5 key areas for tech
leaders to watch in 2020
By O’Reilly's Roger Magoulas and
February 21, 2020
Our annual analysis of the O’Reilly
online learning platform reveals Python’s continued dominance and
important shifts in infrastructure, AI/ML, cloud, and security.
O’Reilly online learning
contains information about the trends, topics, and issues tech
leaders need to watch and explore. It’s also the data source for
our annual usage study, which examines the most-used topics and
the top search terms. 
combination of usage and search affords a contextual view that
encompasses not only the tools, techniques, and technologies
that members are actively using, but also the areas they’re
gathering information about.
signals from usage on the O’Reilly online learning platform
Python is preeminent.
It’s the single most popular programming language on
O’Reilly, and it accounts for 10% of all usage.
This year’s growth in Python usage was buoyed by its
increasing popularity among data scientists and machine
learning (ML) and artificial intelligence (AI)
infrastructure, and operations are each changing rapidly.
The shift to cloud native design is transforming both
software architecture and infrastructure and operations.
operations is trending up,
while DevOps is trending down. Coincidence? Probably
not, but only time will tell.
ML + AI are up, but passions
have cooled. Up until 2017, the ML+AI topic had
been amongst the fastest growing topics on the platform.
Growth is still strong for such a large topic, but usage
slowed in 2018 (+13%) and cooled significantly in 2019,
growing by just 7%. Within the data topic, however,
ML+AI has gone from 22% of all usage to 26%.
Still cloud-y, but with a
possibility of migration. Strong usage in cloud
platforms (+16%) accounted for most cloud-specific
growth. But sustained interest in cloud migrations—usage
was up almost 10% in 2019, on top of 30% in 2018—gets at
another important emerging trend.
Security is surging.
Aggregate security usage spiked 26% last year, driven by
increased usage for two security certifications: CompTIA
Security (+50%) and CompTIA CySA+ (+59%). There’s plenty
of security risks for business executives, sysadmins,
DBAs, developers, etc., to be wary of.
Figure 1 (above). Normalized
search frequency of top terms on the O’Reilly online
learning platform in 2019 (left) and the rate of change for
each term (right).
Figure 2. High-level topics on
the O’Reilly online learning platform with the most usage in
2019 (left) and the rate of change for each topic (right).
Python is preeminent
In 2019, as in
2018, Python was the most popular language on O’Reilly online
learning. Python-related usage grew at a solid 6% pace in 2019,
a slight drop from 2018 (+10%). After several years of steady
climbing—and after outstripping Java in 2017—Python-related
interactions now comprise almost 10% of all usage.
But Python is a
special case: this year, more than in year’s past, its growth
was buoyed by interest in ML. Usage specific to Python as a
programming language grew by just 4% in 2019; by contrast, usage
that had to do with Python and ML—be it in the context of AI,
deep learning, and natural language processing, or in
combination with any of several popular ML/AI frameworks—grew by
9%. The laggard use case was Python-based web development
frameworks, which grew by just 3% in usage, year over year.
Figure 3 (above). Normalized
search frequency of top programming languages on the
O’Reilly online learning platform in 2019 (left) and the
rate of change for each language (right).
Figure 4. Programming languages
on the O’Reilly online learning platform with the most usage
in 2019 (left) and the rate of change for each language
This is consistent
what we’ve observed elsewhere :
Python has acquired new relevance amid strong interest in AI and
ML. Along with
Python is one of the most-used languages for data analysis. From
pre-built libraries for linear or logistic regressions, decision
trees, naïve Bayes, k-means, gradient-boosting, etc., there’s a
Python library for virtually anything a developer or data
scientist might need to do. (Python libraries are no less useful
for manipulating or engineering data, too.)
R itself continues to decline .
R-related usage on O’Reilly online learning fell by 8% between
2017-18 and by 6%, year-over-year, in 2019. It’s likely that
R—much like Scala (-33% in usage in 2018-19; -19% in usage
2017-18)—is a casualty of Python. True, it might seem difficult
to reconcile R’s decline with strong interest in AI and ML, but
consider two factors: first, ML and statistics are not the same
thing, and, second, R is not, primarily, a developer-oriented
language. R was designed for use in academic, scientific, and,
more recently, commercial use cases. As statistics and related
techniques become more important in software development, more
programmers are encountering stats in programming classes. In
this context, they’re more likely to use Python than R.
Interest in some
languages seems to be trending up, and interest in others, down.
Exhibit A: Java-related usage dropped by a noteworthy 13%
between 2018 and 2019. Is this the harbinger of a trend? Not
necessarily: Java-related searches increased by 5% between 2017
appears to be in decline. True, theirs is only a conceptual
really does seem to be waning :
JS-related usage dropped on O’Reilly online learning by 4%
between 2017-2018 and by 7% between 2018-19. It’s possible that
microservices architecture is hastening the move to other
languages (such as Go, Rust, and Python) for web properties.
popularity (+4% in usage) as Angular (-12% in usage) slipped
between 2018 and 2019. Vue.js—a competitor to both React and
Angular—settled down to steady growth (+8% in usage) in 2018-19,
after almost doubling in usage (+97%) between 2017-18.
possible trend-in-the-making is that of a slowing Go,
which—following several years of rapid growth in usage
(including +14% from 2017 to 2018)—cooled down last year, with
usage growing by a mere 2%. But Go is now the sixth most-used
programming language, trailing only Python, Java, .NET, and C++.
Drop .NET from the tally on methodological grounds  ,
and Go cracks the top five.
Trends in software
architecture, infrastructure, and operations
Cloud native design
is a new way of thinking about software and architecture. But
the shift to cloud native has implications not only for
infrastructure and operations ,
too. It exploits new design patterns (microservices) and adapts
existing techniques (service orchestration) with the goal of
achieving cloud-like elasticity and resilience in all
environments, cloud or on-premises. O’Reilly Radar uses the term
Architecture ” to describe this
It’s against this
backdrop that what’s happening in both software architecture and
infrastructure and ops must be understood. In the generic
software architecture topic, usage in the containers topic
increased in our 2019 analysis, growing by 17%. This was just a
fraction of its 2018 growth rate (+56% in usage), but impressive
nonetheless. Kubernetes has emerged as the de facto
solution for orchestrating services and microservices in cloud
native design patterns. Usage in Kubernetes surged by 211% in
2018—and grew at a 40% clip in 2019. Kubernetes’ parent topic,
container orchestrators, also posted strong usage growth: 151%
in 2018, 36% this year—almost all due to interest in Kubernetes
Figure 5. Software architecture
topics on the O’Reilly online learning platform with the
most usage in 2019 (left) and the rate of change for each
This also helps
explain increased usage in the microservices topic, which grew
at a 22% clip in 2019. True, you don’t necessarily need
microservices to “do” cloud native design; at this point,
however, it’s difficult to disentangle the two. Most cloud
native design patterns involve microservices.
These trends are
also implicated in the rise of infrastructure and ops, which
reflects both the limitations of DevOps and the challenges posed
by the shift to cloud native design. Infrastructure and ops
usage was the fastest growing sub-topic under the generic
systems administration topic. Surging interest in infrastructure
and ops also explains declining usage in the configuration
management (CM) and DevOps topic areas. The most popular CM
tools are DevOps focused, and, like DevOps itself, they’re
declining: usage in the CM topic dropped significantly (-18%) in
2019, as did virtually all CM tools. Ansible was least affected
(-4% in usage), but Jenkins, Puppet, Chef, and Salt each dropped
off by 25% or more in usage. It can’t be a coincidence that
DevOps usage declined again (-5%) in 2019, following a 20%
decline in 2018.
Figure 6. Infrastructure and
operations topics on the O’Reilly online learning platform
with the most usage in 2019 (left) and the rate of change
for each topic (right).
The emergence of
infrastructure and ops suggests that organizations might be
having trouble scaling DevOps. DevOps aims to produce
programmers who can work competently in each of the layers in a
system “ stack .”
In practice, however, developers tend to be less committed to
DevOps’ operations component, a fact that gave birth to the idea
site reliability engineering
(SRE). Even if the “full stack” developer isn’t a unicorn, she
certainly isn’t commonplace. Organizations see infrastructure
and ops as a pragmatic, ops-focused complement that picks up
precisely where DevOps tends to fail.
A drill-down into
data, AI, and ML topics
The results for
data-related topics are both predictable and—there’s no other
way to put it—confusing. Starting with data engineering, the
backbone of all data work (the category includes titles covering
data management, i.e., relational databases, Spark, Hadoop, SQL,
NoSQL, etc.). In aggregate, data engineering usage declined 8%
in 2019. This follows a 3% drop in 2018. Both years were driven
by declining usage of data management titles.
When we look more
specifically at data engineering topics, excluding data
management, we see a small share, but solid growth in usage, up
7% in 2018 and 15% in 2019 (see Figure 7).
Within the broad
“data” topic, data engineering (including data management)
continues as the topic with the most share, garnering about
one-twelfth of all usage on the platform. This is almost double
the usage share of the data science topic, which recorded an
uptick in usage (+5%) in 2019, following a decline (-2%) in
in ML and AI keeps growing, albeit at a diminished rate. To wit:
the combined ML/AI topic was up 7% in usage in 2019, about half
its growth (+13%) in 2018.
Figure 7. Data topics on the
O’Reilly online learning platform with the most usage in
2019 (left) and the rate of change for each topic (right).
strength of ML/AI might be less evident in data-specific topics
than in other topic areas, such as programming languages, where
growing Python usage is—to a large degree—being driven by that
language’s usefulness for and applicability to ML. But
ML/AI-related topics such as natural language processing (NLP,
+22% in 2019) and neural networks (+17%) recorded strong growth
in usage, too.
engineering as a task certainly isn’t in decline.
Interest in data engineering probably isn’t declining, either.
If anything, data engineering as a practice area is
being subsumed by both data science and ML/AI  .
We know from
that data scientists, ML and AI engineers, etc., spend an
outsized proportion of their time discovering, preparing, and
engineering data for their work. We’ve seen that popular tools
and frameworks usually incorporate data engineering
capabilities, either in the form of automated/guided
self-service features or (in the case of Jupyter and other
notebooks) an ability to build and orchestrate data engineering
pipelines that invoke Python, R (via Python), etc., libraries to
run data engineering jobs concurrently or, if possible, in
correspond with old-school data engineering—e.g., “relational
database,” “Oracle database solutions,” “Hive,” “database
administration,” “data models,” “Spark”—declined in usage,
year-over-year, in 2019. Some of this decline was a function of
larger, market-driven factors. We know from our research that
Hadoop and its ecosystem of related projects (such as Hive)
are in the midst of a protracted,
years-long decline . This decline is
borne out in our usage numbers: Hadoop (-34%), Hive (also -34%),
and even Spark (-21%) were all down, significantly,
We discuss likely
reasons for this decline in more detail
in our analysis of O’Reilly
Strata Conference speaker proposals .
cloud-related concepts and terms continues to increase on
O’Reilly online learning, albeit at a slower rate. Cloud-related
usage surged by 35% between 2017 and 2018; it grew at less than
half that rate (17%) between 2018 and 2019. This slowdown
suggests that cloud as a category has achieved such a large
share that (mathematically) any additional growth must
occur at a slower rate. In cloud’s case, while growth is slower,
it’s still strong.
Figure 8. Cloud topics on the
O’Reilly online learning platform with the most usage in
2019 (left) and the rate of change for each topic (right).
Interest in cloud
service provider platforms mirrors that of the industry as a
whole: Amazon and AWS-related usage increased by 14%,
year-over-year; Azure usage, on the other hand, grew at a
speedier 29% clip, while Google Compute Platform (GCP) surged by
39%. Amazon controls a little less than half ( per
Gartner’s 2018 numbers ) of the overall
market for cloud infrastructure-as-a-service (IaaS) offerings.
It, too, has reached the point at which rapid growth becomes
mathematically prohibitive. Both Azure and GCP are growing much
faster than AWS, but they’re also much smaller: Azure notched
nearly 61% growth in 2018 (per Gartner), good for more than 15%
of the IaaS market; GCP, at around 60% growth, accounts for 4%
of IaaS share.
cloud-specific interest in microservices and
grew significantly last year on O’Reilly. Microservices-related
usage was up 22%, year over year, following a decline in 2018.
Kubernetes usage was up by 38%, year over year, following a
period of explosive growth (+190%) from 2017 to 2018. Both
trends mirror what we’re seeing via user surveys and other
research efforts :
namely, that microservices has emerged as an important component
of cloud native design and development.
The bigger takeaway
is that the essential tendency of modern software
architecture—namely, the priority it gives to loose coupling in
emphasizing abstraction, isolation, and atomicity—is eliding the
boundaries between what we think of as “cloud” versus
“on-premises” contexts. We see this via sustained interest in
microservices and Kubernetes in both on-premises and
This is the logic
of cloud native design: specific deployment contexts will still
matter, of course—which features or constraints do developers
need to take into account when they’re developing for AWS? For
Azure? For GCP? But the clear boundaries that used to demarcate
the public cloud from the private cloud are starting to
disappear, just as those that distinguish on-premises private
clouds from conventional on-premises systems are falling away,
(+26%) grew significantly in 2019 (see Figure 2). Some of this
was driven by increased usage in the CompTIA Security+ (50%) and
CompTIA Cyber Security Analyst (CySA+, 59%) topics.
Security+ is an
entry-level security certification, so its growth could be
attributed to increased usage by sysadmins, DBAs, software
developers, and other non-specialists. Whether it’s to flesh out
their full-stack bona ides, address new job (or
requirements, or simply to make themselves more marketable,
Security+ is a pretty straightforward certification process:
pass the exam and you’re certified. CySA+, on the other hand, is
relatively new. This
could explain the explosion of
CySA+-related usage in 2018 (+128%), as well as last year’s
strong growth. Unlike
the CISSP and other popular
certifications , CySA+ recommends, but
doesn’t require, real-world experience. Like Security+, it’s
another certification sys admins, DBAs, developers, and others
can pick up to burnish their bonafides.
weren’t the only thing driving security-related usage on
O’Reilly in 2019. A rash of vulnerabilities and potential
exploits, too, had some impact. If 2018 (+5% growth in security
usage; +22% growth in search) gave us
2019 gave us sobering information about
of Meltdown and,
especially, Spectre .
For 2019, security-specific usage (+26%) and search (+25%)
increased accordingly. System and database administrators, CSAs,
CISSPs, and others were keen to acquire expert, detailed
information specific to patching and hardening their vulnerable
systems to protect against
no less than 13 different Spectre
and 14 different Meltdown variants , as
well as to mitigate
the potentially huge performance
impacts associated with these patches .
Developers and software architects had questions about
rewriting, refactoring, or optimizing their code to address
these same concerns. Against this backdrop, the spike in
security-related usage makes sense.
There was a great
deal going on with respect to information security and data
privacy, too. After all, not only was 2019 the first full year
for which the EU’s omnibus GDPR regime was binding, but—as of
January 1, 2019—updates to Canada’s GDPR-like PIPEDA regime
officially kicked in, too. The sweeping
California Consumer Privacy Act
(CCPA), which has been called California’s GDPR, went into
effect on January 1, 2020.
Taken together, an
analysis of these trends seems to support a glass-half-full
assessment of the state of security today. If the sustained
growth in security usage on O’Reilly is a reliable indicator,
it’s possible that security may, finally, be getting
the attention it deserves in an increasingly digital world. It’s
possible that organizations have accepted that the
financial and reputational penalties entailed by a data breach
or high-profile hack are just too costly to risk, and that money
spent on information security is, on balance, money well spent.
The same analysis
also lends itself to a glass-half-empty assessment, however:
namely, that security spending is cyclical; that a confluence of
circumstances has helped to boost security spending; and
that—let’s be honest— organizations
tend to bounce back from high-profile security incidents .
Only time (or future installments of this survey) will tell.
It’s hard to
imagine that the hottest trends of 2019 won’t be reprising their
roles, in more or less the same pecking order, in next year’s
analysis. Programming languages come into and go out of vogue,
but Python appears poised to keep growing at a steady rate
because it’s at once protean, adaptable, and easy to use. We see
this in the widespread use of Python in ML and AI, where it has
supplanted R as the lingua franca of data engineering
The same is true of
ML and AI. Even if (as some naysayers warn) the next
is nigh upon us ,
it’s hard to imagine interest in ML and AI petering out anytime
soon. The same could be said about trends in software
architecture and, especially, infrastructure and operations.
They’re each sites of ceaseless innovation. Their practitioners
will be hard-pressed to keep up with what’s happening.
It’s helpful to
think of what’s hot and what’s not in terms of a modified “ Overton
Window .” The Overton Window
circumscribes the human cognitive bandwidth that’s
available in a certain place at a certain time. No combination
of policies—or issues, or trends—can exceed more than 100% of
available bandwidth. This is true, to a degree, of the activity
on O’Reilly online learning, too. A decline in usage doesn’t
have to correlate with a decline in use (or usefulness) in
practice. It’s just being crowded out by other, emergent trends.
underscores why the decline in a bellwether topic such as
no longer sites of rapid and sustained innovation, they are
likewise important for day-to-day use cases, especially with
respect to general-purpose information-gathering or more
important than it used to be; after all, React, Angular, and
Vue.js are sites of development and innovation, and all three
DevOps and change management—in a different way. We’re
appropriating them differently.
difference that Radar aims to capture. Not the change that’s
obvious for everyone to see, but the coalescence of change
itself as it’s happening.
This article is based on non-personally-identifiable information
about the top search terms and most-used topics on O’Reilly
online learning. We compared aggregated data for the last three
years; a full year of data for 2017 and 2018, and through the
end of October for 2019.
A reasonable enough decision. .NET isn’t so much a language as a
software framework: i.e., a superset of C# and a few related
languages, including Visual Basic .NET, J#, and, C++/CLI, the
latter of which is a .NET-specific implementation of C++.
ML and AI aren’t in any sense the same thing, either. We combine
them here for simplicity’s sake.