a proposal workspace that compounds across deals.
Decision Partner
Lucy noticed leadership decisions made without a structured options view. Pattern type: prompt pattern plus repeatable workflow. Sources: user-provided context, dissent, assumptions, risk checks. Output: options matrix with explicit owner and review questions. Boundary: the call belongs to the decider. Capability gained: better thinking on the decisions that matter most.
Family Logistics
Lucy noticed calendar conflicts and pickup/drop-off churn outside work. Pattern type: light agent with strict approval gate. Sources: permitted personal calendars only. Output: conflict draft and proposed plan. Boundary: humans approve every change before anything is shared. Capability gained: a personal logistics pattern that frees attention for work.
Hands-On Work Support
Lucy noticed field technicians juggling diagnostics, manuals, and customer summaries on the move. Pattern type: voice-first coached workflow. Sources: trusted manuals, prior service notes. Output: checklists, diagnostics, and a customer-summary draft. Boundary: customer-facing communication stays human-reviewed. Capability gained: hands-free support without taking eyes off the work.
Each archetype is a starting shape, not a fixed product. The actual build is yours — coached at your level, against your approved tools, with the boundaries you set.
What Lucy teaches
Universal AI skills + specialized domains.
Roughly 70% of what Lucy teaches is universal — the AI literacy that makes any tool useful. The other ~30% is specialized depth in the domains where you actually work. The split shifts with you over time.
Universal AI skills
Prompting fundamentals
Eight prompting moves Lucy coaches in real work: Iterative & Chain-of-Thought, RTF / RTC, Self-Consistency, Instruction-Tuning-Style (ICE), Expert Persona, Decomposition & Planning (PER), Evaluation & Critique, and the Role / Task / Context / Format structure.
Effective iteration
Improve weak outputs without starting over. Where to add context, what to challenge, when to switch tools, and when to stop iterating because the result is already good enough.
Output evaluation & quality control
Judge whether the result is accurate, useful, complete, and appropriate before you ship it. Spot the failure modes specific to AI output: confident wrong, missing context, wrong audience, unsafe.
Context management
What to include, what to leave out, how to frame examples, and how to keep context tight enough to be useful but small enough to stay in scope. The work behind every good prompt.
AI-aware task decomposition
Break a goal into AI-suitable steps. Some steps need a human; some need a tool; some need a model. Decomposing the right way is most of the difference between a working solution and a stuck one.
Tool selection & orchestration
Pick the right approved tool — and the right combination — for the step in front of you. Avoid the all-purpose-hammer trap. Compose tools so each one does what it is best at.
Collaborative AI work
Treat AI as a partner you can argue with. Surface dissent, push back on weak drafts, ask for the version it didn't give you, and keep your judgment central in the conversation.
Agent design & use
Decide when an agent is the right shape. Scope tools and access. Add HITL and dry-run by default. Know when to retire an agent because the underlying pattern changed.
Specialized skills by domain
Twelve domains where Lucy goes deeper as your work demands it. The list grows with the field.
DomainWhat Lucy coaches
Code GenerationAI for Devs and Vibe Coding — pairing with AI on real code, scoping the prompt, reviewing the diff, and keeping ownership of the architecture.
Agent Building & AutomationDesigning bounded agents and durable automations, including HITL gates, dry-run discipline, and reuse paths.
Visual Content & Media CreationBrand-quality images, mockups, social and video assets — prompt patterns, reference handling, and review against brand standards.
Research SynthesisSource-grounded synthesis, comparison, and decision support that does not hallucinate the citations.
Data AnalyticsQuestion framing, data preparation, model-readable context, and evaluating analytical output before sharing it.
ProductivityBrainstorming, presentation design, and automating workflows — turning repeated personal patterns into saved patterns and small agents.
Personal GrowthInterpersonal relationships, career development, and education — using AI to think more clearly about the work and life around the work.
ITSupport and cybersecurity — runbooks, incident drafts, log triage, and configuration review with appropriate guardrails.
NLP / Whisper / voiceTranscription, summarization, voice-first workflows, and turning audio into structured artifacts you can act on.
Sales & Pipeline ManagementOutreach quality, account research, prep flows, and pipeline discipline without turning the CRM into a writing prompt.
HealthPersonal health information triage, question framing for clinicians, and using AI safely in a domain where accuracy matters.
RecruitingJob descriptions, screening rubrics, candidate research, and interview prep with bias-awareness guardrails.
~70% universal, ~30% specialized. Universal AI skills compound across every domain you ever work in. Specialized skills make today's work measurably better.
Portability
Capability stays with you.
What you build with Lucy is yours. The capability, the prompts, the workflows, the agents — they survive your next role, your next employer, and the next model generation.
Patterns you can take with you
Saved prompts, workflow definitions, and agent configurations belong to whoever built them. When you move, the work moves with you. When the model changes, the pattern adapts — it does not have to be relearned.
Compounds across your career
Universal AI skills compound. Every solution you build with Lucy makes the next one faster and the one after that sharper. The fluency stays even when tools and roles change.
Survives the next model generation
The skills Lucy coaches — prompting, evaluation, decomposition, tool orchestration, HITL — outlast any specific model. The patterns get cheaper and more capable; your judgment about when and how to use them stays valuable.
Boundaries
What Lucy is not.
Just as important as what Lucy does is what she does not do. The boundary is the trust.
Lucy does not act on your behalf
Lucy herself does not write, send, configure, schedule, or trigger anything on your behalf. She coaches you to build solutions; the agents you build are the things that do work, with HITL gates you set.
Not a training program you forget
Lucy is not a training course you click through, pass, and forget. She is not a chatbot that answers AI trivia. She is not a generic productivity tool grafted onto your week.
Not more weight on a full plate
Lucy is not more work on your already full plate. If a Lucy interaction is not creating value for you, that is a Lucy problem, not a you problem — and "not now" carries the same weight as "yes."
Not surveillance, not scoring
Lucy is not employee surveillance. No private prompt feeds. No individual productivity ranking. Aggregate patterns and capability growth flow to leaders; individual recognition is opt-in.
Try Lucy.
Start with a real task on your plate. Lucy will ask a few questions and propose the smallest useful shape. If it lands, you will leave the conversation with a pattern worth keeping — and the capability stays with you.