AI-Native Full-Stack Developer

We are an AI-first company that builds SaaS products with multi-agent systems. We are a small team that loves what we do and are willing to open the door to a like-minded engineer who shares our vision and passion. So far, a fleet of two human devs, a human AI-driven PM, and a fleet of Claude Code and Codex agents under a Linear dashboard shipped 80+ stories across 5 sprints during the last 3 weeks. Real work. Tangible results visualized in PostHog. Humans steer. Agents execute. The harness makes it work. ## What is this about We're hiring a full-stack AI developer. Not a contractor. Not a freelancer we'll forget about. A potential founding team member who will help us excel our existing products for years and scale the team alongside us. You bring a keen eye for crafting engaging frontend experiences (aided by AI-assisted tools like variant.com and Stitch) and the expertise to build robust, secure backends. ## How we actually work (a kinda poetic way of how our days look): Every sprint, we decompose stories into scoped tasks for agents. Planning initially went through the BMAD method, countless Perplexity chats, scratchpad notes, and listening to our Google NotebookLM's generated podcasts. Engineer agents (defined as an .md file) implement features in isolated git worktrees using a shared Vibe Kanban dashboard, tracking both human and agent work in the same pipeline — same statuses, same quality gates, same accountability for all team members. Five reviewer agents run Claude's multi-agent team, each on a topic: adversarial QA, /agent-browser and /dogfood (skills we found and discussed last week on #random Slack channel. They turned out to be amazing AI-assisted reviewing tools). A watchdog agent monitors quality on a 5-minute /loop. When the session context is running out of fuel, you run the command /continuation-prompt (shared and committed in the project's .claude folder) before it hits the auto-compact, to get a contextualized instruction set to continue in a new and fresh session, avoiding hallucinations. All of this is actively working while you go out for a walk. This is not an idyllic post for hiring engineers. This is your way of living, and we want to welcome you as a new member. Why? Because you are like us. Your friends don't understand your work. You spend some of your free time on the tinkerer.club's Discord channel and keeping up with the latest Karpathy X posts. You are made different. ## In other words: This is not "use Copilot to autocomplete faster." This is agentic engineering: you design the harness, scope the work, coordinate agents and humans in parallel, review everything critically, and ship production features at 3-5x the speed of traditional development, with greater quality levels than traditional workflows. This blog post is your mantra: https://openai.com/index/harness-engineering/ In our company, humans orchestrate, review, merge, and make the decisions agents can't. They continuously improve the harness, learning from our (and their) mistakes. ## Please, take this seriously: If you've only read about these workflows on Twitter/Reddit but haven't actually shipped production code this way — working inside a real harness, collaborating with other developers and agents in coordinated sprints — this posting is not for you. We need someone who can prove it in a 5-minute video. (Details below.) ## What You'll Do - Own features end-to-end: requirements analysis, architecture decisions, agent-orchestrated implementation, testing, deployment, production monitoring - Orchestrate agent work: scope tasks, launch parallel agent sessions, review output critically, merge what's good, fix what's not - Build and improve the harness: context files, hooks, skills, MCP servers, GitHub Actions CI/CD pipelines, structural constraints, verification loops — the infrastructure that makes agents reliable - Collaborate in coordinated sprints alongside other humans and agents: participate in sprint planning, daily async standups (Slack), code reviews, and human+agent task coordination via shared dashboards - Make architecture decisions for full-stack applications (API + admin panel + consumer web + mobile) - You focus on the outcome and always apply professional criteria. You don't wait for your boss to make the first call; you just do it. By the end of the week, there will be tangible results on the table. Impactful commits and improvements around the harness - Document your work: process, decisions, deliverables, and delivered functionality — clearly enough that anyone (human or agent) can pick up where you left off. You know how to apply compounding engineering concepts in your day-to-day. - Grow into a team lead: help hire the next developers, take ownership over projects, mentor juniors, and shape how the team works. We are willing to increase compensation based on your results. We have the capacity to do so. It's up to you. ## Our Stack Frontend: Next.js, React, Vite, Tailwind CSS, shadcn/ui Mobile: React Native, Expo (iOS + Android) Backend: Cloudflare Workers (Hono), D1 (SQLite), R2, KV Auth: Supabase Auth Payments: Stripe (Checkout, Billing Portal, Webhooks) and Polar.sh AI Agents (a lot of stuff): Claude Code, Codex, Vercel AI SDK, MCP servers, agent-browser, browser-use CI/CD: GitHub Actions, Wrangler (Cloudflare) Team Collaboration: Vibe Kanban (cloud), Linear (our customer support and marketing push tasks — you run Codex agents from within Linear and manage the sessions until the PR is created), Cline Kanban (we are exploring it) Preferred Infra: Vercel, Cloudflare, Supabase PM Tools: Notion, Slack, Linear, GitHub — all of them deeply integrated Automation: N8N and PostHog ## What You Already Know How to Do (Non-Negotiable) These are not aspirational. These are prerequisites. If you can't demonstrate hands-on experience with most of these in your video pitch, this isn't the right role. Harness Engineering: - You know the formula: Agent = Model + Harness — and you've lived it. You've built the harness, not just used the model. - You've created and maintained context files (CLAUDE.md, AGENTS.md, .cursorrules, or equivalent) that give agents an accurate map of your codebase, conventions, and constraints - You've designed structural constraints — linters, import rules, architectural boundaries — that shrink the agent's solution space so it produces correct output more often - You've built verification loops: pre-commit hooks, automated tests, build checks. Agents verify their own work before anything gets merged. - You live by this: "Every agent mistake becomes a permanent harness fix." Your harness gets better every week. Agent Orchestration & Team Collaboration: - Every day you use agentic coding tools for real multi-file, multi-step work: Claude Code, Cursor, Codex, Cline, or Windsurf — not just autocomplete - You've orchestrated sub-agents with context firewalls: complex tasks broken into scoped sub-tasks, each running in isolation, passing structured results between agents - You've worked with background agents and long-running autonomous sessions - You've used git worktrees or similar isolation patterns for parallel agent execution - You understand context management in practice: context rot, progressive disclosure, when to compact, token budget awareness - Critical: You've collaborated with other humans under these workflows. You know what it takes to coordinate a sprint where multiple developers and multiple agents are working in parallel — task scoping, branch strategies, async communication, conflict resolution, PR review cadence. Solo agent use is table stakes. Team-based agentic collaboration is what we need. Agentic SDLC & Coordination Tools: - You're experienced with using MCP servers and skills from skills.sh in your everyday coding tasks - Maybe you've built or configured custom skills, hooks, or plugins for your agent harness - You've used AI-assisted code review: Greptile, agent reviewers, or custom review pipelines - You've practiced TDD with agents: write failing tests first, agents implement to pass - Full lifecycle ownership: development, staging, production — including CI/CD (GitHub Actions), deployment pipelines (Wrangler/Vercel), hotfixes, and production monitoring Production & Mobile: - Essential: You've deployed, monitored, and maintained live production applications — not just demos or prototypes - Expo (React Native, iOS + Android) experience is a strong plus. If you haven't, you are autonomous enough to learn -by doing- the essentials in a week - You use Notion, Slack, and Linear (or close equivalents) daily ## Please, Do NOT Apply If - You only use Copilot or ChatGPT for code suggestions — that's 2024 - You are not really interested and just for the money. Apathy is your way of being. You say "yes" to everything with no real commitment. If so, please abstain. We're looking for someone energetic who will contribute fresh ideas and actively participate in making our projects successful. If you're seeking a hands-off approach to your work, we'd encourage you to consider other opportunities - You'd suggest ASP.NET, jQuery, or PHP for a new project in 2026 - You need training on cloud-native architectures - You're uncomfortable with AI writing 80%+ of the code while you orchestrate and using AI-assisted workflows - You've read about agentic workflows on social media but haven't shipped production code with them - You've only used AI tools solo — you've never coordinated with other developers in agent-assisted sprints - AND THE MOST IMPORTANT: You're not ready to commit to full-time (40h/week) after the first month if things go well. We're building a team, not filling a slot. If you're looking for a casual side gig with no growth path, this isn't it. ## Requirements - 2+ years of professional full-stack development — years matter less than what you can prove - Fluent English — written AND spoken. You'll communicate daily via Slack and weekly via video call. We need to understand you clearly and you need to understand us. - Spanish is gold — if you speak it, say so. Part of our team operates from Latin America. - Proven experience with our stack or close equivalents — show it in your video, not just your resume - Full SDLC experience: requirements, architecture, implementation, testing, deployment, monitoring - A doer who delivers, documents, communicates proactively, and takes ownership ## The Deal - Days 1-2: Paid fixed-price trial — a real feature from our production backlog. You build it, we evaluate fit. Completable in half a day to one day. This is your chance to prove you are a fit. - The next few days: If the trial goes well, 30-day paid probation at 20h/week (rate between 10-15 USD/h, depending on our criteria). Quick onboarding (our IT Desk team will assist you with all the necessary access, including Google Workspace, Slack, a Claude Max sub, etc). You are ready for action and real sprint work starting next week. Note: full transparency, no hard feelings. Our goal is to find a mutual fit and commit for the long term. If during your probation we are not happy, we reserve the right to terminate the contract at any time. Don't worry: your goals and tasks will be clearly defined from day one. It's up to you. - Month 2: If you deliver, you scale to full-time (40h/week) with better compensation. This is the plan, not a distant maybe. Candidates not willing to scale full time after the first month, please abstain. - Long-term: You will become important in shaping our engineering team. You help hire the next developers. You shape engineering culture. You participate in the long-term success of the products we build together. Salary increase, benefits and other forms of compensation happen as we grow. Important: We're thinking in years, not months. We want someone who's building a career with us — not someone who'll disappear when a shinier contract shows up. ## HOW TO APPLY — READ CAREFULLY Your application MUST include these 3 things. Applications missing any of them will be immediately rejected. 1. STEP 1: Record a Video Pitch (5 minutes max) — THIS DECIDES WHETHER WE READ THE REST Record a screen-share video (Loom, YouTube unlisted, Google Drive — any format). Maximum 5 minutes. This is NOT a talking-head introduction or a slide deck. This is you showing your actual screen, your actual tools, your actual workflow. Your goal: prove in 5 minutes that you already work the way we described above. Please read this post multiple times. Every claim in our posting — harness engineering, agent orchestration, team collaboration, production deploys — show us the evidence. What to demonstrate (pick the strongest combination that fits in 5 min): - Your harness: a real CLAUDE.md, AGENTS.md, .cursorrules, hooks config, or structural constraints you've built and maintained. How do you apply compounding engineering concepts in your workflows? - An agentic workflow: scope a task, launch agents, review their output, iterate — show the real process - Team coordination: a task board, sprint board, or coordination system where human and agent work lives side by side — Vibe Kanban, Linear, or similar - CI/CD and verification: your pipeline, your hooks, your test results — how agents verify their own work in your system. Show your AI-assisted code review practices. - Production: a deploy, a monitoring dashboard, a live app you've maintained - Any MCP server, a great skill, or tool integration you've configured or built. Where did you find it? - Show us that you know what you are talking about. What kills your application: generic screen recordings of ChatGPT conversations, buzzword-heavy narration with no evidence, tools you installed yesterday for this video. We can tell. 2. STEP 2: Answer the Screening Questions (brief answers, 3-5 sentences each) 3. STEP 3: Portfolio Link; A GitHub profile or links showing real projects.