A few examples of how AI is currently changing our work.
Changes in the "Python Developer" profile.
Python has dominated the backend (23% of job offers in Poland), overtaking Java. But searching simply for a "Python Developer" today is a guarantee of a flood of useless applications. The market has split in two:
- Classic Backend (Django/FastAPI): Stable rates, high availability of candidates. Recruitment is fast here.
- AI/LLM Engineer (RAG, finetuning, agentic integration, LangChain): Candidates with these skills cost 35-50% more than traditional backend developers.
Recommendations
- Change the brief with the Hiring Manager: Determine whether you are building another CRUD system or implementing language models. If the latter, an ad stating "Python Developer" will attract people after 3-week AI prompting bootcamps.
- Resume verification: Ignore entries like "AI knowledge". Look for real production deployments in portfolios using vector databases (e.g., Pinecone, Qdrant) and API cost optimization (token management).
The "death" of the junior (5% of offers) and advancement thanks to AI
Statistics from No Fluff Jobs and Just Join IT are clear: 95% of job offers in Poland are for Mid and Senior roles (with Seniors making up as much as 60%). There is almost no room for juniors (5%). But the reason is not just cost-cutting – tools like Cursor or GitHub Copilot have enabled a motivated junior with 1 year of experience to deliver tasks at the level of a former mid. Are Juniors the new Mids?
Recommendation
- Change the technical process: Traditional "take-home" recruitment tasks have lost their purpose. Candidates solve them using LLMs in 5 minutes. Shift to live-coding focused on debugging code generated by AI or analyzing architecture. Test logical thinking, not syntax knowledge.
- "Shadow talent" instead of expensive seniors: Instead of paying 30,000 PLN B2B for a senior who will leave at the first opportunity, implement mentoring programs. A junior supported by AI and a good Senior reaches operational independence in 3 months instead of a year.
Hybrid blackmail (41% vs 38.5%)
For the first time in the history of Polish IT, the hybrid model (41% of offers) has overtaken the fully remote model (38.5%). Boards are pushing Return-to-Office (RTO), believing in "organizational culture building." For a recruiter, it's a nightmare.
Recommendation
- Present the costs to the board: Enforcing a hybrid model (e.g., 3 days in the office) in Warsaw or Krakow shrinks your talent pool by 80-85% (cutting off all regional Poland). Time-to-Hire increases from 25 to an average of 58 days. Show the board simple math: office costs vs the cost of an unfilled position for a quarter.
- Asynchronous model: If the office is necessary, promote a "Smart Hybrid" model – e.g., one fixed week a month for collaborative work and planning, and the remaining 3 weeks 100% remote. This allows recruiting people from all over Poland.
Using Local AI Models in HR Processes
IT recruitment involves processing sensitive personal data (resumes) and code protected by NDAs. Sending this information to cloud APIs (OpenAI, Anthropic) carries a high risk of GDPR violation. The solution is hosting open-weight models locally.
Hardware Solutions Analysis and Recommended Models:
Below is a list of available open-weight models dedicated to programming and analytical tasks in recruitment, sorted by RAM/VRAM requirements:
| Target Platform | Recommended Models | Required RAM/VRAM | Application in HR |
|---|---|---|---|
| Laptop / Workstation (Recruiter) | Qwen-2.5-Coder-7B Llama-3.1-8B |
8 GB – 16 GB | Screening resume text, generating initial interview questions based on the candidate's profile, quick analysis of simple test tasks (run e.g. via Ollama). |
| Developer Desktop (Hiring Manager / Tech Lead) | Qwen-2.5-Coder-32B Codestral-22B DeepSeek-Coder-V2-Lite (16B) |
32 GB – 64 GB | Advanced code analysis, finding subtle logical bugs, evaluating unit test quality, and verifying architectural compliance. |
| Company Server (On-premise / Private Cloud) | LongCat-2.0 (1.6T MoE / 48B active) DeepSeek-Coder-V2 (236B MoE) GLM-5.2 |
128 GB – 256 GB+ | Secure sandbox. With a context window of 1 million tokens (LongCat-2.0), the server can analyze the entire codebase to create a highly specific and unique debugging task (bug hunting) for the candidate. |
Google Autonomous Information Agents for Active Sourcing
Traditional direct sourcing on LinkedIn is becoming less effective. A new approach is utilizing Information Agents (e.g., Google Vertex AI Agent Builder / Agent Space) that can autonomously monitor the internet and actively find candidates based on their actual technical contributions:
- Competency-based sourcing: The agent continuously monitors public GitHub repositories, StackOverflow, and niche tech forums, analyzing pull requests and commit messages.
- Personalized contact: Once the agent identifies a developer solving complex issues in the desired technology, it drafts a personalized message referencing a specific line of code by the candidate.
New AI Models and Future Market Outlook (July 2026)
The AI model market is moving at a breakneck pace. At the end of June 2026, xAI released Grok 4.5 into private beta, and Google is preparing to show its delayed Gemini 3.5 Pro model. Meanwhile, the launch of the Chinese LongCat-2.0 model (June 29, 2026), trained entirely on domestic ASIC chips without Nvidia processors, demonstrated that high-performing, open coding models (scoring 59.5 on SWE-bench Pro) are becoming easily accessible and cheap to run on-premise.
Bibliography and Sources:
- Just Join IT & No Fluff Jobs (2026) – Annual Report: IT Job Market in Poland. Salaries, technologies, and candidate expectations (Q1 2026 edition).
- Meituan AI Lab (June 2026) – LongCat-2.0: Trillion-Parameter Mixture-of-Experts Model for Agentic Coding.
- Google Cloud (2025/2026) – Vertex AI Agent Builder: Building Enterprise-grade Autonomous Sourcing Workflows.
- SWE-bench Community (2026) – Open-source Coding Models Leaderboard & Benchmarks.
- IDC & ITwiz (2026) – Demand for Cloud, AI, and Cybersecurity Competencies in Polish Enterprises.