The Human Side of ML Engineering: Analyzing the Collaboration and Communication Skills in ML Job Postings

Aidin Azamnouri, Phuong Thao Le, Justus Bogner, Stefan WagnerarXiv 2026 · arXiv

TL;DR

This paper presents a large-scale analysis of 1,000 job postings to quantify the demand for collaboration and communication skills in Machine Learning (ML) engineering. By using a hybrid human-LLM annotation pipeline, the researchers found that these "soft skills" are nearly universal requirements (appearing in ~92% of ML ads), though they are often expressed implicitly and oriented toward outward-facing stakeholder interaction rather than internal team practices.

What Problem Does This Paper Address?

While the ML community often discusses the "technical debt" and algorithmic challenges of the field, there is a growing realization that the biggest bottlenecks are often human-centric—coordinating between data scientists, engineers, and domain experts. However, we haven't had a clear, data-driven picture of what employers actually ask for.

Previous attempts to analyze job postings relied on simple keyword searches (e.g., looking for the word "communication"). The authors argue this is too blunt: a recruiter might ask for someone who can "explain complex results to non-technical partners" without ever using the word "communication." This paper aims to bridge that gap by detecting both explicit and implicit requirements and comparing ML roles to traditional Software Engineering (SE) roles to see if the "human side" of ML is truly unique.

Methods

The authors developed a four-stage empirical pipeline:

  1. Data Collection: They used a semi-automated pipeline via SerpApi to scrape 1,000 unique job postings (500 ML, 500 SE) across five continents, ensuring a mix of languages and regions.
  2. Preprocessing: Non-English postings were translated using a multi-model pipeline (Llama-3.1 and SauerkrautLM) with BERTScore-based quality routing to ensure semantic fidelity.
  3. Hybrid Annotation: This is the core innovation. Two human annotators worked alongside an LLM-as-a-Judge (Llama-3.1-8B for collaboration and DeepSeek-R1 for communication). The LLM acted as a third independent rater, helping to surface implicit mentions that humans might overlook and providing reasoning for its labels.
  4. Facet Coding: Beyond binary "yes/no" labels, the team used a deterministic rule set to categorize requirements into "facets," such as cross-functional collaboration, mentoring, or audience-facing interaction.

Datasets

The study utilized a newly collected corpus of 1,000 job postings:

  • 500 ML postings and 500 SE postings.
  • Geographic Diversity: 100 postings per domain from each of five continents (Africa, Asia, Europe, North America, and South America).
  • Languages: Included English, German, French, Japanese, Vietnamese, Portuguese, and Spanish.
  • Benchmarks: The hybrid ground truth was compared against a lexical n-gram baseline and a fine-tuned BERT classifier.

Key Insights and Ideas

  • Integration over Membership: In ML, collaboration isn't just about being a "team player" in a pod. It is predominantly about cross-functional integration—bridging the gap between data engineers, product managers, and domain experts.
  • Outward vs. Inward Orientation: While ML and SE roles demand soft skills at similar rates, their focus differs. SE is "inward-facing," focusing on team practices and agile workflows. ML is "outward-facing," emphasizing the need to present, explain, and interact with stakeholders who don't understand the underlying math.
  • The "Implicit" Gap: Keyword-based tools are failing us. Roughly 21-27% of soft skill requirements in ML are implicit. If you only search for keywords, you miss nearly a quarter of the interpersonal demand.
  • The Specificity Gap: Recruiters are loud but vague. While 72% of ML ads want communication skills, fewer than half specify who the candidate should talk to, and 42% don't specify the channel (written vs. verbal).
  • Role Trumps Region: Within the ML domain, the specific role (e.g., Solutions Architect vs. MLOps Engineer) creates a bigger difference in interpersonal requirements than geographic region or industry.

Key Results

  • Prevalence: Collaboration skills appeared in 88.6% of ML postings; communication in 72.2%.
  • Methodological Superiority: The LLM-as-a-Judge approach achieved substantial agreement with humans (Cohen’s κ ≈ 0.70–0.81), significantly outperforming the n-gram baseline (κ ≈ 0.41–0.58) and a fine-tuned BERT classifier.
  • Regional Variation: North America and Africa showed the highest explicit demand for these skills, while South America and Asia were more likely to leave these expectations implicit.
  • Seniority U-Shape: Demand for these skills is high for entry-level roles (foundational), dips slightly at mid-level (technical focus), and intensifies again at senior/management levels (leadership).

Limitations and Future Work

  • Snapshot in Time: The data is a cross-sectional snapshot; it doesn't show how these requirements have evolved over the years.
  • Geographic Gaps: The study did not include data from Oceania.
  • Translation Risk: While the authors used high-quality translation and automated scoring, there is a residual risk that cultural nuances in job ads were lost during the English translation process.
  • Future Directions: The authors suggest extending this to other soft skills like "critical thinking" or "adaptability" and linking these requirements to actual labor-market outcomes like salary or time-to-hire.

Summary updated

Abstract

Studies of Machine Learning (ML) practice highlight the importance of combining technical expertise with interpersonal capabilities, particularly collaboration and communication. However, evidence on how ML job postings express these skills is scarce, and prior work largely relies on keyword-based analyses that overlook implicit expression. We analyze demand for these skills in ML job postings across regions, industries, and seniority levels, and compare it with software engineering (SE) roles. We collected 1,000 postings (500 ML, 500 SE) from five continents via a semi-automated pipeline. For annotation, we synthesized a taxonomy of collaboration and communication sub-categories from prior work and used an LLM as a third annotator alongside two humans, enabling detection of context-dependent and implicit skill references. Collaboration appeared in 88.6% of ML postings and communication in 72.2%, far above prior keyword-based estimates for AI roles. Roughly one in five collaboration and one in four communication requirements were expressed only implicitly. Decomposing these requirements into the taxonomy's sub-categories reveals what aggregate figures conceal. Collaboration is demanded predominantly as integration across role boundaries rather than team membership. Communication is mostly required in generic terms, with fewer than half of the postings requiring it naming whom the candidate must communicate with. ML and SE postings demand interpersonal competence at statistically indistinguishable levels but orient it differently, SE toward team practices and ML toward presentation and audience-facing interaction. Within ML, the advertised role separates interpersonal profiles more sharply than any other dimension, including the contrast with SE.
BibTeX
@misc{azamnouri2026human,
  author = {Aidin Azamnouri and Phuong Thao Le and Justus Bogner and Stefan Wagner},
  title = {{The Human Side of ML Engineering: Analyzing the Collaboration and Communication Skills in ML Job Postings}},
  year = {2026},
  eprint = {2611.18915},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.18915},
}
Figure