AI user & operator
Improve personal and team workflows before writing code.
Start with prompting, verification, task decomposition, data hygiene, and repeatable playbooks.00 · Field mapMapa del campo
A broad map of AI capabilities, roles, markets, and routes.
Un mapa amplio de capacidades, roles, mercados y rutas de IA.
A field map for learning, building, applying, operating, governing, and commercializing AI
AI is not one job, one model, or one product category. It is a stack of sciences, engineering disciplines, interfaces, operating controls, creative practices, business functions, and emerging markets. Use this landscape to explore broadly, shortlist deliberately, and build evidence.
Un mapa para aprender, construir, aplicar, operar, gobernar y comercializar IA
La IA no es un solo empleo, modelo o categoría de producto. Es una combinación de ciencias, disciplinas de ingeniería, interfaces, controles operativos, prácticas creativas, funciones empresariales y mercados emergentes. Usa este panorama para explorar con amplitud, seleccionar con criterio y construir evidencia.
01 · Choose an entry positionElige una posición de entrada
Start from your strengths, then sample adjacent branches.
Empieza desde tus fortalezas y prueba ramas adyacentes.
The field supports nontechnical operators, software builders, researchers, product designers, security leaders, domain experts, educators, creators, and founders. Pick one primary entry position and one adjacent position to learn across boundaries.
Improve personal and team workflows before writing code.
Start with prompting, verification, task decomposition, data hygiene, and repeatable playbooks.Compose models, data, tools, APIs, and interfaces.
Start with structured output, tool calls, RAG, evals, auth, and one production deployment.Work on prediction, adaptation, experimentation, and model quality.
Start with statistics, ML baselines, datasets, embeddings, fine-tuning, and reproducible experiments.Shape useful, legible, trustworthy AI interactions.
Start with user research, uncertainty, progressive disclosure, multimodal UX, and feedback loops.Make AI systems observable, controlled, compliant, and recoverable.
Start with risk tiers, threat models, identity, audit, eval gates, incident response, and cost controls.Turn domain knowledge and distribution into products or services.
Start with workflow interviews, outcome design, a bounded demo, proof, pricing, and support.Scan every opportunity cluster. Mark what is interesting, valuable, and realistically accessible.
Run three small experiments in different branches. Keep each to one day or one weekend.
Select one 12-week practice lane. Preserve a secondary lane for cross-pollination.
El campo incluye operadores no técnicos, desarrolladores, investigadores, diseñadores, líderes de seguridad, expertos de dominio, educadores, creadores y fundadores. Elige una posición principal y una adyacente para aprender entre fronteras.
Mejora flujos personales y de equipo antes de programar.
Empieza con prompts, verificación, descomposición, higiene de datos y playbooks repetibles.Combina modelos, datos, herramientas, APIs e interfaces.
Empieza con salida estructurada, herramientas, RAG, evals, autenticación y un despliegue.Trabaja en predicción, adaptación, experimentación y calidad.
Empieza con estadística, bases ML, datasets, embeddings, fine-tuning y experimentos reproducibles.Diseña interacciones útiles, legibles y confiables.
Empieza con investigación, incertidumbre, divulgación progresiva, UX multimodal y feedback.Haz que los sistemas sean observables, controlados, conformes y recuperables.
Empieza con niveles de riesgo, amenazas, identidad, auditoría, gates, incidentes y costos.Convierte conocimiento de dominio y distribución en productos o servicios.
Empieza con entrevistas, resultados, demo acotada, evidencia, precio y soporte.Revisa cada grupo. Marca lo interesante, valioso y realmente accesible.
Ejecuta tres experimentos pequeños en ramas distintas. Limita cada uno a un día o fin de semana.
Selecciona una ruta de práctica de 12 semanas. Conserva otra como ruta adyacente.
02 · Navigable mind mapMapa mental navegable
Pan, zoom, filter, focus, and inspect the landscape.
Desplaza, amplía, filtra, enfoca e inspecciona el panorama.
The branches are not isolated careers. They are connected layers and participation modes. Explore across the field, then build depth where interest, access, evidence, and value converge.
Las ramas no son carreras aisladas. Son capas y formas de participación conectadas. Explora el campo y profundiza donde converjan interés, acceso, evidencia y valor.
03 · The system landscapeEl panorama del sistema
Eight layers from model capability to commercial delivery.
Ocho capas desde capacidad del modelo hasta entrega comercial.
Most failures happen at the seams: context that is stale, a tool with excessive privilege, an unmeasured workflow, or an interface that hides uncertainty.
Chat, copilot, voice, embedded widgets, command palettes, APIs, background automation, and physical/edge interfaces.
The surface determines trust, interruption cost, and adoption.Workflows, agents, routers, planners, state machines, queues, retries, approvals, and human handoff.
Explicit control is usually the production default; autonomy is allocated selectively.Functions, APIs, MCP servers, A2A peers, browser/computer tools, event buses, payment rails, and identity.
Protocols reduce custom glue but do not remove authorization or trust design.RAG, hybrid search, embeddings, reranking, memory, knowledge graphs, SQL, files, and enterprise systems.
Context engineering turns generic intelligence into situational capability.Reasoning, language, vision, audio, video, embeddings, rerankers, open weights, local models, routing, and serving.
Choose by measured task fit, not leaderboard prestige.Golden sets, simulations, traces, quality metrics, safety tests, cost, latency, drift, and user feedback.
Without evals, iteration is anecdotal and regressions are invisible.Identity, least privilege, data controls, threat modeling, policy, audit, provenance, red teaming, and incident response.
Agency expands the blast radius; control planes must expand with it.Problem selection, offer, pricing, distribution, contracts, support, outcomes, and defensibility.
A useful model call is not yet a business.La mayoría de fallas ocurre en las uniones: contexto desactualizado, herramientas con privilegios excesivos, flujos sin medir o interfaces que ocultan incertidumbre.
Chat, copiloto, voz, widgets embebidos, paletas de comandos, APIs, automatización en segundo plano e interfaces físicas/edge.
La superficie determina confianza, costo de interrupción y adopción.Flujos, agentes, routers, planificadores, máquinas de estado, colas, reintentos, aprobaciones y transferencia humana.
El control explícito suele ser la base productiva; la autonomía se asigna selectivamente.Funciones, APIs, servidores MCP, pares A2A, herramientas de navegador/computador, buses de eventos, pagos e identidad.
Los protocolos reducen pegamento personalizado, pero no eliminan autorización ni confianza.RAG, búsqueda híbrida, embeddings, reranking, memoria, grafos, SQL, archivos y sistemas empresariales.
La ingeniería de contexto convierte inteligencia genérica en capacidad situacional.Razonamiento, lenguaje, visión, audio, video, embeddings, rerankers, pesos abiertos, modelos locales, routing y serving.
Elige según ajuste medido a la tarea, no por prestigio del ranking.Conjuntos dorados, simulaciones, trazas, métricas de calidad, pruebas de seguridad, costo, latencia, deriva y feedback.
Sin evals, la iteración es anecdótica y las regresiones son invisibles.Identidad, mínimo privilegio, controles de datos, amenazas, políticas, auditoría, procedencia, red team e incidentes.
La agencia amplía el radio de impacto; el plano de control debe crecer también.Selección del problema, oferta, precio, distribución, contratos, soporte, resultados y defensibilidad.
Una llamada útil al modelo todavía no es un negocio.04 · Timeless foundationsFundamentos duraderos
Concepts that remain useful across model and framework cycles.
Conceptos que siguen siendo útiles entre ciclos de modelos y frameworks.
Framework APIs change quickly. The durable mental model is how probabilistic generation interacts with deterministic software, external evidence, state, and risk.
Discrete units processed by a model. They affect context limits, cost, and latency.
Vectors that represent semantic relationships. Useful for search, clustering, recommendation, and matching.
Attention-based architecture that relates elements in a sequence; the foundation of many modern models.
Running a trained model to produce output, optimized across quality, latency, cost, throughput, and hardware.
A response validated against JSON Schema or another contract. It reduces ambiguity between model and software.
The model selects a function and arguments; the application validates, executes, and returns the result.
Retrieves external evidence and adds it to context before generation. It supports freshness, grounding, and provenance.
Reorders retrieved candidates with a more precise model before context assembly.
Adapts weights to patterns, format, or style. It does not replace a dynamic source of truth.
Repeatable tests of quality, safety, cost, and behavior—the analogue of tests for probabilistic systems.
A runtime that uses a model to select steps and tools under policies, state, and stop conditions.
An explicit controlled sequence. Often more reliable than open-ended agency when the process is known.
Las APIs cambian rápido. El modelo mental duradero es cómo la generación probabilística interactúa con software determinista, evidencia externa, estado y riesgo.
Unidades discretas procesadas por el modelo. Afectan contexto, costo y latencia.
Vectores que representan relaciones semánticas. Sirven para búsqueda, agrupación, recomendación y matching.
Arquitectura basada en atención que relaciona elementos de una secuencia; base de muchos modelos modernos.
Ejecución de un modelo entrenado para producir salida, optimizada por calidad, latencia, costo, volumen y hardware.
Respuesta validada contra JSON Schema u otro contrato. Reduce ambigüedad entre modelo y software.
El modelo elige función y argumentos; la aplicación valida, ejecuta y devuelve el resultado.
Recupera evidencia externa y la agrega al contexto antes de generar. Favorece frescura, grounding y procedencia.
Reordena candidatos recuperados con un modelo más preciso antes de ensamblar el contexto.
Adapta pesos a patrones, formatos o estilo. No reemplaza una fuente dinámica de verdad.
Pruebas repetibles de calidad, seguridad, costo y conducta: el equivalente a tests para sistemas probabilísticos.
Runtime que usa un modelo para elegir pasos y herramientas bajo políticas, estado y condiciones de parada.
Secuencia explícita y controlada. Suele ser más confiable que agencia abierta cuando el proceso es conocido.
05 · Models, tools, and inferenceModelos, herramientas e inferencia
Build a measured portfolio instead of betting on one vendor.
Construye un portafolio medido en lugar de apostar por un proveedor.
Use routing: cheap and fast for classification; stronger reasoning for ambiguity; specialist models for speech, vision, embeddings, reranking, or media; local models where privacy, cost, or offline operation justify them.
Usa routing: barato y rápido para clasificación; razonamiento más fuerte para ambigüedad; especialistas para voz, visión, embeddings, reranking o medios; modelos locales cuando privacidad, costo u operación offline lo justifiquen.
06 · Agents, protocols, and transactionsAgentes, protocolos y transacciones
How tools, MCP, A2A, and payment protocols fit together.
Cómo encajan herramientas, MCP, A2A y protocolos de pago.
An agent becomes useful when it has bounded capabilities, trustworthy context, explicit state, observable actions, budgets, stop conditions, and recovery.
Start with deterministic steps. Let the model handle interpretation, ranking, transformation, and exceptions. Expand autonomy only where evaluation shows a benefit.
A host connects to servers that expose tools, resources, and prompts. Treat every server as a privileged integration boundary.
Use for reusable connectors, enterprise data access, IDE tools, and app extensions.Agents advertise capabilities and exchange tasks, messages, status, and artifacts.
Use when independently operated agents need delegation or long-running collaboration.A server can require a payment in an HTTP-native flow before serving a resource or capability.
Use for metered APIs, data, inference, or agent services where tiny transactions make economic sense.A trust and mandate layer for agent-mediated purchases and payments.
Use for commerce scenarios requiring user intent, merchant trust, and verifiable authorization.Un agente es útil cuando tiene capacidades limitadas, contexto confiable, estado explícito, acciones observables, presupuestos, condiciones de parada y recuperación.
Empieza con pasos deterministas. Deja al modelo interpretación, ranking, transformación y excepciones. Amplía autonomía solo donde las evaluaciones demuestren beneficio.
Un host se conecta a servidores que exponen herramientas, recursos y prompts. Trata cada servidor como límite privilegiado.
Úsalo para conectores reutilizables, datos empresariales, IDEs y extensiones de apps.Los agentes publican capacidades e intercambian tareas, mensajes, estados y artefactos.
Úsalo cuando agentes operados de forma independiente necesiten delegación o colaboración larga.Un servidor puede exigir pago mediante un flujo HTTP antes de entregar un recurso o capacidad.
Úsalo para APIs, datos, inferencia o servicios medidos cuando las microtransacciones tengan sentido.Capa de confianza y mandato para compras y pagos mediados por agentes.
Úsalo cuando se requieran intención del usuario, confianza del comercio y autorización verificable.07 · Modalities and surfacesModalidades y superficies
Where multimodal capability becomes useful interaction.
Dónde la capacidad multimodal se vuelve interacción útil.
The gap is rarely “support another media type.” The opportunity is a complete loop—for example, hear a Spanish call, retrieve policy, coach in English, update CRM, and preserve an auditable summary.
Chat, extraction, classification, transformation, planning, SQL and JSON contracts.
Best first modality: cheap to test, easy to evaluate, broad demand.Speech recognition, synthesis, interruption, turn-taking, translation, call summaries and live coaching.
Strong bilingual opportunity in sales, hospitality, healthcare administration, and field operations.OCR, layout, tables, forms, images, screenshots, quality inspection and visual grounding.
High-value when documents or visual evidence currently require manual review.Creative generation, localization, storyboards, editing, avatars, product media and synthetic data.
Commercially active but quality control, rights, brand consistency, and disclosure matter.Repository edits, tests, browser operation, desktop workflows, command execution and UI automation.
Treat as privileged automation. Prefer APIs; constrain computer use to isolated environments.Cameras, microphones, telemetry, devices and local models.
Useful for privacy, latency, intermittent connectivity, and physical workflows.Chat, voice, messaging, email
Good for ambiguous intent and progressive clarification.Inside CRM, IDE, support desk, browser, document, or operations console
Best when context and action are already present.Model selects or composes cards, forms, tables, charts, and controls
Useful for dynamic tasks; must preserve accessibility and action clarity.Event-triggered or scheduled process
High leverage, but requires observability, approvals, and exception queues.A reusable machine-consumable service
Best path toward repeatable products and agent-to-agent markets.Dialog, sheet, drawer, popover
A presentation pattern—not an AI modality. Use for bounded decisions, confirmation, or focused detail.La brecha rara vez es “soportar otro medio”. La oportunidad es cerrar el ciclo: escuchar una llamada en español, recuperar políticas, orientar en inglés, actualizar CRM y guardar un resumen auditable.
Chat, extracción, clasificación, transformación, planeación, SQL y contratos JSON.
La mejor primera modalidad: barata de probar, fácil de evaluar y con demanda amplia.Reconocimiento, síntesis, interrupción, turnos, traducción, resúmenes y coaching en vivo.
Gran oportunidad bilingüe en ventas, hotelería, administración de salud y operaciones de campo.OCR, layout, tablas, formularios, imágenes, capturas, inspección y grounding visual.
Alto valor cuando documentos o evidencia visual requieren revisión manual.Generación creativa, localización, storyboards, edición, avatares, medios de producto y datos sintéticos.
Mercado activo, pero importan calidad, derechos, marca y divulgación.Cambios en repositorios, pruebas, navegador, escritorio, comandos y automatización UI.
Trátalo como automatización privilegiada. Prefiere APIs y aísla el uso del computador.Cámaras, micrófonos, telemetría, dispositivos y modelos locales.
Útil para privacidad, latencia, conectividad intermitente y flujos físicos.Chat, voz, mensajería y correo
Buena para intención ambigua y aclaración progresiva.Dentro de CRM, IDE, soporte, navegador, documento o consola
Mejor cuando contexto y acción ya están presentes.El modelo elige o compone tarjetas, formularios, tablas, gráficos y controles
Útil en tareas dinámicas; debe conservar accesibilidad y claridad.Proceso por evento o programación
Gran apalancamiento, pero requiere observabilidad, aprobaciones y colas de excepción.Servicio reutilizable consumido por máquinas
Camino fuerte hacia productos repetibles y mercados agente-a-agente.Diálogo, hoja, drawer o popover
Patrón de presentación, no modalidad de IA. Úsalo para decisiones acotadas, confirmación o detalle.08 · Opportunity domainsDominios de oportunidad
Thirty-two ways to learn, build, apply, operate, govern, or commercialize AI.
Treinta y dos formas de aprender, construir, aplicar, operar, gobernar o comercializar IA.
These domains overlap. A voice product needs data, UX, evaluation, security, and operations. A governance practice needs technical literacy. A robotics team needs simulation, models, sensors, safety, and product design. Filter the map, then sample several branches.
Teach people to frame tasks, verify outputs, protect data, and integrate AI into daily work.
First proof: Workshop + workflow playbookDesign instructions, examples, schemas, context, guardrails, and conversational flows.
First proof: Prompt suite + eval setCreate courses, labs, coaching, assessments, and role-specific enablement.
First proof: Bilingual mini-courseFind valuable workflows, capability boundaries, adoption risks, and product wedges.
First proof: Opportunity brief + prototypePrepare ingestion, quality, lineage, labeling, metadata, permissions, and feedback data.
First proof: Traceable data pipelineForecast, rank, detect anomalies, segment, optimize, and make decisions from structured data.
First proof: Baseline model + monitoringBuild reliable text and reasoning applications with structured outputs and tools.
First proof: Task API + UI + evalsFine-tune, distill, quantize, prompt-tune, or train adapters for specialized behavior.
First proof: Measured before/after benchmarkRetrieve, rerank, cite, abstain, manage freshness, and combine search with graphs or SQL.
First proof: Cited answer systemCoordinate tools, state, plans, queues, approvals, retries, and human handoffs.
First proof: Bounded agent workflowExpose and consume tools through APIs, MCP, A2A, events, identity, and capability catalogs.
First proof: Secure capability serverImprove code discovery, generation, review, testing, migration, documentation, and operations.
First proof: Repository agent harnessBuild realtime speech, translation, turn-taking, coaching, intake, and call automation.
First proof: Realtime voice prototypeUnderstand images, screens, forms, tables, video, diagrams, and physical evidence.
First proof: Reviewable extraction workflowCreate and edit images, video, audio, music, 3D, avatars, and localized content.
First proof: Controlled media pipelineDesign chat, copilots, ambient assistance, adaptive UI, uncertainty, and approval experiences.
First proof: Usability-tested interactionBuild datasets, rubrics, judges, simulations, regression gates, red-team suites, and human review.
First proof: Reusable evaluation harnessTrace prompts, context, tools, costs, latency, failures, drift, and user feedback.
First proof: Operational dashboard + runbookServe, route, cache, batch, scale, benchmark, and govern access to models and capabilities.
First proof: Multi-model gateway benchmarkControl token, compute, storage, review, support, and failure costs against business value.
First proof: Unit-economics modelDefend against prompt injection, data exfiltration, tool abuse, supply-chain risk, and unsafe autonomy.
First proof: Threat model + attack suiteCreate policies, inventories, risk tiers, impact assessments, controls, audits, and accountability.
First proof: Control matrix + evidenceKeep sensitive processing on-device or within controlled boundaries for latency, privacy, and resilience.
First proof: Private local deploymentMake trusted information understandable, attributable, current, and retrievable by answer engines and agents.
First proof: Entity/content evidence mapCombine language interfaces with BI, forecasting, simulation, and decision support.
First proof: Decision cockpitTransform email, documents, CRM, ERP, tickets, approvals, and back-office workflows.
First proof: Before/after workflow proofBuild domain-specific systems for health administration, legal, finance, insurance, logistics, education, retail, and more.
First proof: Domain workflow MVPAssess, prototype, integrate, secure, evaluate, and operate AI for organizations.
First proof: Diagnostic + fixed pilotPackage repeatable capability into SaaS, APIs, plugins, internal platforms, or marketplaces.
First proof: Metered capability productLet software discover, authorize, purchase, and account for data, tools, compute, or services.
First proof: Auditable paid capabilityApply models to biology, chemistry, materials, climate, engineering, simulation, and research workflows.
First proof: Reproducible research assistantConnect perception, planning, control, simulation, sensors, and embodied systems.
First proof: Simulated embodied taskEstos dominios se superponen. Un producto de voz requiere datos, UX, evaluación, seguridad y operación. Una práctica de gobierno necesita alfabetización técnica. Un equipo de robótica necesita simulación, modelos, sensores, seguridad y producto. Filtra el mapa y prueba varias ramas.
Enseña a plantear tareas, verificar resultados, proteger datos e integrar IA al trabajo diario.
Primera evidencia: Workshop + workflow playbookDiseña instrucciones, ejemplos, esquemas, contexto, límites y flujos conversacionales.
Primera evidencia: Prompt suite + eval setCrea cursos, laboratorios, coaching, evaluaciones y capacitación por rol.
Primera evidencia: Bilingual mini-courseEncuentra flujos valiosos, límites, riesgos de adopción y oportunidades de producto.
Primera evidencia: Opportunity brief + prototypePrepara ingestión, calidad, linaje, etiquetado, metadatos, permisos y feedback.
Primera evidencia: Traceable data pipelinePronostica, clasifica, detecta anomalías, segmenta, optimiza y decide con datos estructurados.
Primera evidencia: Baseline model + monitoringConstruye aplicaciones confiables de texto y razonamiento con salidas estructuradas y herramientas.
Primera evidencia: Task API + UI + evalsAjusta, destila, cuantiza o entrena adaptadores para comportamiento especializado.
Primera evidencia: Measured before/after benchmarkRecupera, reordena, cita, abstiene, gestiona vigencia y combina búsqueda con grafos o SQL.
Primera evidencia: Cited answer systemCoordina herramientas, estado, planes, colas, aprobaciones, reintentos y transferencia humana.
Primera evidencia: Bounded agent workflowExpone y consume herramientas mediante APIs, MCP, A2A, eventos, identidad y catálogos.
Primera evidencia: Secure capability serverMejora descubrimiento, generación, revisión, pruebas, migración, documentación y operación de código.
Primera evidencia: Repository agent harnessConstruye voz en tiempo real, traducción, turnos, coaching, recepción y automatización de llamadas.
Primera evidencia: Realtime voice prototypeComprende imágenes, pantallas, formularios, tablas, video, diagramas y evidencia física.
Primera evidencia: Reviewable extraction workflowCrea y edita imágenes, video, audio, música, 3D, avatares y contenido localizado.
Primera evidencia: Controlled media pipelineDiseña chat, copilotos, asistencia ambiental, UI adaptativa, incertidumbre y aprobaciones.
Primera evidencia: Usability-tested interactionCrea datasets, rúbricas, jueces, simulaciones, gates, red team y revisión humana.
Primera evidencia: Reusable evaluation harnessTraza prompts, contexto, herramientas, costos, latencia, fallas, deriva y feedback.
Primera evidencia: Operational dashboard + runbookSirve, enruta, cachea, agrupa, escala, compara y gobierna acceso a modelos y capacidades.
Primera evidencia: Multi-model gateway benchmarkControla costos de tokens, cómputo, almacenamiento, revisión, soporte y fallas.
Primera evidencia: Unit-economics modelDefiende contra inyección, exfiltración, abuso de herramientas, cadena de suministro y autonomía insegura.
Primera evidencia: Threat model + attack suiteCrea políticas, inventarios, niveles de riesgo, evaluaciones, controles, auditorías y responsabilidad.
Primera evidencia: Control matrix + evidenceMantén procesamiento sensible en dispositivo o límites controlados por latencia, privacidad y resiliencia.
Primera evidencia: Private local deploymentHaz información confiable comprensible, atribuible, vigente y recuperable por motores y agentes.
Primera evidencia: Entity/content evidence mapCombina interfaces de lenguaje con BI, pronóstico, simulación y apoyo a decisiones.
Primera evidencia: Decision cockpitTransforma correo, documentos, CRM, ERP, tickets, aprobaciones y back office.
Primera evidencia: Before/after workflow proofConstruye sistemas específicos para salud administrativa, legal, finanzas, seguros, logística, educación, retail y más.
Primera evidencia: Domain workflow MVPEvalúa, prototipa, integra, protege y opera IA para organizaciones.
Primera evidencia: Diagnostic + fixed pilotEmpaqueta capacidades repetibles en SaaS, APIs, plugins, plataformas o mercados.
Primera evidencia: Metered capability productPermite que software descubra, autorice, compre y contabilice datos, herramientas, cómputo o servicios.
Primera evidencia: Auditable paid capabilityAplica modelos a biología, química, materiales, clima, ingeniería, simulación e investigación.
Primera evidencia: Reproducible research assistantConecta percepción, planificación, control, simulación, sensores y sistemas físicos.
Primera evidencia: Simulated embodied task09 · Roles, markets, and business modelsRoles, mercados y modelos de negocio
Ways to participate and exchange value across the AI economy.
Formas de participar e intercambiar valor en la economía de IA.
A person can participate as a domain expert, builder, operator, designer, researcher, or educator. The same expertise can be delivered through employment, consulting, a managed service, a product, an API, a course, open source, or transactions.
Uses AI to improve work without owning the underlying platform.
Supplies workflow knowledge, judgment, constraints, and quality standards.
Composes applications from models, data, tools, and interfaces.
Builds datasets, experiments, adaptations, and predictive systems.
Shapes value, adoption, interaction, trust, and human control.
Provides scalable, observable, economical, reusable foundations.
Defines controls, risk, policy, assurance, and accountability.
Packages capability, domain insight, distribution, and support.
Join a product, data, platform, research, operations, or domain team.
Provide evaluation, architecture, security, data, content, or enablement part-time.
Sell a bounded outcome with fixed scope, evidence, and support.
Discover workflows, design operating models, pilot, and scale.
Monitor quality, cost, content, incidents, vendors, and controls monthly.
Package a repeatable workflow for a narrow user or industry.
Expose a specialized model, dataset, tool, connector, or evaluation service.
Distribute capabilities inside ecosystems where users or agents already work.
Teach, explain, benchmark, curate, localize, or create trusted content.
License or operate datasets, synthetic data, rubrics, benchmarks, or human review.
Build adoption publicly and monetize hosting, enterprise features, or expertise.
Charge by verified action, successful task, usage, or agent-mediated purchase.
Una persona puede participar como experta de dominio, constructora, operadora, diseñadora, investigadora o educadora. La misma experiencia puede entregarse mediante empleo, consultoría, servicio administrado, producto, API, curso, código abierto o transacciones.
Usa IA para mejorar el trabajo sin operar la plataforma.
Aporta conocimiento del flujo, juicio, restricciones y calidad.
Compone aplicaciones con modelos, datos, herramientas e interfaces.
Construye datasets, experimentos, adaptaciones y sistemas predictivos.
Define valor, adopción, interacción, confianza y control humano.
Proporciona bases escalables, observables, económicas y reutilizables.
Define controles, riesgo, política, aseguramiento y responsabilidad.
Empaqueta capacidad, conocimiento, distribución y soporte.
Join a product, data, platform, research, operations, or domain team.
Provide evaluation, architecture, security, data, content, or enablement part-time.
Sell a bounded outcome with fixed scope, evidence, and support.
Discover workflows, design operating models, pilot, and scale.
Monitor quality, cost, content, incidents, vendors, and controls monthly.
Package a repeatable workflow for a narrow user or industry.
Expose a specialized model, dataset, tool, connector, or evaluation service.
Distribute capabilities inside ecosystems where users or agents already work.
Teach, explain, benchmark, curate, localize, or create trusted content.
License or operate datasets, synthetic data, rubrics, benchmarks, or human review.
Build adoption publicly and monetize hosting, enterprise features, or expertise.
Charge by verified action, successful task, usage, or agent-mediated purchase.
10 · Learning routesRutas de aprendizaje
Free and low-cost paths spanning concepts, engineering, operations, and trust.
Rutas gratuitas y económicas sobre conceptos, ingeniería, operación y confianza.
Do not attempt to complete the internet. Build a T-shaped plan: broad literacy across the whole system, working competence in two adjacent branches, and depth in one branch. Every learning block should produce a note, benchmark, diagram, dataset, evaluation, or working artifact.
Conceptual, nontechnical introduction to AI and its societal implications.
ML vocabulary, data, loss, generalization, embeddings, neural networks.
Code-first deep learning with practical projects.
Production framing from data and training to deployment and product.
Model capabilities, prompting, APIs, and applied learning.
Code-first patterns for tools, orchestration, and multi-agent systems.
Open-source agents, tools, frameworks, and exercises.
LangGraph, agent patterns, tracing, and framework implementation.
Retrieval, indexing, evaluation, and production trade-offs.
Short applied courses and deeper programs across the stack.
Cloud AI architecture, services, operations, and certifications.
A practical vocabulary for govern, map, measure, and manage.
Risk taxonomies, threats, and mitigation guidance.
Learn interoperability directly from specifications and examples.
Spanish-language public cohorts and practitioner communities where available.
Read or watch only enough to explain the concept and its trade-offs.
Implement a small reference example without hiding behind a large framework.
Change data, model, prompt, constraints, or interface. Observe failure.
Create expected cases, metrics, adversarial tests, and a baseline.
Write a concise architecture note and record a bilingual walkthrough.
Connect the concept to one domain workflow and one user outcome.
No intentes completar internet. Construye un plan en T: alfabetización amplia del sistema, competencia funcional en dos ramas adyacentes y profundidad en una. Cada bloque debe producir una nota, benchmark, diagrama, dataset, evaluación o artefacto funcional.
Conceptos introductorios de IA y sus implicaciones sociales.
Vocabulario ML, datos, pérdida, generalización, embeddings y redes.
Deep learning práctico orientado a código y proyectos.
Visión de producción desde datos y entrenamiento hasta despliegue y producto.
Capacidades, prompting, APIs y aprendizaje aplicado.
Patrones de código para herramientas, orquestación y multiagente.
Agentes abiertos, herramientas, frameworks y ejercicios.
LangGraph, patrones, trazas e implementación.
Recuperación, indexación, evaluación y trade-offs productivos.
Cursos cortos y programas profundos en todo el stack.
Arquitectura, servicios, operación y certificaciones cloud.
Vocabulario práctico para gobernar, mapear, medir y gestionar.
Taxonomías de riesgo, amenazas y mitigaciones.
Aprende interoperabilidad desde especificaciones y ejemplos.
Cohortes públicas en español y comunidades cuando estén disponibles.
Lee o mira lo suficiente para explicar el concepto y sus trade-offs.
Implementa un ejemplo pequeño sin esconder todo dentro de un framework.
Cambia datos, modelo, prompt, restricciones o interfaz. Observa fallas.
Crea casos esperados, métricas, pruebas adversariales y una base.
Escribe una nota de arquitectura y graba un recorrido bilingüe.
Conecta el concepto con un flujo de dominio y un resultado de usuario.
11 · Tackle boardLista para abordar
A persistent, filterable backlog from orientation to operation.
Un backlog persistente y filtrable desde orientación hasta operación.
This board moves from orientation to foundations, branch sampling, a complete build, trust controls, published proof, market connection, and sustained operation. Progress is stored only in this browser. Filter without losing completed work.
Map the field and your leverage.
Learn durable concepts and basic instrumentation.
Run small experiments before committing.
Create one complete and useful system.
Make quality, risk, and recovery explicit.
Make the work inspectable and credible.
Choose users, delivery, economics, and distribution.
Sustain the system and select the next depth.
Este tablero avanza desde orientación y fundamentos hasta pruebas de ramas, una construcción completa, controles de confianza, evidencia publicada, conexión con mercado y operación sostenida. El progreso se guarda solo en este navegador.
Mapea el campo y tu ventaja.
Aprende conceptos duraderos e instrumentación.
Ejecuta experimentos antes de comprometerte.
Crea un sistema completo y útil.
Haz explícitos calidad, riesgo y recuperación.
Haz el trabajo verificable y creíble.
Elige usuarios, entrega, economía y distribución.
Sostén el sistema y elige la siguiente profundidad.
12 · Governance, security, and trustGobierno, seguridad y confianza
The control plane required for production AI.
El plano de control necesario para IA productiva.
A chatbot that only drafts text has a limited blast radius. An agent that reads customer data, executes code, sends messages, or initiates payment needs a materially stronger control plane.
Authenticate users, services, agents, and tool servers. Preserve actor and delegation chains.
Default deny. Scope every tool and data source. Recheck permissions at execution time.
Classify input/output, minimize retention, redact secrets/PII, control residency, and prevent cross-tenant leakage.
Validate arguments, constrain destinations, make writes idempotent, isolate computer use, and require approvals.
Pin and scan dependencies, verify MCP/agent servers, inventory capabilities, sign releases, and patch quickly.
Treat retrieved content and tool output as untrusted data. Separate instructions, apply policies, and constrain tools.
Trace prompts, context, model decisions, tools, results, approvals, cost, and errors without logging prohibited data.
Test normal, boundary, adversarial, multilingual, and degraded cases before and after every material change.
Name owners, risk tiers, acceptable uses, review gates, exception processes, and retirement criteria.
Provide kill switches, token revocation, rollback, containment, investigation evidence, and user notification paths.
Un chatbot que solo redacta tiene impacto limitado. Un agente que lee datos, ejecuta código, envía mensajes o inicia pagos exige un plano de control mucho más fuerte.
Autentica usuarios, servicios, agentes y servidores. Conserva cadenas de actor y delegación.
Niega por defecto. Limita herramientas y datos. Revalida permisos al ejecutar.
Clasifica entradas/salidas, minimiza retención, redacta secretos/PII, controla residencia y tenants.
Valida argumentos, limita destinos, hace escrituras idempotentes, aísla uso del computador y exige aprobación.
Fija y escanea dependencias, verifica servidores, inventaría capacidades, firma releases y parchea.
Trata contenido recuperado y resultados como datos no confiables. Separa instrucciones y limita herramientas.
Traza prompts, contexto, decisiones, herramientas, resultados, aprobaciones, costo y errores sin registrar datos prohibidos.
Prueba casos normales, límites, adversariales, multilingües y degradados antes y después de cambios.
Define dueños, niveles de riesgo, usos aceptables, gates, excepciones y retiro.
Incluye kill switches, revocación, rollback, contención, evidencia y comunicación a usuarios.
13 · Where the industry is headingHacia dónde va la industria
Directional signals, durable bets, and emerging protocols.
Señales, apuestas duraderas y protocolos emergentes.
The 2026 AI Engineer World’s Fair program is a useful directional sample: harness and context engineering, AI-native enterprise, inference, local AI, generative media, agentic commerce, graphs, security, and software factories all appear as distinct practice areas. Conference programming is a signal, not a forecast. S01S02
The agent runtime, repository context, tools, checkpoints, and feedback loop become a product surface of their own.
Attention shifts from bigger prompts to deliberate context selection, compression, permissions, memory, and provenance.
MCP standardizes application-to-capability connections; A2A targets agent collaboration; identity and trust remain active design areas.
Teams redesign workflows around model capability rather than attaching chat to old processes.
Routing, caching, batching, smaller models, local inference, and hardware-aware serving become core engineering skills.
Voice, vision, screen context, and generated interfaces make AI less like a separate chat box.
Capabilities, data, and services become discoverable and payable by software agents; mandates and audit become differentiators.
Task suites, simulations, judges, trace analytics, and release gates become normal software-delivery assets.
Prompt injection, tool abuse, identity delegation, supply-chain risk, and data leakage move into mainstream architecture.
Open-weight and small models expand deployment choices for privacy, cost, customization, and edge use.
Agents increasingly create code, workflows, reports, and situational interfaces; review and provenance become essential.
As capabilities diffuse, trusted brands, communities, proprietary workflows, and embedded channels matter more.
El programa 2026 de AI Engineer World’s Fair sirve como señal: ingeniería de harness y contexto, empresa nativa, inferencia, IA local, medios generativos, comercio agéntico, grafos, seguridad y fábricas de software aparecen como prácticas separadas. Es una señal, no un pronóstico. S01S02
El runtime del agente, contexto del repositorio, herramientas, checkpoints y feedback se vuelven una superficie propia.
El foco pasa de prompts grandes a selección, compresión, permisos, memoria y procedencia deliberadas.
MCP estandariza conexiones aplicación-capacidad; A2A colaboración; identidad y confianza siguen en diseño.
Los equipos rediseñan flujos alrededor de capacidades del modelo, no agregan chat a procesos antiguos.
Routing, caché, batching, modelos pequeños, inferencia local y serving consciente de hardware se vuelven centrales.
Voz, visión, pantalla e interfaces generadas hacen que IA deje de ser una caja de chat separada.
Capacidades, datos y servicios se vuelven descubribles y pagables por agentes; mandatos y auditoría diferencian.
Suites, simulaciones, jueces, trazas y gates se vuelven activos normales de entrega.
Prompt injection, abuso de herramientas, delegación, supply chain y fuga de datos entran a la arquitectura principal.
Modelos de pesos abiertos y pequeños amplían opciones por privacidad, costo, personalización y edge.
Agentes crean código, flujos, reportes e interfaces situacionales; revisión y procedencia son esenciales.
Al difundirse capacidades, marca, comunidad, flujos propietarios y canales embebidos importan más.
14 · GlossaryGlosario
Operational definitions for the most useful terms.
Definiciones operativas de los términos más útiles.
Terms are defined operationally. Vendor terminology varies, and emerging standards may evolve.
Optimizing trustworthy content and structure so answer systems can understand, retrieve, and cite it.
A proposed practice for visibility in generative answers. Evidence and measurement remain less mature than conventional SEO.
A controlled runtime in which a model can choose steps or tools under state, policy, budgets, and stop conditions.
A workflow that delegates selected decisions to a model while retaining explicit controls.
A bounded action or service an application or agent can discover and invoke.
Designing what information, instructions, state, tools, and evidence reach the model at each step.
A numeric representation useful for semantic similarity and retrieval.
A repeatable test or measurement of system behavior.
A broadly trained model adapted to many downstream tasks.
A control that constrains input, output, data, tools, or action. It is one layer, not a complete safety system.
Fluent output that is unsupported, incorrect, or inconsistent with required evidence.
The runtime and surrounding infrastructure that provides context, tools, state, feedback, and control to an agent.
Running a trained model to produce a result.
Protocol for exposing tools, resources, and prompts to AI hosts.
Protocol for agents to discover capabilities and exchange tasks, status, messages, and artifacts.
A form of input or output: text, audio, image, video, code, action, sensor data.
A dialog-like interface layer that temporarily focuses attention. Not the same as modality.
A system that understands or produces more than one modality.
Untrusted content attempts to redirect model behavior or exploit connected tools/data.
Retrieving external evidence before generation.
A model that reorders retrieved candidates by relevance.
Model output constrained and validated against a schema or grammar.
A model proposes a function and arguments; application code validates and executes it.
HTTP-native machine payment pattern using the 402 status family.
Protocol concepts for verifiable user mandates and agent-mediated commerce.
Las definiciones son prácticas. La terminología de proveedores varía y los estándares emergentes pueden cambiar.
Optimización de contenido y estructura confiables para que sistemas de respuesta puedan comprenderlos, recuperarlos y citarlos.
Práctica propuesta para visibilidad en respuestas generativas. La evidencia y medición son menos maduras que SEO.
Runtime controlado donde un modelo elige pasos o herramientas bajo estado, políticas, presupuestos y parada.
Flujo que delega decisiones seleccionadas a un modelo y conserva controles explícitos.
Acción o servicio limitado que una aplicación o agente puede descubrir e invocar.
Diseño de información, instrucciones, estado, herramientas y evidencia que llegan al modelo en cada paso.
Representación numérica útil para similitud semántica y recuperación.
Prueba o medición repetible de la conducta del sistema.
Modelo ampliamente entrenado que se adapta a muchas tareas posteriores.
Control que limita entrada, salida, datos, herramientas o acciones. Es una capa, no un sistema completo.
Salida fluida sin soporte, incorrecta o inconsistente con la evidencia requerida.
Runtime e infraestructura que entrega contexto, herramientas, estado, feedback y control a un agente.
Ejecución de un modelo entrenado para producir un resultado.
Protocolo para exponer herramientas, recursos y prompts a hosts de IA.
Protocolo para que agentes descubran capacidades e intercambien tareas, estados, mensajes y artefactos.
Forma de entrada o salida: texto, audio, imagen, video, código, acción o sensores.
Capa UI tipo diálogo que enfoca temporalmente la atención. No es lo mismo que modalidad.
Sistema que entiende o produce más de una modalidad.
Contenido no confiable intenta redirigir al modelo o explotar herramientas/datos conectados.
Recuperación de evidencia externa antes de generar.
Modelo que reordena candidatos recuperados por relevancia.
Salida del modelo limitada y validada contra esquema o gramática.
El modelo propone función y argumentos; la aplicación valida y ejecuta.
Patrón de pago máquina-a-máquina nativo de HTTP usando la familia 402.
Conceptos de protocolo para mandatos verificables y comercio mediado por agentes.
15 · Sources and methodologyFuentes y metodología
Traceability, limitations, and update-sensitive evidence.
Trazabilidad, límites y evidencia sensible al tiempo.
Primary sources were preferred for protocols, APIs, standards, security guidance, courses, and public programs. Marketplaces and conference agendas are treated as directional signals rather than universal forecasts.
Durable concepts are separated from time-sensitive availability. Inferences are labeled as synthesis. Course prices, cohorts, model portfolios, specifications, and market conditions should be rechecked.
The landscape is globally applicable. Local language, time-zone overlap, domain access, regulation, infrastructure, and distribution determine which opportunities are practical in a given place.
Generated 2026-07-18. Version 2.1. Self-contained HTML. No external scripts, fonts, analytics, or runtime network requests. Source IDs connect claims to the registry below.
Official conference and schedule: tracks across harness/context engineering, AI-native enterprise, inference, local AI, generative media, security, commerce, and software factories.
https://www.ai.engineer/worldsfair/2026Session-level evidence for the industry-direction map.
https://www.ai.engineer/worldsfair/scheduleOpen specification and reference repositories for connecting AI applications to tools, data, prompts, and resources.
https://github.com/modelcontextprotocol/modelcontextprotocolOAuth-oriented authorization guidance for HTTP-based MCP deployments.
https://modelcontextprotocol.io/specification/2025-06-18/basic/authorizationCurrent public roadmap for protocol evolution.
https://modelcontextprotocol.io/development/roadmapOpen protocol for discovery, task delegation, messaging, artifacts, and long-running agent collaboration.
https://a2a-protocol.org/latest/Adoption and governance signal for interoperable agent ecosystems.
https://www.linuxfoundation.org/press/a2a-protocol-surpasses-150-organizations-lands-in-major-cloud-platforms-and-sees-enterprise-production-use-in-first-yearThreats and mitigations for MCP deployments: access control, token handling, tool trust, sandboxing, logging, inventory, and data protection.
https://media.defense.gov/2026/Jun/02/2003943289/-1/-1/0/CSI_MCP_SECURITY.PDFRisk taxonomy for agentic systems.
https://genai.owasp.org/resource/owasp-top-10-for-agentic-applications-for-2026/Govern, map, measure, and manage framework for trustworthy AI risk practices.
https://www.nist.gov/itl/ai-risk-management-frameworkInterleaving reasoning and actions with external environments.
https://arxiv.org/abs/2210.03629Language models learning when and how to call tools.
https://arxiv.org/abs/2302.04761Evidence that long-context placement and retrieval strategy matter.
https://arxiv.org/abs/2307.03172Official agent runtime concepts and tools.
https://developers.openai.com/api/docs/guides/agentsRealtime multimodal interaction patterns.
https://developers.openai.com/api/docs/guides/realtimeComputer-use capability and operational cautions.
https://developers.openai.com/api/docs/guides/tools-computer-useMCP servers as app backends with tool and UI surfaces.
https://developers.openai.com/apps-sdk/concepts/mcp-serverGoogle’s realtime multimodal API documentation.
https://ai.google.dev/gemini-api/docs/live-apiOpen and commercial model portfolio and capability comparison.
https://docs.mistral.ai/models/overviewBroad taxonomy of transformer tasks and modalities.
https://huggingface.co/docs/transformers/main/en/task_summaryEnterprise language and retrieval model capabilities.
https://docs.cohere.com/reference/aboutOfficial guidance: AI-search visibility still depends on solid SEO fundamentals and useful, accessible content.
https://developers.google.com/search/docs/fundamentals/ai-optimization-guideResearch survey examining Generative Engine Optimization claims and evidence quality.
https://arxiv.org/abs/2607.14035HTTP-native payment protocol for machine-payable services and agent transactions.
https://docs.x402.org/introductionProtocol concepts for delegated, verifiable agent payments.
https://ap2-protocol.org/Payment and trust considerations for agent-mediated commerce.
https://stripe.com/en-es/guides/agentic-commercePlatform-specific demand signal for AI integration, AI video, annotation, and fractional specialists.
https://www.upwork.com/research/in-demand-skills-2026Platform-specific signal for automation, video, data transformation, and ecommerce services.
https://www.fiverr.com/resources/guides/reports/business-trends-index-june-2026Labor-market signal on accelerating skill change and skills-based profiles.
https://economicgraph.linkedin.com/research/work-change-reportExample of a major Medellín public training call in 2026; the cited call is closed, but it demonstrates ecosystem capacity.
https://rutanmedellin.org/en/news/google-c4ir-medell%C3%ADn-and-ruta-n-open-20.000-free-spots-for-training-in-artificial-intelligenceMedellín innovation and startup ecosystem updates.
https://rutanmedellin.org/noticiasLow-cost/public learning option in Colombia; availability varies by cohort.
https://betowa.sena.edu.co/oferta/iniciacion-a-la-inteligencia-artificial?location=57011001&modality=P&offertype=company&programId=168454Colombian public digital-skills program and completion signal.
https://www.mintic.gov.co/portal/inicio/Sala-de-prensa/Noticias/439554:Mas-de-104-mil-colombianos-se-formaron-en-Inteligencia-Artificial-y-tecnologias-avanzadas-con-Talento-TECHLocal practitioner community for demos, peer learning, and networking.
https://medellin.aitinkerers.org/Free, code-oriented agent curriculum.
https://github.com/microsoft/ai-agents-for-beginnersFree foundational ML curriculum.
https://developers.google.com/machine-learning/crash-course/Free open-source agent course.
https://huggingface.co/learn/agents-course/en/unit0/introductionFree and public learning materials where available.
https://academy.openai.com/pages/coursesCloud and generative AI learning paths; free and paid content vary.
https://skillbuilder.aws/generative-aiShort courses and deeper programs across generative AI, RAG, agents, and ML; pricing varies.
https://www.deeplearning.ai/coursesFocused RAG learning path.
https://www.deeplearning.ai/courses/retrieval-augmented-generationFree framework-oriented courses for agents, LangGraph, and observability.
https://academy.langchain.com/Free production-oriented deep-learning and LLM systems material.
https://fullstackdeeplearning.com/Cross-industry signal on growing demand for AI, big data, cybersecurity, technological literacy, and complementary human skills.
https://www.weforum.org/publications/the-future-of-jobs-report-2025/Occupational projections and analysis for data, software, and computer-related work influenced by AI adoption.
https://www.bls.gov/opub/btn/volume-15/artificial-intelligence-and-employment-projections.htmFree conceptual AI curriculum designed for broad public participation.
https://www.elementsofai.com/Free, code-first deep-learning course and learning materials.
https://course.fast.ai/Se priorizaron fuentes primarias para protocolos, APIs, estándares, seguridad, cursos y programas públicos. Mercados freelance y agendas se tratan como señales, no pronósticos universales.
Los conceptos duraderos se separan de disponibilidad temporal. Las inferencias se marcan como síntesis. Precios, cohortes, portafolios, especificaciones y mercado deben verificarse.
El panorama es aplicable globalmente. Idioma, zona horaria, acceso al dominio, regulación, infraestructura y distribución determinan qué oportunidades son prácticas en cada lugar.
Generado 2026-07-18. Versión 2.1. HTML autónomo. Sin scripts externos, fuentes, analítica ni solicitudes de red. Los IDs conectan afirmaciones con el registro.
Official conference and schedule: tracks across harness/context engineering, AI-native enterprise, inference, local AI, generative media, security, commerce, and software factories.
https://www.ai.engineer/worldsfair/2026Session-level evidence for the industry-direction map.
https://www.ai.engineer/worldsfair/scheduleOpen specification and reference repositories for connecting AI applications to tools, data, prompts, and resources.
https://github.com/modelcontextprotocol/modelcontextprotocolOAuth-oriented authorization guidance for HTTP-based MCP deployments.
https://modelcontextprotocol.io/specification/2025-06-18/basic/authorizationCurrent public roadmap for protocol evolution.
https://modelcontextprotocol.io/development/roadmapOpen protocol for discovery, task delegation, messaging, artifacts, and long-running agent collaboration.
https://a2a-protocol.org/latest/Adoption and governance signal for interoperable agent ecosystems.
https://www.linuxfoundation.org/press/a2a-protocol-surpasses-150-organizations-lands-in-major-cloud-platforms-and-sees-enterprise-production-use-in-first-yearThreats and mitigations for MCP deployments: access control, token handling, tool trust, sandboxing, logging, inventory, and data protection.
https://media.defense.gov/2026/Jun/02/2003943289/-1/-1/0/CSI_MCP_SECURITY.PDFRisk taxonomy for agentic systems.
https://genai.owasp.org/resource/owasp-top-10-for-agentic-applications-for-2026/Govern, map, measure, and manage framework for trustworthy AI risk practices.
https://www.nist.gov/itl/ai-risk-management-frameworkInterleaving reasoning and actions with external environments.
https://arxiv.org/abs/2210.03629Language models learning when and how to call tools.
https://arxiv.org/abs/2302.04761Evidence that long-context placement and retrieval strategy matter.
https://arxiv.org/abs/2307.03172Official agent runtime concepts and tools.
https://developers.openai.com/api/docs/guides/agentsRealtime multimodal interaction patterns.
https://developers.openai.com/api/docs/guides/realtimeComputer-use capability and operational cautions.
https://developers.openai.com/api/docs/guides/tools-computer-useMCP servers as app backends with tool and UI surfaces.
https://developers.openai.com/apps-sdk/concepts/mcp-serverGoogle’s realtime multimodal API documentation.
https://ai.google.dev/gemini-api/docs/live-apiOpen and commercial model portfolio and capability comparison.
https://docs.mistral.ai/models/overviewBroad taxonomy of transformer tasks and modalities.
https://huggingface.co/docs/transformers/main/en/task_summaryEnterprise language and retrieval model capabilities.
https://docs.cohere.com/reference/aboutOfficial guidance: AI-search visibility still depends on solid SEO fundamentals and useful, accessible content.
https://developers.google.com/search/docs/fundamentals/ai-optimization-guideResearch survey examining Generative Engine Optimization claims and evidence quality.
https://arxiv.org/abs/2607.14035HTTP-native payment protocol for machine-payable services and agent transactions.
https://docs.x402.org/introductionProtocol concepts for delegated, verifiable agent payments.
https://ap2-protocol.org/Payment and trust considerations for agent-mediated commerce.
https://stripe.com/en-es/guides/agentic-commercePlatform-specific demand signal for AI integration, AI video, annotation, and fractional specialists.
https://www.upwork.com/research/in-demand-skills-2026Platform-specific signal for automation, video, data transformation, and ecommerce services.
https://www.fiverr.com/resources/guides/reports/business-trends-index-june-2026Labor-market signal on accelerating skill change and skills-based profiles.
https://economicgraph.linkedin.com/research/work-change-reportExample of a major Medellín public training call in 2026; the cited call is closed, but it demonstrates ecosystem capacity.
https://rutanmedellin.org/en/news/google-c4ir-medell%C3%ADn-and-ruta-n-open-20.000-free-spots-for-training-in-artificial-intelligenceMedellín innovation and startup ecosystem updates.
https://rutanmedellin.org/noticiasLow-cost/public learning option in Colombia; availability varies by cohort.
https://betowa.sena.edu.co/oferta/iniciacion-a-la-inteligencia-artificial?location=57011001&modality=P&offertype=company&programId=168454Colombian public digital-skills program and completion signal.
https://www.mintic.gov.co/portal/inicio/Sala-de-prensa/Noticias/439554:Mas-de-104-mil-colombianos-se-formaron-en-Inteligencia-Artificial-y-tecnologias-avanzadas-con-Talento-TECHLocal practitioner community for demos, peer learning, and networking.
https://medellin.aitinkerers.org/Free, code-oriented agent curriculum.
https://github.com/microsoft/ai-agents-for-beginnersFree foundational ML curriculum.
https://developers.google.com/machine-learning/crash-course/Free open-source agent course.
https://huggingface.co/learn/agents-course/en/unit0/introductionFree and public learning materials where available.
https://academy.openai.com/pages/coursesCloud and generative AI learning paths; free and paid content vary.
https://skillbuilder.aws/generative-aiShort courses and deeper programs across generative AI, RAG, agents, and ML; pricing varies.
https://www.deeplearning.ai/coursesFocused RAG learning path.
https://www.deeplearning.ai/courses/retrieval-augmented-generationFree framework-oriented courses for agents, LangGraph, and observability.
https://academy.langchain.com/Free production-oriented deep-learning and LLM systems material.
https://fullstackdeeplearning.com/Señal intersectorial sobre la demanda creciente de IA, big data, ciberseguridad, alfabetización tecnológica y habilidades humanas complementarias.
https://www.weforum.org/publications/the-future-of-jobs-report-2025/Proyecciones ocupacionales y análisis sobre trabajo de datos, software y computación influido por la adopción de IA.
https://www.bls.gov/opub/btn/volume-15/artificial-intelligence-and-employment-projections.htmCurrículo conceptual gratuito de IA diseñado para participación pública amplia.
https://www.elementsofai.com/Curso y materiales gratuitos de deep learning práctico orientado a código.
https://course.fast.ai/