3 July 2026 · databerlin.net
From 'Big Data' to 'LLM': 10 Years of Buzzwords in Berlin's Data Job Postings
How the magic words in data job postings have changed over the last decade — and what it tells us about where we're going
Cover: Photo by Ayman Mahmoud / Pexels. A candidate unimpressed with the job market.
There’s a moment, in every technology cycle, when a word stops being a technical term and becomes a cultural icon. You see it in newspaper headlines, in conversations at the bar, in LinkedIn posts. And you see it — perhaps more than anywhere else — in job postings.
“Data is the new oil.” Clive Humby said it in 20061, and for years it was the phrase stamped on every corporate slide deck. The problem is that oil, before it’s refined, is crude — valuable only after processing. Data is the same. But the phrase worked: it gave data an aura of strategic resource, of a gold rush.
“Data Scientist: the sexiest job of the 21st century.” Harvard Business Review wrote it in 2012, and for a decade it was the title everyone wanted. The reality was less glamorous: mostly cleaning datasets and writing SQL queries. But the phrase did its job: it filled universities with aspiring data scientists.
These phrases aren’t just marketing. They’re the thermometer of a larger phenomenon: the hype cycle. Gartner has called it that since 1995: every new technology goes from a peak of inflated expectations to a trough of disillusionment, before settling into a productivity plateau. Job postings are the most faithful reflection of this cycle — because companies put in them the words they think will attract talent. And the words that attract talent are the ones everyone is saying.
The result? A perfect stratigraphy. Each wave of hype leaves a layer of buzzwords in job postings. Some words stay (Python, SQL, Kubernetes). Others — “Big Data”, “Hadoop”, “Web 2.0” — disappear without a trace, as if they were never the center of the universe.
We analyzed 674 active job listings on databerlin.net as of July 3, 2026. We searched for the words that have dominated the last ten years — from “Big Data” to “LLM”, from “Data Science” to “Agentic AI” — and measured what’s left, what’s dead, and what’s emerging.
Spoiler: “Big Data” doesn’t appear in a single title. Zero. Meanwhile, “AI” is in 75.4% of descriptions. But the interesting story isn’t in today’s numbers — it’s in how we got here.
Let’s take a step back.
The Big Data Era
“Big Data” was everywhere. Hadoop, Spark, NoSQL — these were the words that opened doors. Every company wanted its “Big Data Architect” or “Big Data Engineer”.
It was the period when:
- “Big Data” dominated job titles
- Hadoop was considered indispensable
- Data Scientist was called “the sexiest job of the 21st century” (Harvard Business Review)
- Companies hired data scientists without knowing what to do with them
Then the word began to disappear — not because data had shrunk, but because it had become a commodity. Nobody posts a “Big Data Engineer” ad anymore because everyone works with big data. The word lost its distinguishing power.
The Rise of Machine Learning
The late 2010s marked a turning point. “Machine Learning” and “ML” began to climb. They were no longer just researcher stuff — they became required skills in regular job postings.
In Berlin, this is the period when:
- TensorFlow vs PyTorch was the main technology debate
- Computer Vision and NLP became sought-after specializations
- The first ML Engineer roles emerged, distinct from Data Scientist
- Companies realized a data scientist isn’t automatically an ML engineer
It’s also the period when “Data Science” peaks. Then, like Big Data, it slowly begins to normalize.
The Remote Shift
Remote work becomes normal, and job postings double. During this period:
- ML Engineer becomes a stable, recognized title
- MLOps emerges as a discipline (and a buzzword)
- Kubernetes, Airflow, dbt become the infrastructure standard
- The “Data Scientist” title begins to fragment into specializations
This is also when “Big Data” dies for good as a buzzword. Searching for “Big Data” in a Berlin job title became like searching for “Web 2.0” — archaeology.
The LLM Explosion
Late 2022: ChatGPT launches. Then, over the next year, AI coding assistants — Copilot, Cursor, Claude — begin to reshape what it means to write software. The job market follows.
In 12 months:
- LLM goes from an insider term to a must-have on CVs
- Prompt Engineering becomes a real job title
- RAG (Retrieval-Augmented Generation) enters the technical vocabulary
- LangChain becomes the most talked-about library of the moment
- GPT, Transformer, Fine-tuning explode in job postings
In Berlin, AI-native startups begin hiring at breakneck speed. Traditional companies realize they need people who can work with AI, not just talk about it.
The Age of AI (and the Data to Prove It)
And here we are. We analyzed 674 active listings on databerlin.net. Here’s what the numbers say:
Words in job titles
Click any label to explore that skill on the job board.
Words in job descriptions
Click any label to explore that skill on the job board.
Most requested skills (from our structured data)
Click any label to explore that skill on the job board.
What This Story Tells Us
Four things stand out:
1. “AI” is everywhere, but it’s already a commodity
75.4% of listings mention AI. It’s become like saying “we use computers.” The term is so widespread it’s lost its filtering power. When everyone says AI, the real question becomes: what kind of AI?
2. LLM is the new “Big Data”
LLM appears in 23.1% of descriptions but only 0.6% of titles. That’s exactly where “Big Data” was at its peak: ubiquitous in requirements, but not yet distinctive enough to put in the title. In two years, it will probably disappear from descriptions too — because it’ll be taken for granted.
3. Agentic AI is the newcomer
Agentic AI has climbed to fifth place (189 listings) — and it’s the fastest-growing skill in our dataset. It’s the buzzword of 2026. In a year, it’ll probably be where LLM is now in the rankings. The cycle repeats.
4. The “boring” skills always win
Python, SQL, AWS, Kubernetes, Airflow, dbt — these are the foundations. They don’t make headlines, they’re not sexy, but they’re in almost every listing. While buzzwords come and go, these stay.
The Active vs. Expired Signal
One of the most revealing ways to read this data is to compare active listings against expired ones. Our database tracks 1,325 total jobs (674 active + 651 expired). The difference between the two groups shows which buzzwords are gaining momentum and which are fading.
Rising: AI and LLM
| Buzzword | Expired | Active | Change |
|---|---|---|---|
| AI in descriptions | 68.9% | 75.4% | +6.5pp |
| LLM in descriptions | 19.5% | 23.1% | +3.6pp |
| LLM as structured skill | 40.8% | 47.2% | +6.4pp |
AI and LLM are not just present — they’re growing. Newer listings are significantly more likely to mention them than older, expired ones. The trend is accelerating.
Falling: GenAI and Spark
| Buzzword | Expired | Active | Change |
|---|---|---|---|
| GenAI in titles | 1.5% | 0.3% | −1.2pp |
| Spark in descriptions | 9.3% | 8.5% | −0.8pp |
| SQL in descriptions | 35.2% | 31.0% | −4.2pp |
GenAI is the most interesting case. It appeared in 1.5% of expired job titles but only 0.3% of active ones. The term peaked in 2024–2025 and is already on its way out — replaced by the simpler, broader “AI.” It’s following the same trajectory as “Big Data,” just in fast-forward.
Spark is in a slow decline. Once the darling of big data processing, it’s being displaced by newer tools and managed services. The 0.8pp drop between expired and active listings suggests the trend will continue.
SQL appears to be falling, but that’s misleading. SQL isn’t disappearing — it’s becoming implicit. Fewer job descriptions bother listing it because it’s assumed. The structured skill data tells a different story: SQL is still the third most requested skill overall.
Stable: the infrastructure backbone
Python, Kubernetes, Airflow, dbt — these show minimal difference between expired and active. They’re not hype-driven. They’re the concrete floor of the job market — nobody notices them until they’re gone.
The Pattern: Every Hype Cycle Leaves a Residue
Every hype cycle is like a glacier: when it retreats, it leaves deposits. Big Data left Spark and Hadoop. ML left PyTorch and scikit-learn. AI is leaving LLM, RAG, and Agentic AI. The next wave will leave something else.
The lesson for job seekers: learn the buzzwords, but invest in the foundations.
- The buzzword gets you past ATS screening
- The foundations get you through the technical interview
- The “boring” skills (SQL, dbt, Airflow) keep you employed over time
What to Expect in 2027
If the pattern holds:
- “AI” in titles will start to decline — it’ll become implicit
- “Agentic” will explode as the main buzzword
- LLM will follow the same arc as “Big Data”: from must-have to commodity to invisible
- Something new will take the hype slot
- Python, SQL, and infrastructure skills will still be there, unchanged
Data in this article comes from 674 active listings on databerlin.net as of July 3, 2026. The historical timeline is based on industry trends and cross-referenced public sources — approximate, not precise.
Want to see the live data? Explore current openings at databerlin.net.