AI-Native Software Engineer

Canaria is looking for next-generation software engineers.
The way software gets built is changing. A new kind of engineer no longer types every line: they direct AI agents, check their work, and ship far more than one person could before. While most teams are still debating it, we already build this way every day. We are looking for engineers who want to be at the front of this shift, with unlimited AI and a real, large-scale problem to apply it to.
Company
Canaria is a technology startup providing high-quality, large-scale labor market data to the B2B market. Our mission is to power the next generation of solutions in recruitment, HR analytics, workforce planning, lead generation, healthcare, academic research, and finance by delivering the most comprehensive and accurate job market data available. Our advanced data mining, computing optimizations, and state-of-the-art NLP techniques let us process job market data at scale: over 1 billion unique job postings drawn from 200,000+ data sources, with 100+ fields per record. Our clients use this data to understand market trends, identify skill gaps, and connect people with the right opportunities.
About the Position
Full-time, remote from Turkey. Early-career candidates are welcome.
We collect job data with our own scraping platform from 200,000+ company career pages and job boards, and they change continuously. Every change can break something. We want LLMs doing most of the development that keeps up with them, so a small team can run a fleet this size. You will drive Claude Code, Codex, Cursor, Grok and whatever comes next to get us there. We give you the problem, unlimited AI, compute and direct access to the founder. You find the path.
How we work: small, fully remote, async team working in English, with evening syncs to overlap with the US. Nobody tracks hours; we look at what shipped.
What You Will Do
Onboard new sources at a pace a traditional engineer can't match: drive coding agents to work out a site's page or API shape, write the source spec, test it live, and ship a working spider.
Fix broken sources fast: when a site changes, use LLMs to diagnose why, write the fix, verify it live against the real site, and open a pull request.
Turn repeat work into reusable agent workflows (skills, prompts, project context) so the next source or fix goes faster than the last.
Make the agents prove their work: tests, field-level checks against the live site, and regression gates. A fix that quietly drops a field is worse than no fix.
Work Python-first in a codebase we already build with coding agents every day, with thousands of tests, strict typing and CI gates.
Find the next problem yourself. Nobody will hand you a ticket queue.
Who You Are (Required)
You have shipped real software that you built mainly by driving coding agents: something people used, or something running in production. You can walk us through it live, including what the agents got wrong.
Coding agents and LLMs (Claude Code, Codex, Cursor, Grok, Gemini or similar) are how you build every day. You plan before you prompt, split work across agents, and check what comes back.
You make agents prove their work: you write or demand tests that would fail if the code were wrong, and you have caught agent-written code that passed its own tests and was still wrong.
Strong Python and software fundamentals: Git, testing, SQL, APIs, Linux, Docker.
Professional English, written and spoken. Our team works in English across the US and EU.
No degree or years-of-experience requirement. Show us what you built.
Nice to Have
Running several coding agents in parallel on one task: splitting the work, having one review another, merging the results.
You maintain the context your coding agents work from (project rules files, skills, prompts) and can show a change that fixed a repeated mistake.
Web scraping or browser automation (HTTP clients, Playwright, anti-bot handling).
Knowing which model or agent fits which task, and when plain code beats an LLM.
Knowing when to let an agent run on its own and when to stop and check by hand.
You have built a small LLM-powered tool yourself. A bonus, not the job.
You can name a model or tool you adopted or dropped in the last six months, and why.
Hackathon wins, open-source contributions, or issues you filed that someone fixed.
AI automation outside coding (research, operations, writing workflows).
This Role May Not Be the Right Fit If
You prefer to start from a detailed spec rather than shape the problem yourself.
You would rather use agent output as it comes than review it.
You would like to ship without review. Here, every change goes through tests and review, whether a person or an agent wrote it.
You are looking to build LLM products or train models. Here, LLMs are tools for building software rather than the product itself.
Benefits
Unlimited AI: company-paid Claude Code, Codex, Cursor, Grok and other LLM tools, with no usage cap.
On-prem and cloud compute.
Work directly with the founder. No management layer: you own the outcome and choose the path.
High-end laptop.
Annual performance bonus.
Work with global teams (US, EU).
Founders and colleagues with FAANG and top-institution backgrounds.
Hiring Process
First call (15 mins): about the role & expectations, plus a 5-minute walkthrough of something you built with coding agents. Tell us what you did, what the agents did, and one problem in it you found that nobody asked you to look for.
Live session on a slice of our real codebase, any AI tools allowed (1 hour). We watch how you read, verify and debug, not how fast you type.
Deep dive with the founder on something you built (1 hour)
Only shortlisted candidates will be contacted.

Work arrangement
Yes