Assemble provides a team of 34 specialized AI agents, each with a distinct role, expertise, and personality. Together, they cover the full spectrum of software product development — from architecture to marketing.
An agent is a role-specific system prompt that transforms a general-purpose LLM into a domain expert. Each agent has:
A name and identity (Marvel-inspired for memorability)A domain of expertise (backend, frontend, security, SEO, etc.)An approach (how they think and work)An intervention sequence (their step-by-step methodology)Anti-patterns (what they never do)Quality rules (standards they enforce)
This is deliberate prompt engineering . Research shows that LLMs respond better to well-defined personas with rich backstories. By using Marvel characters:
Each agent has a distinct personality that prevents role confusion
The metaphors reinforce behavior (Spider-Man cares about users like Peter Parker cares about people)
It makes the system memorable and fun to use
It creates natural team dynamics (Tony Stark and Captain America disagree constructively)
Agent Alias Domain Specialty Spider-Man @spider-manFrontend React, Next.js, UI/UX, accessibility, performance Bruce Banner @bruce-bannerBackend APIs, Node.js, Python, microservices Mr. Fantastic @mr-fantasticFullstack End-to-end features, debugging, integration Doctor Strange @doctor-strangeDatabase Schema design, queries, migrations, optimization Ant-Man @ant-manMobile React Native, Flutter, mobile-first design Thor @thorDevOps/SRE CI/CD, Docker, Kubernetes, infrastructure
Agent Alias Domain Specialty Hawkeye @hawkeyeQA/Testing Unit tests, E2E, regression, coverage Punisher @punisherSecurity Vulnerability audits, pentesting, hardening Microchip @microchipRed Team Offensive security, exploits, injection testing
Agent Alias Domain Specialty Professor X @professor-xProduct Requirements, specs, user stories, OKRs Nick Fury @nick-furyAnalysis Business analysis, requirements gathering Captain America @captain-americaScrum Sprint planning, agile ceremonies, task breakdown Pepper Potts @pepper-pottsCustomer Success Onboarding, churn prevention, NPS Iron Fist @iron-fistFinance Pricing, P&L, budget, financial modeling
Agent Alias Domain Specialty Tony Stark @tony-starkArchitecture System design, scalability, tech decisions Vision @visionAI/ML LLM integration, fine-tuning, ML pipelines Quicksilver @quicksilverAutomation Workflow automation, RPA, n8n/Zapier
Agent Alias Domain Specialty Star-Lord @star-lordMarketing GTM strategy, positioning, campaigns Rocket Raccoon @rocket-raccoonGrowth Acquisition funnels, conversion optimization Gamora @gamoraPaid Media Google Ads, Meta Ads, campaign management Black Widow @black-widowTechnical SEO Crawl optimization, indexation, Core Web Vitals Storm @stormContent SEO Blog strategy, articles, keyword research Jean Grey @jean-greyGEO/AIO Generative search optimization
Agent Alias Domain Specialty Loki @lokiCopywriting Headlines, taglines, messaging frameworks Black Panther @black-pantherBrand Brand identity, tone of voice, guidelines Silver Surfer @silver-surferStorytelling Narrative arcs, brand storytelling Ms. Marvel @ms-marvelSocial Media Community management, content calendar Phil Coulson @phil-coulsonPR/Comms Press releases, media relations
Agent Alias Domain Specialty Invisible Woman @invisible-womanUX Design Wireframes, user flows, design systems Beast @beastData/Analytics Dashboards, metrics, data pipelines
Agent Alias Domain Specialty She-Hulk @she-hulkLegal GDPR, compliance, terms of service
Agent Alias Domain Specialty Jarvis @jarvisOrchestrator Routes requests, manages workflows, coordinates agents Deadpool @deadpoolContrarian Devil’s advocate, challenges assumptions, flags risks Doctor Doom @doctor-doomStress-Tester Strategic demolition of critical decisions
Agents don’t work in isolation. Jarvis orchestrates them in sequences called workflows . For example, a /feature workflow might chain:
@professor-x writes the spec@tony-stark designs the architecture@captain-america breaks it into tasks@bruce-banner + @spider-man implement@hawkeye writes tests@punisher reviews security
Each agent receives context injection — they know what previous agents produced, what’s expected of them, and what comes next.
You can bypass Jarvis and talk to any agent directly:
@spider-man refactor the dashboard to use Server Components
@bruce-banner add rate limiting to the API
@loki write a tagline for the pricing page
The agent stays in character until you dismiss them with /dismiss.