Job description

We're a stealth holding company acquiring and operating software businesses in overlooked, mission-critical industries - the systems that run the physical economy but have been ignored by Silicon Valley. We're rebuilding them as AI-native from the inside out. We're small, well-capitalized, and moving fast. This is not a research lab. This is a build shop with real customers, real revenue, and real P&Ls. The Role You will be the AI engineer. You'll own: 1. Building the industry-specific GPT: the reasoning layer that encodes the domain expertise of each vertical we operate in. 2. Embedding AI-native functionality directly into our AI-native system of record: not bolt-on chatbots. Core workflow replacement. Agents that do the job. 3. Moving legacy systems of record datasets into new ai-native ones: Core workflow replacement. Agents that do the job. 4. Making it repeatable: Every new company we bring into the portfolio should benefit from the playbook, model stack, eval harness, and infra you build. 5. Managing AI talent across our operating companies: you'll be a player-coach for the AI engineers embedded at each sub. Split: 90% building. 10% managing. If that ratio scares you, this isn't your role. If it excites you, keep reading. What You'll Do… - Ship a working domain-specific agent into a live customer environment in your first 30 days - Build the eval infrastructure that lets us measure whether these agents are actually replacing human workflows (not vibes) - Fine-tune, distill, or route across frontier + open models depending on the economics of each vertical - Sit with customers. Watch them work. Instrument their workflows. Rebuild them. - Write the AI engineering playbook that every future portfolio company inherits on day one Who You Are… - You are a builder first. You've shipped things people actually use. Alone or as the clear technical lead. - You are a forward-deployed engineer. You'll fly to a customer site, sit in their office, and understand the business before you write a line of code. - You've been called a 10x engineer and it wasn't in a self-review. - You have demonstrated experience building and deploying production-grade AI/LLM harnesses that connect models to customer data, tools, APIs, and business workflows, with appropriate evaluation, observability, guardrails, and failure handling. - You have strong opinions about when to fine-tune vs. prompt vs. RAG vs. agent vs harness and you can defend them with numbers. - You've built with the current frontier stack (Claude, GPT, Gemini, open models) in production, not just in a Jupyter notebook. - You are allergic to slide decks and comfortable with ambiguity, small teams, and unglamorous industries. What This Role Is Not… - A research role. We ship. - A management role. You build. - A "Head of AI" title for LinkedIn. You will be in the codebase daily. - A place for people who need process, roadmaps handed to them, or a 20-person team to feel productive. We are focused on building a diverse and inclusive workforce. If you’re excited about this role, but do not meet 100% of the qualifications listed above, we encourage you to apply. ----- Atomic is an Equal Opportunity Employer and considers applicants for employment without regard to race, color, religion, sex, orientation, national origin, age, disability, genetics or any other basis forbidden under federal, state, or local law. Please review our CCPA policies atomic.vc here.

Requirements

Must-haves

We're small, well-capitalized, and moving fast. This is not a research lab. This is a build shop with real customers, real revenue, and real P&Ls.
You will be the AI engineer. You'll own:
Building the industry-specific GPT: the reasoning layer that encodes the domain expertise of each vertical we operate in.
Embedding AI-native functionality directly into our AI-native system of record: not bolt-on chatbots. Core workflow replacement. Agents that do the job.
Moving legacy systems of record datasets into new ai-native ones: Core workflow replacement. Agents that do the job.
Making it repeatable: Every new company we bring into the portfolio should benefit from the playbook, model stack, eval harness, and infra you build.
Managing AI talent across our operating companies: you'll be a player-coach for the AI engineers embedded at each sub.
Split: 90% building. 10% managing. If that ratio scares you, this isn't your role. If it excites you, keep reading.
What You'll Do…
Ship a working domain-specific agent into a live customer environment in your first 30 days
Build the eval infrastructure that lets us measure whether these agents are actually replacing human workflows (not vibes)
Fine-tune, distill, or route across frontier + open models depending on the economics of each vertical
Sit with customers. Watch them work. Instrument their workflows. Rebuild them.
Write the AI engineering playbook that every future portfolio company inherits on day one
Who You Are…
You are a builder first. You've shipped things people actually use. Alone or as the clear technical lead.
You are a forward-deployed engineer. You'll fly to a customer site, sit in their office, and understand the business before you write a line of code.
You've been called a 10x engineer and it wasn't in a self-review.
You have strong opinions about when to fine-tune vs. prompt vs. RAG vs. agent vs harness and you can defend them with numbers.
You've built with the current frontier stack (Claude, GPT, Gemini, open models) in production, not just in a Jupyter notebook.
You are allergic to slide decks and comfortable with ambiguity, small teams, and unglamorous industries.
What This Role Is Not…
A research role. We ship.
A management role. You build.

Nice-to-haves

None flagged.

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

aillm

Platform: ashby
Effort: quick