Junior Software Engineer

Software Engineer: Agentic Workflow \& Craftsmanship

Department: Engineering

Classification: Full-time, Exempt

Reports to: Senior Engineer / CTO

Location: US-only. No visa sponsorship. [Remote]

About the role:

Monarch Technology is a startup building enterprise agentic software. We're a team of senior engineers who use AI to magnify our skills, which is a different sport than vibe-coding. In enterprise software, cost of change dominates everything else. If we were shipping mobile games or marketing sites, we'd have a different opinion. Modern best practices are changing every few months. The durable skills are related to craftsmanship and tinkering: taste, willingness to delete work, comfort with empirical iteration, the instinct to try the weird thing on a Saturday, and the discipline to measure rather than assume. You'll write code with agents, write code that orchestrates agents, build the surrounding quality machinery, and grow into the practices that hold up under real load. You'll have senior engineers around you who've shipped this stuff before; your job is to learn fast, contribute hard, and develop your own opinions through real work.

We live in the agentic world. It's different over here.

We want someone who:

• (A) Is hungry for the XP / Craftsmanship model, including the human interactions.
• (B) Likes startup speeds; build it today, release it tonight, refactor it tomorrow. Speed and maintainability.
• (C) Has jumped into the AI space and tinkered their way to a personal point of view on how they work in a modern workflow.
• (D) Communicates well in English and in code. Massive collaboration, mostly verbal.
• (E) Has wrestled with modern agentic development beyond prompt refinement, even if that wrestling happened on side projects, not at scale yet.

Specifically:

• Practical, recent experience using coding agents for real work, with honest views about where they help and where they generate confident garbage.
• Curiosity about building tools and pipelines that tighten the agent loop and shorten feedback. Has built and shipped real, working AI agents, at a job, on the side, or as personal projects, as opposed to just reading about them or playing with tutorials.
• Starting to understand that prompts shift distributions; they do not enforce bounds. You don't need a finished architectural answer; you need to be looking for one.
• Refactoring instinct: able to take a messy module to a cleaner one in small, safe steps. We'll help you sharpen this.
• Open to methodology: XP, Lean, and related traditions. Willing to try what the team uses and form your own opinions through experience.
• Comfortable enough with statistical thinking to learn it on the job; confidence intervals, composition arithmetic, residual risk.
• Clear written communication: can explain a tradeoff in a paragraph.

#1 Requirement Recency over duration:

You have hands-on agentic work from this year. In 2022, AI was different from ML because AI was written in PowerPoint. The best ways to use these tools, and the best tools themselves, have changed half a dozen times this year. Pre-2026 experience with agentic coding is background, not signal. For this role, your years of professional software experience matter less than what you've been doing in the last six months. Prompt-engineering, intent-engineering, and algorithmic guardrails around LLM outputs are warmups. A motivated child could do those. We want to hear about what you've tried that's actually interesting; even if it broke, even if it was small.

What we're not impressed by:

• Spec-centric solutions; "I can simply prompt the agent better and it'll do what I want." Specs are a useful component inside a stacked architecture. They are not the architecture. If your answer is always "make the spec better," we're going to push.
• Massive hardcoded limiters; agent does X, but 3,000 lines of algorithmic control on 20 lines of spec. Not a real solution either. If your answer is always "put guardrails on the spec," we're going to push.
• Resume-decorated alphabet soup. "I've done RAG, evals, tool use, MCP, observability." Cool; most people have, and most of it sucks. Tell us about specific things you've built and what changed when you ran them for real.
• Performative obsession with token cost. We'll monitor it eventually. It's not a focus today. If you lead with "I cut prompt tokens by 40%," we'll roll our eyes.
• Two failure modes we screen for: only ever doing spec-centric work, or rejecting specs entirely. The candidate we want is somewhere in the middle of that journey; has tried spec-centric, started bumping into the limits, and is actively figuring out how specs fit as components inside a stacked architecture.
• You don't need to be done with that journey. You need to be on it.

Minimum requirements:

• Education: Bachelor's in Computer Science, related field, or equivalent demonstrable experience.
• Experience: Some production or production-adjacent software experience with visible craftsmanship discipline; internships, junior roles, serious open-source contributions, or substantial personal projects all count. Recent (THIS-year) hands-on engagement with current-generation coding agents (Claude Code, Codex, or comparable), and evidence of having changed your mind at least once about how to use them. Duration is not the metric. Recency and disposition are.
• Systems: Working proficiency in at least one mainstream language (JavaScript/TypeScript, Python, Go, or similar). Comfortable with Git, CI pipelines, and modern test tooling. Familiar with at least one agent-coding tool beyond surface-level use.
• Location: US-only. No visa sponsorship.

How to apply:

AIs generate resumes. AIs read resumes. People have been "exaggerating" since resumes were invented. On average, there isn't a ton useful information in one. We still want yours. It's a paper trail and we read it as such. The actual conversation starts after you apply. That's where we find out whether we should be talking.