Senior Engineering Manager, Data & AI

Description

The team builds four connected products: an embedded agent platform with a unified chat that answers questions and takes actions, conversational analytics grounded in each customer's operating data, proprietary data intelligence from third-party, licensed, and internal sources, and AI voice and chat agents for load status, driver check calls, and follow-ups. You will lead the engineering group behind all of them.

You will own the group's technical direction, delivery, and quality, hire and develop its engineers, and stay technical by prototyping, writing evals, reading traces, and reviewing architecture in depth.

The group has about eight engineers and is growing. It also maintains the shared base the products run on: a context graph linking loads, trips, carriers, drivers, customers, documents, and conversations; event-sourced platform data streaming into the warehouse; and the evals, tracing, guardrails, and model routing every agent relies on.

The company moves more than $1 billion in funds each month through invoices, carrier payments, and driver settlements, which produces one of the richest operational datasets in freight. The team turns that data into agents that complete real work and into insights customers cannot get elsewhere.

Engineering leaders here still build, spending about a third of their time on it. You will set the norm for an AI-first team: coding agents in daily use, evals before scale, and production evidence over demos.

The stack is.NET (C, and increasingly F), TypeScript, and Python on Azure and AKS, with Cosmos DB, Postgres, Elasticsearch, Snowflake, dbt, and Fivetran.

The company is building a cloud-native operating system for freight that covers dispatch, driver mobile, safety and compliance, billing and settlements, native EDI, reporting, integrations, and automation. It was founded in 2020 by former freight insiders, more than 3,000 trucking and brokerage companies use it, and it has raised $77M to date, including a $40M Series B.

U.S. trucking is worth roughly $1 trillion today and is projected to grow about 60% to nearly $1.5 trillion over the next decade, yet most freight work is still manual. Nearly 580,000 active U.S. motor carriers operate in a fragmented market: 91.5% run 10 or fewer trucks and 99.3% run 100 or fewer, so few will ever build a data team. A single load involves brokers, carriers, drivers, shippers, and documents, each with its own identifiers, and the facts are spread across EDI, phone calls, and free text. That is why the company builds a context graph based on how freight actually moves.

Agents can now take calls, read documents, query operations data, and act inside a TMS. In freight, those actions move trucks and money, so they must be correct, auditable, and safe.

The team works from a few principles. Engineers own what they ship. Generating code and verifying it are separate steps, with tests and review run independently of the agent that wrote the code. An agent earns more autonomy only through measured precision against human-verified outcomes. Customer data stays on approved infrastructure under managed identity. Cost is an engineering constraint. Reviewed intent is stored once, and vendor-specific tooling is generated from it. Multiple model vendors are assumed, and repeated work is automated and shared within 48 hours.

Responsibilities

• Lead, coach, and develop a team of about eight engineers, including senior engineers and tech leads, and grow it as the portfolio expands.
• Hire senior and staff engineers and keep the bar high as the team scales.
• Set a standard for AI-written code: how it is reviewed, tested, and shipped, and how coding agents fit into the daily workflow.
• Set technical direction for agents, data, and the context graph, and make build, buy, and model decisions with cost, quality, and speed in view.
• Decide trade-offs across the four products on one platform, including what to build, what to cut, and where to invest.
• Own the agent lifecycle from prototype to production to continuous improvement.
• Build self-learning loops that improve agents from outcomes, user corrections, and transcripts, and measure that improvement.
• Make evals a release gate. Every agent needs tests for quality, safety, latency, and cost before it ships.
• Own the context graph and the entity resolution that keeps it accurate.
• Work with the data platform team on event streams, CDC into Snowflake, data quality, and lineage, so agents and analytics use one source of truth.
• Record decision traces covering the inputs gathered, the rule applied, any exception, who approved it, and why, so precedent is searchable while rules stay explicit.
• Combine third-party, licensed, and proprietary data into intelligence customers can act on, respecting the provenance, licensing terms, and tenant boundaries of each source.
• Design agents that change data using scoped permissions, human approval, and audit trails, and guard against prompt injection and tool misuse.
• Enforce tenant isolation and least privilege across every agent, tool, and data path.
• Own reliability and unit economics by tracking cost per task, per call, and per tenant, and keep margins healthy as usage grows.
• Spend about a third of your time building: prototyping, writing evals, reading traces, and debugging production.
• Review your team's designs and code in depth, and track model and tool changes, separating what works in production from what only demos well.
• Work as a peer to product and design leadership on direction, architecture, sequencing, and delivery.
• Work directly with customers and the Forward Deployed Engineering team, and turn repeated patterns into platform features.
• Design multi-agent systems on paper first. For each step, choose a rule, a small model, a frontier model, or a person, and measure cost per outcome rather than cost per token.
• Start with code-orchestrated workflows and add model-led autonomy only where a task needs exploration. Keep agent handoffs few, and give each agent defined inputs, allowed tools, and an output schema.
• Put limits in code, not prompts: hooks, sandboxed worktrees, and a stop condition, retry cap, and budget on every loop.
• Keep the agent that writes separate from the agent that checks, and give the checker fresh context.
• Run work in a loop of explore, plan, execute, verify, and compound, turning every correction into a rule, skill, hook, or test.
• Work fluently across Claude Code, Codex, and Cursor, and pick an editor-first or agent-first surface based on who should control the next step.
• Move every engineer to goal-based loops where specs and checks come first, are never weakened to pass, and are run by the agent itself.
• Shift the team from steering each turn to designing the system that steers the agent: its context, harness, and loop.
• Build toward continuously running agents in hosted sandboxes, each with its own least-privilege identity, a named owner, a budget, and a tested kill switch, with risk-scaled human review on financial actions, auth, settlements, and customer data.
• Close the improvement loop: capture traces, analyze real failures every week, encode fixes, and require evals and a human approver for any change to rules, memory, or evals.

Requirements

• 8+ years building software, including 3+ years managing engineers, with a record of growing senior engineers and tech leads while staying a hands-on builder who can prototype, debug production, and review code.
• Experience shipping LLM-powered products to real users, and the ability to explain what broke, what you measured, and what you changed.
• Daily use of coding agents, with examples of multi-agent workflows, contracts, hooks, and loops you put into production and the checks that kept them reliable.
• Working knowledge of tool use, function calling, MCP, orchestration, memory, and retrieval, and an understanding of where agents fail.
• Experience building eval systems that combine offline suites, online metrics, human review, and transcript analysis, and using them to gate releases.
• Solid data engineering judgment with event streams, CDC, cloud warehouses, dbt, and data quality, and an understanding of how data shape limits what models can do.
• Experience designing and running production services with real reliability requirements.
• Experience securing multi-tenant SaaS, isolating customer data, and treating agent permissions as a security boundary.
• Fluent in Python or TypeScript and comfortable reading C, F, and.NET.
• Experience operating at Series B or C pace amid ambiguity, thin support, competing priorities, and a bias toward shipping and learning.
• Ability to explain a trade-off clearly to an engineer, a customer, and an executive.
• Preferred: knowledge graphs, context graphs, entity resolution, or GraphRAG in production.
• Preferred: real-time voice agents, including telephony, speech-to-text, text-to-speech, turn-taking, and latency budgets.
• Preferred: fine-tuning, reinforcement learning from feedback, or prompt optimization against eval sets.
• Preferred: embedded automation or workflow platforms such as Activepieces, n8n, or Temporal.
• Preferred: SRE experience, or a record of running systems other teams depend on.
• Preferred: logistics or freight experience, or experience in another vertical where data is closely tied to operations.
• Preferred: founder or early-stage engineering leadership experience.
• Preferred: a track record of hiring, growing, and retaining strong engineers.
• Preferred: experience functional-language work with F or another functional language.

Conditions

Remote, based in the United States.

Some travel is required for team onsites, customer visits, and industry events.

Relocation assistance is not provided.

You report to the Head of Engineering and have a direct line into company planning.

Equal-opportunity boilerplate omitted.