Junior AI Developer (Agentic Engineer)

The role, and who it is with

A publicly traded company is building a dedicated AI Unit and is hiring Junior AI Developers for it. Werkit is running this search: we select on evidence and connect Zambian talent with global work like this. You would work remotely on the company's team, report into its Technical Director, and be part of a small unit where your output is visible immediately. This is a ground-floor position on a unit that is being built now.

To be straight about expectations: this is work with the company, not employment by Werkit. Werkit finds you, tests you, and backs you through the process — the offer, the engagement, and the pay come from the company.

What "Agentic Engineer" means

An agentic engineer directs AI coding agents to produce most of the implementation, while remaining the engineer of record. The agent is your workforce. You are the architect, the reviewer, and the person responsible when something breaks in production.

Three things are yours and cannot be delegated:

• Architecture. What gets built, how it is divided into components, what the interfaces and data contracts are, how it fails and what happens when it does.
• Orchestration. How the work is broken into agent-sized tasks, what context and constraints each agent receives, what it is forbidden to touch, and how its output is verified and integrated.
• Verification. Proof that the system does what it is supposed to do and fails safely when it does not. Tests, manual checks, logs, and confirmation in the deployed environment.

What you will do

• Own features end to end: from a one-paragraph requirement to a deployed, monitored, documented system.
• Design the architecture of small and medium systems before a single line is generated: components, data flow, interfaces, failure handling.
• Decompose work into agent-executable tasks and run AI coding agents against them (Claude Code, Cursor, Copilot, etc.).
• Personally own the security-critical, data-critical and correctness-critical parts of every system, line by line, no exceptions.
• Review every line of agent-generated code as if a stranger wrote it and you are the last reviewer before production.
• Build internal automation: AI agents, MCP servers, tool pipelines, and integrations with the company's operational systems (Jira, Slack, Zoho, Google Workspace).
• Integrate LLM APIs into production with attention to cost, latency, failure modes and data privacy.
• Build evaluation harnesses: how do we know an AI feature is still correct after the model or the prompt changes?
• Deploy and operate what you build (Vercel, Cloudflare, AWS, Azure, Docker).
• Write technical specifications and knowledge-base entries.

The entry bar

• 1+ year of hands-on software development: commercial, freelance, or substantial personal projects with real users.
• You can write working code in an empty editor with no AI assistance. We will test this directly.
• Solid in at least one of TypeScript/JavaScript or Python, and able to read the other.
• Practical Git: branches, merge and rebase, conflict resolution, readable history, pull requests.
• SQL and relational data modelling: joins, indexes, transactions, and why a query is slow.
• HTTP and API fundamentals: methods, status codes, headers, authentication, pagination, idempotency.
• You have deployed something to the internet and kept it running.
• Daily use of an AI coding assistant, and a considered opinion about where it fails.
• Strong written English. This is a remote, written-first team.

Nice to have

• Public repositories where the history shows AI-assisted work you clearly directed.
• Experience building or consuming MCP servers.
• Next.js, Node.js, FastAPI, or .NET/C#.
• Hands-on AWS, Azure or GCP work, or a cloud certification.
• Familiarity with accessibility standards (WCAG): one of the company's products is in that space.
• Open-source contributions or published technical writing.

How the team works

• Agent-first, engineer-owned. Default to delegating implementation. Never delegate accountability.
• Understand everything you ship. "The AI wrote it" is not an explanation and not a defence.
• Verified beats fast; fast beats perfect. Shipping is the goal. Unverified shipping is a rollback waiting to happen.
• Simplicity wins. Agents over-engineer by default. Deleting generated code is part of the job.

Written-first team: Slack for day-to-day, Jira for task state, a knowledge base for anything durable. Daily written status. Escalate blockers early with what you already tried. Camera on for team calls.

The stack

• AI: Claude and Claude Code, Cursor, GitHub Copilot, OpenAI and open-source models; MCP servers and custom agents.
• Languages: TypeScript/JavaScript, Python; .NET/C# on some projects.
• Frameworks: Next.js, Node.js, React, FastAPI.
• Data: PostgreSQL, MSSQL, Redis, vector stores.
• Cloud and infrastructure: Azure, AWS, Oracle Cloud, Vercel, Cloudflare; Docker; GitHub Actions.
• Process: Jira, Confluence, Slack, Git/GitHub.

How we select

Werkit runs the selection end to end. The final decision and the offer come from the company.

1. Apply online. Profile links plus a few short written answers. Takes about 10 minutes.
2. Online assessment, about 1 hour. Code reading, a debug-and-fix exercise, and judgment calls about working with AI agents. AI tools are explicitly allowed in the sections where using them well is the skill being tested. Do it when it suits you; your progress saves as you go.
3. Fundamentals check, 60 minutes, live. You write and debug code in an empty editor, no AI. We are testing whether you are a developer, not whether you can prompt.
4. Agentic take-home, up to 8 hours, AI encouraged. A small system to design, generate, harden and deploy, with a short write-up of what you delegated and what you wrote yourself.
5. Architecture and review interview, 60 minutes. A walk through your take-home: why this structure, what happens when this fails, what would you delete.
6. Offer from the company, with a defined probation period.