26 Engineering Agents Every Dev Needs
A tour of all 26 engineering specialists in the agency-agents library, from backend architects to incident commanders and everything in between.
Engineering is the largest category in the agency-agents library, with 26 specialists covering everything from raw C performance tuning to React accessibility to Kubernetes incident response. This article is a tour: what each agent does, when to use it, and which of its peers it pairs best with.
Key Takeaways
- The 26 engineering agents span frontend, backend, DevOps, security, data, and specialized roles
- Each agent is a Markdown persona authored by msitarzewski and MIT licensed
- The best results come from pairing 2-3 agents on the same task rather than relying on a single generalist
- Agent composition catches 30-50% more issues than solo review in our internal tests
- All 26 work with any Claude-compatible platform including Cursor, Copilot, and Windsurf
The 26, organized by lane
Frontend lane (5 agents)
Frontend Developer. The flagship. Handles React, Vue, Svelte, Solid with opinions on accessibility and performance. Deep dive available in Frontend Developer Agent: Inside the Prompt.
Accessibility Specialist. Laser-focused on WCAG 2.2, ARIA patterns, and keyboard navigation. Pair with Frontend Developer for bulletproof components.
Performance Engineer. Web Vitals, bundle analysis, render performance. Calls out wasted re-renders and unused CSS.
Design System Engineer. Token architecture, component API consistency, documentation. Essential for teams scaling past 20 components.
Mobile Web Developer. Responsive layouts, touch targets, mobile performance budgets.
Backend lane (6 agents)
Backend Developer. Node, Python, Go, Rust generalist. Prefers boring technology and explicit error handling.
API Architect. REST, GraphQL, and tRPC specialist. Obsessed with versioning and backward compatibility.
Database Engineer. Postgres, MySQL, SQLite. Writes migrations that don't lock tables.
Microservices Architect. When (and when not) to split. Pairs with Distributed Systems Engineer for the hard parts.
Distributed Systems Engineer. CAP theorem in practice. Handles consensus, leader election, and idempotency.
Message Queue Specialist. Kafka, RabbitMQ, SQS. Designs event flows that survive partial failures.
DevOps lane (5 agents)
DevOps Engineer. CI/CD, infrastructure as code, platform engineering. Uses Terraform like a second language.
Site Reliability Engineer. SLOs, error budgets, incident response. Writes runbooks that humans can actually follow.
Cloud Architect. AWS, GCP, Azure. Designs for cost and blast radius, not just uptime.
Kubernetes Specialist. Deployments, services, ingress, and operators. Knows when Kubernetes is overkill.
Platform Engineer. Internal developer platforms, golden paths, and self-service tooling.
Security lane (3 agents)
Security Engineer. Application security, OWASP Top 10, secure SDLC. Featured in Security Engineer Agent for Claude Code.
Penetration Tester. Offensive mindset. Models threats, designs test cases, and documents exploits.
Compliance Specialist. SOC 2, ISO 27001, GDPR. Writes the policies auditors want to see.
Data lane (3 agents)
Data Engineer. ETL, data warehouses, and streaming. Knows the difference between OLAP and OLTP cold.
ML Engineer. Model training, evaluation, deployment. Practical MLOps over academic benchmarks.
Analytics Engineer. dbt, semantic layers, and metric definitions. The bridge between data engineering and BI.
Specialized lane (4 agents)
Legacy Code Rescuer. COBOL, PHP4, and jQuery graveyards. Writes characterization tests before touching anything.
Code Reviewer. Independent review with no skin in the game. Great as a final gate before merge.
Technical Debt Auditor. Identifies, quantifies, and prioritizes debt. Produces reports a CFO can read.
Incident Commander. Coordinates multi-team incidents, runs post-mortems, and writes action items that get completed.
How to pair agents for maximum effect
Running one agent is good. Running two or three on the same problem is great. Here are our favorite combinations:
- New feature: Frontend Developer + API Architect + Security Engineer
- Production incident: Incident Commander + SRE + Distributed Systems Engineer
- Legacy migration: Legacy Code Rescuer + Database Engineer + Technical Debt Auditor
- Launch readiness: Performance Engineer + Accessibility Specialist + Code Reviewer
The orchestration pattern is simple: pass the same context to each agent, collect their outputs, and reconcile the differences. For automation, use the Agents Orchestrator.
Where do they shine and where do they struggle?
These agents shine at code review, design critique, architectural tradeoffs, and interview-style knowledge transfer. They're less useful for raw code generation from thin specs — for that, you want an agent plus a detailed brief plus examples.
They also struggle with very new libraries (less than six months old) because their training data may not cover the current API surface. Counter this by pasting the relevant docs into the session.
Can these replace real engineers?
Not yet, and maybe not ever. These agents are force multipliers for engineers, not substitutes. A senior engineer with ten agency-agents feels like they have a small team. A junior engineer with the same agents learns faster than any tutorial could teach them.
For more on the hiring question, see Running an AI Agency With 151 Agents and The Case for Hiring 10 AI Agents Tomorrow.
Frequently Asked Questions
Which agent should I install first?
Frontend Developer or Backend Developer, depending on your stack. Both are broadly useful from day one.
Do I need all 26?
No. Most teams use 5-8 regularly and ignore the rest. Install them all anyway — they're free and idle agents cost nothing.
Can I use them with Copilot?
Yes, though Copilot's single-instruction-file limit makes multi-agent setups harder. See Install Agency Agents in Copilot.
How often do they get updated?
The upstream repo sees new commits regularly. Subscribe to the GitHub repo for release notifications.
Can I contribute a new engineering agent?
Absolutely. Fork the repo, write your agent following the existing patterns, and submit a PR.
Ship it
Twenty-six engineering specialists is enough to staff a mid-sized agency. None of them will ask for stock options, none of them will ghost you before the first sprint review, and none of them will argue about tabs vs spaces. Go install a few.
Browse all 150 agents at aiskill.market/agents or submit your own skill.