Senior Software Engineer, Agentic Development

The role

We’re hiring Software Engineers who build production software through AI coding agents and design the system that makes that reliable. This posting has no level and no years requirement on purpose. If you have the evidence, we want to talk, whether you’re two years in or twenty. You’ll write specs, not boilerplate. You’ll review, test, and own what ships. Most of your code will be generated. All of it will be yours.

Why this role, why now

Total Expert builds the customer engagement platform that banks, credit unions, and mortgage lenders use to turn data into relationships. Our customers are regulated, run on systems they can’t rip out, and need software that fits how their people work. The code has to be right the first time it touches a customer.

Execution isn’t expensive anymore. Engineers who treat agents as their primary way of building are shipping faster than people with twice their tenure, and the gap is widening. We want those engineers, and we want them to raise how the whole team builds.

This is a hybrid role based in our St. Louis Park, MN office. We ask that you live within commuting distance or be willing to relocate at your own expense.

What you'll be doing:

• Ship production features with agents writing most of the code. You own the spec, the verification, the review, and the result. If it breaks, it’s yours.
• Build and maintain the harness. Instruction files (CLAUDE.md, AGENTS.md), skills and commands, MCP tools, hooks, and the tests, linters, type checks, and security scans the agent has to pass before you ever look at a diff.
• Decompose before you delegate. Turn ambiguous business problems into small, verifiable slices an agent can execute. Plan first. Review the plan before any code exists.
• Hold the quality line. Tests before implementation. Review for correctness and design, not just syntax. Security and compliance by default. Observability on everything that ships.
• Measure what works. Track cycle time, defect escape rate, and rework on agent-generated code, and change the harness when the numbers say to.
• Make the team faster. Turn your setup into the team’s setup. Teach it, document it, pair on it.
• Get close to customers. Watch how loan officers and marketing teams actually use what you build, and let that change what you build next.
• Build the thing nobody asked for. When you see a business problem worth solving, bring a working prototype, not a proposal.

How we level

One posting, every level. We don’t screen on years because years no longer predict impact in this kind of work. Your title, level, and place in the compensation range are set at offer from what you’ve shipped, the harness you’ve built, and how you handle the agent being wrong. An engineer with three years and that evidence will level above a fifteen-year veteran without it.

What we're looking for:

Required

• Shipped production software where coding agents wrote most of the code. You can name the product, your share of it, and what moved as a result.
• You’ve designed or materially improved a harness that other engineers use, and you can show it or describe it file by file.
• Strong fundamentals in JavaScript/TypeScript and modern frameworks (Vue, React), object-oriented languages (C#, PHP, Java), relational databases (MySQL, PostgreSQL, SQL Server), and queuing technologies (Kafka, RabbitMQ), plus data modeling, APIs, and distributed systems. Deep enough to know when the agent is wrong.
• A real story about an agent being confidently wrong, how you caught it, and what you changed so that class of error gets caught automatically now.
• Testing, CI/CD, code review, and observability as your quality system, not as chores you do after the fact.
• Public evidence of your work: repos, prototypes, writeups, or talks. Portfolios beat resumes here.
• Bachelor’s degree in computer science, software engineering or related field
• Hunger is required.

Preferred

• Experience in fintech, regulated industries, or with compliance constraints.

How you work

• Reality is the referee. You argue from outcomes and evidence, not from “the model handles it” or “that’s how we did it at my last company.”
• Disagreement is a duty. You push back on the agent, on the spec, and on the team when something’s wrong, and you make it safe for others to do the same.
• Fix the system, not the person. When an agent ships a bug, you fix the harness so the whole class of bug is gone.
• You learn fast. You picked up your current tools in days, and you’ll say “I don’t know” without flinching.
• You build to communicate. You’d rather show a working prototype than write the deck.

What this role isn't

• Not a prompt-and-paste role. Generating code is the easy part. Owning it is the job.
• Not a research role. You ship to real customers in a regulated industry.
• Not a solo act. If your workflow only works on your laptop, it doesn’t count.