Lo que pidió Juan.
Este 360 nace de un encargo concreto de Juan (Hive · brazo técnico). Su audio es el brief — no un workflow del equipo. Acá está, punto por punto, lo que pidió y cómo lo entregamos.
Las 6 cosas que pidió
Compilar + agregar
"Pegar todo lo que mandaron sobre cada rutina de cada uno y agregarlo." Juntar los 10 workflows sueltos en un solo lugar coherente.
Iterar en el cockpit
"Iterar no cockpit da Atómica." No empezar de cero: este reporte vive en el mismo sistema y diseño del cockpit que ya existe.
Mapear lo repetitivo
"Mapear os processos repetitivos que cada um tem." Detectar, persona por persona, qué se repite todos los días.
Quebrar y organizar
"Quebra isso em partes, divide, organiza, bota achados." Romper el flujo en piezas, ordenarlo y dejar hallazgos claros.
Aplicar inteligencia
"Aplica uma inteligência com o Claude." Pasar todo por Claude para cruzar los workflows y proponer dónde automatizar.
El 360 de toda la empresa
"Fazer um 360 de todas as pessoas da empresa, agora que tem contexto de todos os workflows." Una vista completa, área por área.
El entregable, tal cual lo pidió
"Um relatório completo, separado por área, por workflow, por rotina — do que está repetitivo, o que toma tempo, o que dá pra automatizar, tudo." Así está armado el 360:
Por área
Cuentas · Contenido · Diseño & DA · Audiovisual. Una pestaña por área.
Por workflow
Cada persona, su flujo real de punta a punta, paso a paso.
Por rutina
Cada rutina con su frecuencia, herramientas y si se puede automatizar.
De los workflows al plan de 3 meses
El equipo · quién mandó su workflow
| Persona | Rol · Área | Workflow |
|---|---|---|
| Stefany Delgado | AE senior · Cuentas | ✓ enviado |
| Lourdes González | AE junior · Cuentas | ✓ enviado |
| Nico Sosa | Director Creativo · Contenido | ✓ enviado |
| Santiago Cruz | Director de Arte · Diseño & DA | ✓ enviado |
| Federica Méndez | Diseñadora Junior · Diseño & DA | ✓ enviado |
| Juana Ramirez | Com. Visual & AV · Audiovisual | ✓ enviado |
| Giuliana Pastorino | Coord. Paid Media · Paid Media | ✓ enviado |
| Lucia Correa | Paid Media Specialist · Paid Media | ✓ enviado |
| Florencia Rey | Community Manager · Contenido | pendiente |
| Agustín Caraballo | Account Exec · Cuentas | pendiente |
| Paula Pagani | Account Exec · Cuentas | pendiente |
| Estefanía | Diseñadora · Diseño & DA | pendiente |
| Maricel | Account Exec · Admin · Cuentas | pendiente |
| Verónica Rey | CEO · Founder | dirección |
| Analicia Sánchez | Partner · Co-decisora | dirección |
| Marielena Echenagucia | Partner · Tech-savvy | dirección |
10 de 16 recibidos · 6 pendientes (3 operativos + 3 de dirección) · Juan (Hive) no entra — es el brief, no un workflow.
Get it in one minute.
This is what the team does by hand today, how much time it eats, and what Steve does to take it off their plate. Read it top to bottom: ① the priority · ② the time · ③ how it works · ④ the chain.
The performance report, done on its own.
Across the 29 active accounts, logging into Meta, Google, TikTok, LinkedIn and Analytics, copying numbers into sheets and building each status or report is the single most repeated manual work. Account execution and campaign management alone eat about 490 hours a month, and a big share of that is this grind. Verónica herself calls it obsolete. Steve builds it on its own and leaves it ready to review. One automation gives hours back across the whole book.
Every bar is one full working day. The colored part is the repetitive, mechanical work Steve can take over (pulling data from Meta, Google and Analytics, filling sheets, building status and reports, gathering files, chasing approvals). The light part stays with the team: strategy, judgment, creative, the client relationship.
Paid MediaGiuliana · LuciaStays with the teamSteve handles this60%of the day
Admin · FinanceMaricelStays with the teamSteve handles this60%of the day
AccountsStefany · LourdesStays with the teamSteve handles this55%of the day
Design & ADSantiago · FedericaStays with the teamSteve handles this55%of the day
AV ProductionJuaniStays with the teamSteve handles this50%of the day
ContentNicoStays with the teamSteve handles this45%of the day
StrategyVerónicaStays with the teamSteve helps20%of the dayLonger colored part means more of the day is mechanical and absorbable by Steve. Strategy sits low on purpose: that work stays human.
One single assistant that lives in Slack and runs on our server 24/7. The team talks to it in plain language; it does the heavy lifting and returns a draft to review.
1. The team asks
In Slack, in plain language: "Steve, build today's UCM status".
2. The AI works
On our server (24/7). Reads that client's context (one file per account, 29 already loaded) and connects to the tools.
3. Gathers the data
Meta · Google · Analytics · Drive · Asana · Sheets. Pulls the numbers and builds the piece.
4. Returns the draft
In Slack, ready to review. The person edits and approves. Judgment stays human.
Is it many robots?
No. It is one single AI assistant (we call it Steve) with several skills (pull reports, build the daily, invoice). The team talks to just one.
Where does it live?
On our server, on 24/7. It is not installed on your computer, does not slow it down, does not shut off when you close the laptop.
How do you talk to it?
In Slack, like a teammate. You ask in plain language, it returns the draft, you review. Nothing goes out without a person approving.
Lo que hace cada área.
Abrí la página de cada área para ver su workflow a fondo — cada persona, su flujo del día paso a paso y sus rutinas. Cada área es su propia página.
Lo repetitivo y lo que duele.
El cruce de los 10 workflows deja dos listas claras: lo que se repite con patrón fijo (sale de las manos del equipo) y los sumideros que rompen la planificación (impredecibles o de embudo).
Matriz de lo repetitivo
| Rutina | Quién | Frecuencia | Automatizable |
|---|---|---|---|
| Extracción de métricas → planillas/status | Stefany ×4 · Lourdes | diaria | Sí · alto |
| Planillas de Seguimiento de Inversión | Lourdes ×5 | diaria | Sí · alto |
| Daily de Diseño | Santiago | diaria | Sí · alto |
| Revisión multicanal de arranque (mail+Slack+Asana+calendario) | Todos | diaria | Parcial · resumen |
| Chequeo de posts publicados 9hs | Lourdes | diaria | Sí |
| Pauta de contenidos (Plan de Medios → Ads Mgr) | Lourdes + Stefany | diaria | Parcial |
| Generación de imágenes IA | Fede + Juani | por tarea | Sí · con skill |
| Documentación de cliente | Lourdes | de fondo | Sí · es el Hub |
| Onboarding técnico · Cotización de extras | Stefany | por cliente / demanda | Sí |
Los sumideros
Post-edición IA con packaging
(Fede) Etiquetas, logos y sellos se deforman → retoque manual de duración imposible de estimar. Desbalancea la tarea.
Esperas de aprobación del cliente
(Todo el pipeline) El ritmo lo marca el cliente. Se acumulan piezas y se desordena la agenda interna.
Aprobación única de DA
(Santiago) Nada sale sin su OK. Calidad garantizada, pero único punto de paso que cuellea a fin de mes.
Ida y vuelta paid-media
(Stefany) Los AE no son expertos en pauta → consultas constantes + research extra antes de cada reunión.
Briefings ambiguos
(Diseño/AV) Algunos cronos llegan claros, otros piden consultas → pasos extra invisibles en el flujo.
Fragmentación de herramientas
(Todos) La info de un cliente vive en Asana + Canva + Drive + Slack + Sheets + Meta/Google. CAT, sin planilla, suma desorden.
10 skills para Steve-Atómica.
Cada hallazgo se vuelve una skill concreta del agente, con prioridad. P0 ya iniciado o mayor ahorro · P1 alto valor, sigue · P2 cuando haya base.
| # | Skill de Steve | Qué resuelve | Skill / ruta | Prioridad |
|---|---|---|---|---|
| 1 | Reporte de performance | Meta+Google+Analytics → status y planillas solos, con insight en Slack. Mata el sumidero de cuentas. | /reporte-cliente | P0 |
| 2 | Daily de Diseño automática | Lee Asana + calendario + deadlines y arma la Daily por persona. Santiago la suelta en 2 min. | /daily-diseno | P0 |
| 3 | Documentación viva de cliente | El context.md por cliente, auto-armado desde Slack/Drive/manuales. Es el corazón del Hub. | /digest | P1 |
| 4 | Inteligencia post-reunión | Notas de Meet → contexto del cliente + action items + tareas en Asana. Pedido de Stefany. | /post-meeting | P1 |
| 5 | Carruseles ágiles de Atómica | Fondos fijos/gradientes + tipografía de marca, sin el ciclo full de diseño. Pedido de Stefany. | /gerar-carrossel-cliente | P1 |
| 6 | Triangulación paid-media | Conecta necesidad del cliente con performance y explica el "por qué". Corta el ida y vuelta del AE. | /insight-campana | P1 |
| 7 | Generación de imágenes IA | Encapsula Flow/Magnific/ChatGPT con prompts por marca. El packaging se ataca aparte. | /generar-imagen | P2 |
| 8 | Onboarding técnico | Checklist de accesos, mail automático, verificación. Hoy Stefany lo hace cliente por cliente. | /nuevo-cliente | P2 |
| 9 | Cotización de extras | Horas → precio/impuestos/margen, desde el Excel que ya usa Stefany. | /cotizar | P2 |
| 10 | Verificación de publicación | Chequea que los posts de las 9hs salieron y avisa solo si algo falla. | /check-publicacion | P2 |
Skills que ya viven en el Hub
No se arranca de cero: el repo ya tiene skills y comandos que las automatizaciones de arriba reutilizan. Steve los orquesta.
The whole book, one brain.
These 29 accounts are the agency. Each one now has its own living context.md inside the Hub: industry, voice, what is allowed and not, approval flow, hours. That is the multi-tenant memory that feeds Steve, so any request ("a carousel for Sika", "the UCM weekly report") already knows the client. Hours are the real average of the last 4 months.
| Account | Industry | Executive | Hrs / month | Profile |
|---|---|---|---|---|
| Grupo Gramón | Pharma | Stefany / Paula | 106 | partial |
| Uncle Jim | Textile | Stefany | 98 | empty |
| UCM | Emergency medical services | Stefany | 96 | partial |
| Sika | Construction / chemicals | Maricel | 84 | partial |
| RPA + RPD | Fishing | n/d | 69 | partial |
| CUATROASES & LA DOLFINA | Textile | Ana | 54 | empty |
| Werba | Recycling | Stefany | 52 | partial |
| Paloma | Textile | Stefany | 40 | partial |
| Medi Import | n/d | Maricel | 39 | incomplete |
| Axell | Food | Lourdes | 32 | partial |
| Corteva | Agro | n/d | 29 | empty |
| Balsoma | Medical | Stefany | 23 | empty |
| Cebala | Textile / Food | Ana | 20 | empty |
| Imperio de las importaciones | n/d | n/d | 20 | empty |
| Manto terminaciones | n/d | Maricel | 19 | incomplete |
| Urgenterap | Textile | Lourdes | 17 | partial |
| Tilsor | Textile | Marie | 15 | empty |
| Mitra | Textile | n/d | 6 | empty |
| Pepeganga | Textile | n/d | 6 | empty |
| Marcas Famosas | n/d | n/d | 4 | empty |
| Roemmers | Textile | n/d | 4 | empty |
| UYTECH | Textile | n/d | 4 | empty |
| Diora | Textile | n/d | 3 | empty |
| Atómica | n/d | Stefany | n/d | empty |
| Balsoma Dental | Medical | Stefany | n/d | empty |
| CENTRO DE ALTA TECNOLOGÍA (CAT) | Medical | Ana / Lourdes | n/d | empty |
| Kroser Bentancur | n/d | n/d | n/d | empty |
| Mentis | n/d | n/d | n/d | empty |
| Quatum | Transport | Lourdes | n/d | empty |
We leave today with this running.
One decision, one start, clear access. Every workflow needs to connect its sources (the accounts spreadsheet, Meta/Google, Drive, Asana). No access, no automation, so each row says exactly what to connect. Where the live feed is not connected yet, it runs on the data we already have and we mark the rest as pending access.
The performance report, on UCM.
It is the number 0 the team chose and the biggest time sink (account exec + campaign = 45% of the time). UCM already has its ad accounts connected. Steve builds the weekly status on its own, Analicia reviews.
Worked example: the UCM weekly report, end to end
Exactly how it works, the people side and the technical side, today with the data we already have. Plug the live feed later and the same flow runs on its own.
What lands in the channel, sample with real UCM numbers from this week:
The 90 minutes of pulling numbers becomes a 2 minute review.
The 3 steps this week
Connect UCM's sources
Meta Ads + Google Ads + Analytics of UCM into the system (Pipeboard is ready). This is the access the workflow needs.
Steve builds the first report
UCM weekly status (spend, CPL, leads, top ads, 3 actions). Analicia reviews and sets the tone.
Book the 1:1s per area
One call per area (Paid, Accounts, Content, Design) to connect access and build its agent.
The workflows and what each one needs connected
| Workflow | What it does | What to connect (access) | When |
|---|---|---|---|
| Performance report P0 | Meta + Google + Analytics → weekly status + insight, on its own | UCM's Meta Ads, Google Ads and Analytics + Slack channel | This week |
| Living client doc | The context.md per account, auto-built and always current | Accounts spreadsheet + Drive + client Slack | 30 days |
| Design Daily | Reads Asana + deadlines and builds the Daily per person | Asana (client project) + Google Calendar | 30 days |
| Agile carousels | Fixed backgrounds + brand type, without the full design cycle | Brand assets per account (Drive / brand book) | 30-60 days |
| Paid triangulation | Connects client need with performance and explains the why | Performance data (Meta/Google) + spreadsheet | 30-60 days |
| Orchestration + dashboard | Multi-account view for Verónica: what each agent is doing | All of the above connected + agents per area | 60-90 days |
De workflows a sistema.
El 360 es el insumo directo del ciclo Q2. No se arranca de cero: estos workflows son el contexto que entrena a Steve. El plan ataca primero lo que más duele y más se mide.
Cómo arranca
Contexto + Steve base
El context.md de cada cliente vivo + Steve escuchando Slack. Primera skill: el reporte de performance (mayor ahorro, ya iniciado con los MCPs en la fase base).
Una skill por área
Cuentas, contenido, diseño y AV — cada área con su skill, todas leyendo el mismo contexto. La Daily automática y la documentación viva entran acá.
Orquestación + dashboard
Steve propaga prioridades entre clientes, dashboard ejecutivo para Verónica, y el camino white-label. Infra objetivo < USD 500/mes, Slack como interfaz.
El equipo opera más solo
L1 → L2 → L3. La meta no es dependencia: es que Atómica vuele con el sistema en la mano. El 360 se re-prioriza con cada workflow nuevo.
El 360 está vivo.
Mandá los workflows que faltan (Flor, el Loom de Stefany, el resto del equipo) y se suman en la próxima vuelta. Cada uno hace a Steve más preciso.
Client data, handled with care.
Using client data with AI needs a clear policy, not a black box. This is how Atómica's system treats it, aligned with the principles of Uruguay's data protection law (Ley 18.331). The short version: the data stays in Atómica's house, only what a task needs reaches the model, and a person always approves before anything goes out.
Where the data lives
| What | Where | Access |
|---|---|---|
| Client context (brand books, assets, briefs) | Atómica's private GitHub repo (the brain) + Atómica's own server | Team only, version controlled |
| API keys / secrets | Environment variables on the server | Never in the repo, never in chat |
| Day to day work | Slack + the server (24/7) | No client data required on personal laptops |
What reaches the AI, and what does not
Only the task context
For each request, the agent sends just what that task needs (the relevant brand book, the brief). Not the whole book, not unrelated clients.
No training on your data
The model runs on Anthropic's commercial API, which does not use business inputs or outputs to train its models. Calls are per request.
No end-consumer personal data
We work with brand and marketing assets, not sensitive personal data of clients' customers. Anything sensitive is opt in, per account.
Transparency with clients
Two simple moves keep trust and compliance clean:
A clause in the agreement
A short line stating that AI assists in content production under human supervision. Clear, honest, standard.
Per client opt in for sensitive data
If an account involves anything beyond marketing assets, we ask first and log the consent. Default is minimal data.
Rights respected
Data minimization, purpose limitation, and the client's right to access, correct or delete their data. Always.
The legal frame, three laws, one standard
Atómica operates from Uruguay, serves clients in the region, and the system is built by a Brazilian partner on US infrastructure. So the policy is written to satisfy the three frameworks at once, by design, not patch by patch.
| Law | Scope | What it demands of us | How we comply |
|---|---|---|---|
| Ley 18.331 (Uruguay) | Atómica and its clients' data | Prior consent, purpose limitation, security measures, registered databases, rights of access/rectification/deletion | Minimal data per task · consent clause · data stays on Atómica's server |
| LGPD (Brazil, Lei 13.709) | Any data touching Brazilian operations or data subjects | Legal basis for processing, data minimization, controller/processor roles, incident reporting, DPO when applicable | Same minimization policy · Atómica = controller, Hive = operator under contract |
| GDPR (EU, reference standard) | Gold standard, applies if any EU client/data subject enters | Lawful basis, data processing agreements, international transfer safeguards, privacy by design | Policy already written to GDPR level · Anthropic offers DPA with standard contractual clauses |
Who touches the data
Atómica (controller) → Hive (operator, under the service agreement) → Anthropic API (sub-processor, no training on inputs, DPA available). No other party. No data brokers, no resale, ever.
If something goes wrong
Any suspected leak: the affected client is informed without undue delay, the scope is documented, and access is rotated. Plain language, no burying.
The data is yours
Any client (or Atómica itself) can ask for export or deletion of their folder at any time. It is files in a repo: deleting it is real deletion, verifiable.
Everything you sent, digested.
Nothing the team shares sits in a drive folder. Every document gets read, structured and folded into the system: the client book became 29 living context files, the Asana manual became the agent's methodology, the meetings became the plan. This tab is the audit trail: what came in, from whom, and what it turned into.
| Document | From | When | What it became in the system |
|---|---|---|---|
| Client book (Accounts Dept doc, 29 profiles) | Marielena | Jun 3 | 29 living context.md files (one per account) + the Accounts tab + Steve's account catalog |
| Hours per client (spreadsheet) | Marielena | Jun 3 | Real numbers in the whole report: 1,093 h/mo, the load map, the 45% sink finding |
| Client distribution / team allocation (spreadsheet) | Marielena | Jun 3 | Executive per account in every profile (Stefany, Lourdes, Maricel, Ana, Marie) |
| Asana survival guide (PDF, 5 sections) | Team (via Marielena) | Jun 3 | Distilled to methodology/: project structure, roles, flows, tags. Steve knows your process |
| Team structure email + team presentation (PDF) | Marielena | Jun 3 | Org chart by area (C-level, AEs, Paid, Content, Design) + team canvas |
| 10 workflow write-ups (each person's routine) | Whole team | May | The Workflows tab, the findings, and the 10-skill automation backlog |
| Session 4 (full transcript, 74 min) | All hands | May 27 | The agent architecture (Agent Zero, areas, playbooks) + Q2 plan framing |
| Working meeting (Verónica · Marielena · Analicia) | C-level | Jun 3 | Priority zero confirmed (performance report) + the Security & data tab (delivered commitment) |
| UCM weekly summaries (Slack) | Stefany / Paid | Weekly | Live metrics in the report: Meta 128 conv @ $4.67 · Google 49 @ $5.15 · budget USD 9,500 |
| Audio feedback (team canvas layout) | Verónica | May 28 | Org map rebuilt horizontal, C-level at the center, color by area |
How digestion works (the pipeline)
Drop anything
PDFs, spreadsheets, emails, audios, transcripts go into the Hub inbox, no order needed. The agent reads and classifies.
It becomes context
Each piece lands where it belongs: client profiles, methodology, team docs, metrics. Markdown, versioned, auditable.
Steve learns it
The agent's brain syncs from the repo every 5 minutes. Whatever you send today, Steve knows it today. Ask him in Slack.
Todo el material, en un lugar.
Lo que siempre acompaña una sesión con Atómica: preparación, slides y las sesiones pasadas. Esta es la pestaña fija — se actualiza cada ciclo.
Preparación · Sesión 5
Para que la Sesión 5 (wrap-up + arranque del plan) sea rápida, traer esto:
Validar el 360 por área
Cada uno revisa su workflow en su pestaña: ¿falta un paso? ¿sobra algo?
Elegir la primera skill
Recomendado: reporte de performance (mayor ahorro). Voto de Verónica + Marielena.
Los tiempos de Asana
Lourdes (y Stefany) traen los minutos cronometrados para dimensionar el ahorro real.
Slides de cada sesión
Slides · Sesión 5newwrap-up + plan 3 meses→
Reporte 360 · deck S5esta página · wrap-up→
Slides · Sesión 4Meta + Google Ads MCP→
Slides · Sesión 3motor de contenido · copy→
Slides · Sesión 2contexto & marca→
Slides · Sesión 1fundamentos + setup→
Sesiones pasadas · cockpits
Cockpit vivo de la alianzaen vivo · ES/EN/PT→
Cockpit · Sesión 4MCPs Meta + Google→
Cockpit · Sesión 3motor de contenido→
Cockpit · Sesión 2contexto & marca→
Team Canvas · 16 rolesroles confirmados (Fathom S4)→
Preparaciones anteriores
Preparación · Sesión 4accesos + Slack + campaña→
Preparación · Sesión 3voice samples + KPIs→
Preparación · Sesión 2briefs UCM + Werba→
Preparación · Sesión 1kickoff + setup inicial→
Otros recursos
Repo · hub-atomicaGitHub privado de la alianza→
Propuesta · Q2scope · pricing · alianza→
Discovery callFathom · 58 min · 17 mar→
Catch-up setupsi te trabaste · Mac/Windows→