About this role
Summary: Own the architecture and technical direction of Distru's AI product suite, including multi-agent systems, MCPs, and context layers. Build production-grade agentic harnesses, ensure reliability and observability, and drive AI research and experimentation within a remote, high-agility engineering team.
About Distru
Distru is the [#1 ERP in the cannabis industry](https://www.g2.com/products/distru/reviews) with $15B in sales processed through our platform. We've built a battle-tested ERP and B2B eCommerce platform that powers inventory, manufacturing, sales, and logistics, and we're now building AI infrastructure on top of it to give commercial businesses their all-in-one AI stack.
We're growing and profitable, with [5/5 stars on Glassdoor](https://www.glassdoor.com/Reviews/Distru-Reviews-E2957092.htm). Our investors include [Felicis Ventures](https://www.felicis.com/), [Global Founders Capital](https://www.globalfounders.vc/), [Poseidon Asset Management](https://www.poseidonassetmanagement.com/), and [Village Global](https://www.villageglobal.vc/), who have funded companies such as Shopify, Cruise, HelloFresh, Facebook, Slack, and Opendoor.
Get to know the team, our culture and our full benefits at https://distru.com/careers!
The Role
AI is at an inflection point for enterprise use, and AI assistants like Claude, Codex, and Grok are seeing record adoption at work. We believe the system of record has never been more valuable for organizing the data AI acts on.
We started with our AI Order Agent, now rolled out across our customer base, which lets buyers submit orders from any source in any format. It's the first piece of a much bigger agentic ecosystem we’re building, and we're moving most of our engineering effort to AI-first work to build it.
You'll own the architecture and technical direction of our AI product suite, partnering with our Director of AI Engineering, including:
- Agentic harness: a multi-agent system that safely operates across a business's core tools (ERP, CRM, spreadsheets, calendars, email, analytics, compliance systems)
- MCP: our MCP that connects Distru to external AI harnesses
- Context Layer: our context layer that ingests customer data, builds memories and profiles, and feeds back into our harness to create more personalized AI experiences for our customers
- Evals: the eval suites that make our agents reliable in production
- And more…
You’ll be working directly on the AI system from the ground up and make strong, opinionated calls on the agentic stack and where it goes next. The AI space moves fast and the specifics will change, so this technical vision is something you'll mold.
You'll report to the VP of Engineering and the Director of AI Engineering, and work closely with the rest of the product team, eng team, and our customers.
### What you'll do
Agentic engineering
- Architect and build a production multi-agent system that can safely take actions across Distru and our customer’s core systems
- Iterate and maintain the MCP that connects Distru to Claude & Codex.
- Own reliability, observability, and durable execution end to end. The harness takes real actions on customers' data, so the bar is high.
- Write evals to improve our agentic systems from real customer use cases
- Build a unified data system that pulls data from multiple sources into a singular data layer that can be exposed to Claude & Codex via an MCP
- Build a contextual memory system that ingests our customers' data and improves how our agents work for them
- Set the architectural standards for how agents are built, tested, and shipped at Distru, and bring the rest of eng along
AI research & experimentation
- Keep pace with the latest agentic frameworks, harnesses, toolkits, models, and bring in better approaches as they emerge
- Rapidly prototype new ideas to present to our team and our customers
Engineering workflow
- Build hands-on with Claude Code and similar tools as part of your own workflow
- Improve our agentic coding workflow and bring new discoveries to the team
### What success looks like at 6 months
- You've shipped core improvements to our harness in production that our customers use daily, with clear observability and iteration for improvement
- You’ve shipped and iterated on many MCP tools and improvements
- Every agent you own has an eval suite and observability, so the team catches regressions before customers do
- You own the architecture of the agentic stack, and other engineers build against patterns you set
- You’ve built prototypes on how our agentic infrastructure can improve (i.e. a new way to retrieve memories, an agentic onboarding process for our customers)
- You’ve raised the technical bar on at least one thing the team was doing the slow way
### What we look for
AI fluency
For every AI role, we look for three things:
- Tool fluency: you build with Claude Code or similar every day, you've built your own skills, agents, or workflows, and you can explain why your setup looks the way it does.
- Workflow fluency: you write evals before you trust a prompt, and you know when a model is the wrong tool and plain code is the right one.
- Build fluency: you've put agents in front of real users, with guardrails, observability, and a way to roll back.
You use AI tools daily and can tell us specifically where your fluency is strong and where you're still building it.
Must have
- 7+ years in full-stack development, ideally on B2B SaaS applications
- Multi-agent harnesses: hands-on experience building multi-agent harnesses in production, not just prototypes or personal projects
- MCPs: you've built MCPs that are used in production
- Production systems: a strong grasp of reliability, observability, permissions, and safe workflow execution
- Product and engineering: you've operated as both, and you're comfortable owning scoping and decisions, not just implementation
- Modern stack: proficiency in React, TypeScript, Node, PostgreSQL, GraphQL, and LangGraph, and you pick up new tools and frameworks quickly
- Startup agility: you've worked at an early-stage startup (under 20 engineers) and you’ve been the one setting direction in that ambiguity, not waiting for it
- Low ego: you're team-first and measure success by what shipped and whether it worked, not by internal politics
- Influence: you’ve shaped roadmap, not just executed it
- Clear communication: you raise concerns and propose solutions before being asked
Strong signal
- Background building supply chain, ERP, CRM, B2B marketplace, or accounting software
- UX design experience
- Agentic systems work beyond the must-haves, such as evaluation frameworks or agent-to-agent protocols
- Experience using Claude Code (or similar) to build products on your own
- Cannabis industry experience
### How we work
- We love building with AI. Trying new models the week they drop, building agents for fun, and forming strong opinions on what works. We invest in it with company-wide AI growth with classes and workshops, AI demo days, AI hackathons, and access to best-in-class models and tools.
- Taste in building. A lot of this role comes down to knowing what to build, what to cut, and what a great agentic experience feels like for the user.
- We stay close to the customer. You'll talk to customers and our GTM team directly, not through a chain of tickets.
- You'll build on what we've learned. You'll have access to the team's combined AI knowledge and the processes and skills we've built.
For the full picture of our culture and who thrives here, see https://distru.com/careers.
Where you land in the range depends on your experience and skills.
Benefits
Our [careers page](http://distru.com/careers) is the best way to get to know us: the full benefits, our yearly offsites, who thrives here, and what the team says about working at Distru.
Here are a few highlights (exact benefits vary by country and employment type):
- Fully remote across 4 continents
- Employee-friendly option terms: after 2 years, a 10-year window to exercise if you leave
- 18 days PTO + 11 holidays (sick days don't count toward PTO)
- Medical, dental, vision + 401(k) for US employees
- The best AI tools available and a generous AI stipend for everything else
- Yearly in-person team offsites
Interview process
- Application questions: three short written questions in Lever on how you build with LLMs
- Recruiter screen (30 min): your background, the most complex AI product you've worked on, and your personal AI tool stack
- Technical interview (45–60 min): a deep dive into a complex product you've built, a live harness design discussion, and a screen-share walkthrough of your day-to-day setup
- Harness take-home (3–4 hours): spec a real feature on our agentic harness across product and UX, system design, and implementation details
- Take-home review (60 min): a technical deep dive into your submission with two engineers, including how your design holds up when a constraint changes
- References and offer
Distru was formed by an environment of mutual respect and trust made up by people working together without regard to race, color, religion, sex, pregnancy (including childbirth, lactation and related medical conditions), national origin, age, physical and mental disability, marital status, sexual orientation, gender identity, gender expression, genetic information (including characteristics and testing), military and veteran status, and any other characteristic protected by law. We believe that diversity and inclusion among our team members is critical to our success and we proactively strive to recruit, develop and retain the best and brightest people from the most diverse candidate pools we can find all over the world.