AI Agent Engineer

Dexisphere; United States; Remote; USD 120,000 - 190,000; Full time

Benefits offered by this job

Equity with long exercise window

Remote-first across Europe or Africa

Learning budget $1,500 per year

Home-office stipend $1,000

Laptop of choice

Job summary

Dexisphere is seeking an engineer to make AI agents reliable at real tasks, owning instructions, tool calls, evaluations, and model choice for server-side systems that find prospects, verify emails, move deals, book appointments, and send outreach.

You will focus on planning, tool calls, approval flows, and recovery when things go wrong, comparing Claude, GPT, and Gemini on cost, speed, and quality, and building guardrails from failed turns.

Qualifications

• 3+ years of software engineering with an LLM-powered feature shipped to real users.
• Hands-on experience with tool or function calling and structured outputs.
• An evaluation mindset: measuring before believing an improvement.
• Solid backend fundamentals including APIs, queues, retries, and observability.
• Good judgment about when an agent should act and when it should ask.

Responsibilities

• Improve how agents convert plain-language tasks into correct sequences of tool calls.
• Build evaluations from real workflows and track them over time.
• Tune instructions and approval behavior for precise pausing points.
• Compare Claude, GPT, and Gemini models on cost, speed, and quality.
• Investigate failed turns end to end and turn findings into fixes and guardrails.

Skills

LLM feature delivery

Tool calling

Structured outputs

Backend fundamentals

APIs

Observability

Evaluation mindset

Decision making

Tools

LLM APIs

Job description

Role overview

Make AI agents reliable at real tasks, owning instructions, tool use, evaluations, and model choice for server-side systems that find prospects, verify emails, move deals, book appointments, and send outreach. The work focuses on planning, tool calls, approval flows, and recovery when things go wrong.

Responsibilities

• Improve how agents convert plain-language tasks into correct sequences of tool calls.
• Build evaluations from real workflows, such as prospecting runs and campaign sends, and track them over time.
• Tune instructions and approval behavior so agents pause and ask at exactly the right moments.
• Compare Claude, GPT, and Gemini models on cost, speed, and quality for different types of work.
• Investigate failed turns end to end and turn findings into fixes and guardrails.

Requirements

• 3+ years of software engineering with at least one LLM-powered feature shipped to real users.
• Hands‑on experience with tool or function calling and structured outputs.
• An evaluation mindset: measuring before believing an improvement.
• Solid backend fundamentals including APIs, queues, retries, and observability.
• Good judgment about when an agent should act and when it should ask.

Nice to have

• Experience with token budgeting and cost controls in production.
• Background in sales, marketing, or CRM automation.
• Contributions to open-source agent or evaluation tooling.

Benefits and work setup

• Meaningful equity with a long exercise window.
• Remote-first across Europe or Africa, async by default with a few overlap hours.
• Private health cover (or a stipend where appropriate), $1,500 yearly learning budget, $1,000 home-office stipend plus a laptop of choice, and flexible time off with a 25-day minimum.