Coding agents that take real tickets to merged PRs β daily, in production, at Fortune-500 scale. Not demos.
1,300+ GitHub stars Β· 3 Adobe-distributed SDKs Β· 57 MCP tools Β· 78 AI skills
Shipping developer tooling since long before agents could write it for me: ssh-deploy since 2019, Adobe's AEM SDKs, 16+ years of enterprise platforms. I build agent-ready codebases.
I make enterprise codebases agent-ready β the rules, skills, MCP servers and guardrails that let AI coding agents work in large, complex repositories. Then I redesign delivery around them: find where it breaks when agents arrive (QA bottlenecks, poorly scoped requirements, review debt), fix it, and bring the whole team along, not just the engineers already excited.
Under that sits tooling built for the constraints enterprises actually have β governance, multi-environment deployments, team-scale coordination:
- the protocol layer (MCP servers),
- the orchestration layer (skills, agents and plugins shared across Claude Code, Copilot CLI and VS Code Chat),
- the execution layer (agents running as CI pipelines, triggered from the ticket),
- the measurement layer (evals with known-answer fixtures).
- Issue-to-PR agents β requirements, readiness checks, bug fixing, development, QA, docs; each triggered by a comment on the ticket and running in CI. 1/3 of all tickets now ship through this spec-to-PR workflow. Presented to the COO, who made it a program priority the same day.
- Autonomous AI PR review β hundreds of pull requests reviewed in daily delivery, at about $5 per review β a fraction of senior-reviewer time.
- 5 enterprise repositories made agent-ready β skills, agents, rules files, guardrails (policy-as-code, token budgets, secret-blocking hooks).
- Evals that changed decisions β an LLM judge passed a report its own rubric marked as a fail; the deterministic grader caught it, so deterministic graders now run first.
- Model tiering β Opus for deep review, Sonnet for execution, Haiku for lookups.
- Scale β 100+ markets, millions of users: a multi-brand AEM platform with 500+ components and 4+ market teams.
The orchestration layer. 78 skills, 13 agents, 4 plugins running identically across Claude Code, GitHub Copilot CLI and VS Code Chat. Ticket to PR, autonomous CI agents, AI PR review, eval layer. Config-driven, never hardcoded. Shipped ~6 months before Salesforce productized the "Agentic Engineering" category.
The protocol layer. 57 MCP tools that give AI agents direct access to AEM β JCR content, components, dialogs, page operations. One of the first full MCP servers for AEM, live since July 2025. Listed on PulseMCP.
AI parliament simulation. 6 autonomous agents model German political parties using Claude + Grok and the Vercel AI SDK. Multi-provider orchestration, circuit breakers and coalition negotiation, running a full 4-year parliamentary term for ~$44. The project people find me through.
GitHub Action for SSH deployments. 1.3k+ stars, used in CI/CD pipelines worldwide.
A self-hosted n8n agent that triages new GitHub issues: code-aware triage comment, labels, possible duplicates. About one cent per issue; 8 security layers against prompt injection on public input; GitOps deploy.
Lead contributor to Adobe's official AEM Headless SDK (one of 3 Adobe-distributed SDKs I delivered) Β· the frontend toolchain adopted across Netcentric Β· a zero-config mock API server.
- AI developer platforms β agents for engineers: skills, subagents, hooks, plugins, MCP
- Context engineering β rules files, AGENTS.md, skills for large, complex repos
- Agent safety β policy-as-code, token budgets, review gates, prompt-injection defense
- Evals / LLMOps β known-answer fixtures, deterministic graders before LLM judges
- AEM platform engineering β AEM is the depth; making it AI-operable is the edge
- adaptTo() 2023 β "AEM Headless: A Glimpse of Developer Tools"
- Frontend Coffee Break β Episode #25: "Open Sourcing Our FE Build"
- Netcentric Blog β "AI-Driven Code Generation: AEM Component Development from Figma" (2025)
- ThoughtWorks Berlin 2019 β ML prototype turning hand-drawn wireframes into rendered web components
- Company-wide talks on AI-enabled development (audiences up to 200+)
Enterprise delivery where agents handle execution end-to-end, and humans set direction and own the merge.
Making a large codebase agent-ready? Let's talk β LinkedIn
Berlin Β· @draganfill Β· frontenddot.com