Rising experience demands don't mean skills no longer matter

Revelio Labs' hiring data shows two lines moving in opposite directions: since early 2025, the years of experience required for technical roles have risen while the number of listed skills has fallen. The chart controls for seniority level, so the interesting signal is how requirements are shifting within the same kind of role.

For similar technical roles, experience demands rise while skill requirements fall

Starting in 2025, the two lines clearly separate; the red line reads skill count on the left axis, the blue line reads years of experience on the right axis. Job seniority levels are controlled. The intersection of the lines does not imply equal values.

Data and systems skills show the strongest gains in experience requirements

The comparison looks at the increase in years of experience associated with each skill mention since 2023; data organization and data research rank in the top two.

This "experience premium" reflects what job ads ask for, not what employers actually pay. Job postings only reveal how companies describe their ideal candidate; they can't alone prove who got hired or which capabilities drove better outcomes.

If companies list fewer skill keywords but demand more experience, what are they really trying to filter for?

When work methods haven't settled, judgment is harder to put on a list

Skill lists work best when the job is well defined: which tools to use, what tasks to complete, what standard to meet. But when the work itself is still taking shape, a list can't easily capture what's actually hard. A person might know every listed tool and still not know which problem to tackle first or how to judge whether a result is trustworthy.

Sternstein's argument is that AI is making old approaches less reliable while new ones haven't yet crystallized. In that situation, people who have worked through messy problems may be better equipped to make progress with incomplete information. The shifting experience requirements in data and systems roles are the evidence he points to.

Data work is a clear example. AI has made organizing and processing data more urgent, but what form it should take and how pipelines should be arranged are still moving targets; no best practice has settled. Sternstein's inference: when even the shape of the task is still being figured out, companies can't easily write an accurate skill list first. Someone who can judge as they go becomes more valuable. That helps explain why the roles with the biggest jumps in experience requirements cluster around data organization, data research, and systems architecture.

One key caveat applies here: experience only counts when it translates into fresh judgment. Repeating the same operation isn't the same as having weighed options, debugged failures, and checked results. Years on a resume are easy to count; the quality of someone's judgment only shows up in actual work.

For job seekers, a useful exercise is to frame project experience as a verifiable chain: what constraints you faced, why you chose a particular approach, what went wrong, and how you confirmed a fix actually worked. That doesn't guarantee an offer; it just turns "I have experience" into evidence someone else can evaluate.

Whether AI actually caused this hiring shift is still a hypothesis. The mix of roles companies post, the reach of remote hiring, and the natural aging of the workforce could all influence what the data shows. Once new working methods stabilize, employers may well translate these capabilities back into concrete skill requirements. The greater-than sign in the original headline describes a preference at this stage, not a permanent ranking of experience over skills.

Netflix's pressure: separate usage, share, and revenue

Apptopia's data shows US social media time still creeping up, driven mostly by users aged 17 to 25, while those aged 36 and over have been pulling back since October 2025. Meanwhile, since April 2026, social media's slice of total phone time has fallen about 3 percentage points. People are still scrolling, but they're giving more of their new screen time to other things.

US social app total time rises, but age groups diverge

Overall usage remains above 2023 levels, with the 17–25 age group still increasing; other age groups have recently plateaued or declined.

Share of phone time falls, year-over-year growth near zero

The dark green line reads the share of phone time on the right axis, the light green line reads the year-over-year growth rate on the left axis; growth near zero refers to the light green line.

Usage can rise while share falls. Suppose a category grows from 40 minutes a day to 45, but total phone time climbs from 100 to 125 minutes; its share drops from 40% to 36%. That hypothetical illustrates why absolute growth alone doesn't prove a competitive position is improving. Sternstein floats other apps, AI, and short-form video as likely destinations, leaning toward YouTube Shorts as a key rival. These are interpretations; the share change itself doesn't identify the cause.

Cross-usage is asymmetric: YouTube users open Netflix less

US mobile app index, set to 100 in January 2025; the index for Netflix users using YouTube is roughly stable, while the reverse direction clearly declines.

US weekly active users: Netflix declines, Spotify rises

Indexed to the week starting December 26, 2022; these are separate change indices and cannot be used to compare the total user bases of the two services.

US time spent: Netflix falls more sharply

Indexed to January 2023; Netflix drops to around 70 while Spotify remains above the base period.

Netflix faces a more direct squeeze: the index of YouTube users who also use Netflix is trending down, while the index of Netflix users who also use YouTube is roughly flat. Sensor Tower's US data also shows Netflix's weekly active users and time spent weakening over a long stretch. With their respective starting points in late 2022 and early 2023 set to 100, weekly actives sit in the 80s and time spent around 70, meaning viewing time has eroded more sharply. Short-form competition is one possible explanation; subscription pricing and other streaming rivals could also play a role.

The two cross-usage directions use different user groups as denominators, so they can't be swapped. An index simply records a starting value as 100 and tracks changes from there; it doesn't represent user counts. These charts show usage patterns in a US sample and can't be converted into global subscriber losses.

Even under usage pressure, Netflix's high-tier subscriber revenue estimate leads

As of June 2026, the lifetime revenue estimate for Netflix's premium monthly plan in the US is $1,427; this is a revenue estimate and does not subtract content or other costs.

At the same time, Owl & Co.'s estimate of US subscriber lifetime revenue puts Netflix's premium monthly plan at $1,427, ahead of the next highest, Disney+ Premium annual at $871. Higher payment tiers combined with longer subscriptions support a higher per-subscriber revenue value. Reduced usage and high lifetime value can therefore coexist, and lifetime revenue is not the same as profit after content and other costs.

Higher prices can also make users rethink whether a subscription is worth keeping. Sternstein links Netflix's push into live sports, clips, and short-form video to this attention pressure: the platform needs more reasons for people to open the app, to protect retention and the value of a paid subscription. Whether the new content actually wins back users' time remains to be seen.

The author draws an analogy to Blockbuster: Netflix upended the video rental giant about two decades ago, and now it may itself face a challenge from shifting viewing habits. The comparison raises the question of whether the business model can adapt, but the charts don't prove Netflix is headed for the same outcome. When evaluating, it's worth asking separately whether users are opening less, watching less, or paying less.

Security risk: the real question is how much response time remains

Epoch AI's tally of high-severity and critical vulnerability reports rises noticeably after spring 2026. Meanwhile, ZeroDayClock's data shows exploits happening earlier and earlier among vulnerabilities that have been exploited. Put together, defense teams aren't just facing more problems to fix; they have a tighter window to respond.

High and critical severity vulnerability reports increase sharply in 2026

The orange line represents high severity, and the red line represents critical severity; samples come from vendors listed below the chart, and reporting processes, labels, and frequencies vary.

Among exploited samples, the share already exploited by disclosure rises

In 2026, it stands at 86.7%; the denominator is only CVEs that were exploited, with the event occurring on or before the public disclosure date.

The 86.7% figure in the ZeroDayClock chart refers to the share of exploited vulnerabilities, as of the time of the chart's data cutoff in 2026, that were already being exploited on or before the day of disclosure. The comparison points are 2020, at 22.8%, and 2025, at 53.6%. The year-over-year trend shows exploitation clustering much more tightly around the disclosure moment; 2026 isn't over, and the sample size differs from a full year. The finding means disclosure doesn't necessarily give defenders a head start, but it shouldn't be read as "nearly all vulnerabilities are already under attack." The rise in reported vulnerabilities could also reflect better detection or changes in how disclosures are recorded.

The median exploitation threshold is extrapolated down to the minute level

Based on extrapolation from 2018–2024 trends; the 'one hour' and 'one minute' values for 2027 are labeled as projections.

For vulnerabilities that eventually get exploited, activity is more concentrated right after disclosure

The horizontal axis measures time from public disclosure, the vertical axis shows the share not yet exploited; the 2026 curve is lower, and the sample excludes vulnerabilities never exploited.

ZeroDayClock's data on the gap between disclosure and exploitation has already compressed to days; the line reaching "one minute by 2027" is an extrapolation of historical trends. Another chart makes the point more starkly: among vulnerabilities that will eventually be exploited, the 2022 sample still had roughly half unexploited about six weeks after disclosure, while the 2026 sample was near zero at the same point. Waiting for a regular patch cycle may no longer keep pace with risk, and this doesn't imply every vulnerability will be exploited within some fixed window.

Sternstein also emphasizes that AI cuts both ways in defense: faster vulnerability discovery can serve defenders. The practical question for the defense side is how long each step takes: spotting the risk, assessing its impact, deciding on a course of action, and actually deploying a fix. Even if detection tools speed up, a slow handoff downstream keeps the overall response time from improving on its own.

Security and observability lead other software groups in year-to-date stock performance

As of September 1, 2026, weighted by then-current market capitalization across categories; these curves reflect capital market pricing.

Security group commands a higher forward revenue multiple

The vertical axis is enterprise value to forward 12-month revenue, the horizontal axis is expected revenue growth, and bubble area represents market capitalization; this is a valuation comparison.

Among the software stocks in this week's charts, cybersecurity and observability names are clearly leading year-to-date, and forward revenue valuations reflect high market expectations. Capital is already pricing in this demand. Share prices and valuations don't prove a product actually defends anything; both incumbents and newer platforms still have to answer the same question: how do you convert faster risk detection into faster risk reduction?

Data centers carry AI demand onto the construction site

AI infrastructure demand ripples beyond the software industry, into data centers, power systems, cooling, and construction. The annualized pace of data center construction spending jumped from roughly $50 billion to about $75 billion within six months, an increase equivalent to roughly two years' worth of prior gains. US construction job openings have also ticked up recently. Using other construction employment trends as a baseline, Goldman Sachs estimates that since 2022, data center-related construction and skilled trades have added about 300,000 workers; the article cites a nearly 75,000 increase over the past year.

Data center construction spending accelerates and construction job openings recover

The red line reads seasonally adjusted annualized construction spending on the right axis; the blue line reads the three-month average of construction job openings on the left axis; job openings cover the construction sector.

Related construction trades exceed other construction employment trends by over 300,000

The left panel uses 2022 as the comparison base and includes categories such as HVAC, electrical, utilities, and commercial construction; the right panel shows year-over-year growth separately.

The relevant job categories in the chart include HVAC, electrical, utilities, and commercial building construction. Construction spending is reported as a seasonally adjusted annual rate, meaning it's the projected one-year total if the current pace continued, after stripping out seasonal effects. A higher annualized pace shouldn't be read as money already spent within a given period, and estimates relative to employment trends shouldn't be mistaken for net new jobs across the entire economy.

To understand what these gains mean, you have to look at the broader construction environment. After interest rate increases, much other construction activity slowed and labor demand came under pressure. Sternstein's read is that data center construction is propping up a blue-collar job market that would otherwise be weak; without this building boom, employment would likely look worse. The article mentions a wage premium for related work, but this week's charts don't show the specific figures.

It's important to distinguish "what work a project requires" from "how many net new jobs the economy gains." A facility brings construction demand, but it doesn't automatically tell you whether workers transferred from other projects or how many jobs remain once construction ends. Conversely, it would be a mistake to focus only on office automation and ignore the jobs being created by infrastructure buildouts.

Coming back to the hiring question, the concrete preparation individuals can do is to audit their own experience: which parts were just familiarity with a tool, and which parts show judgment under changing conditions, correcting mistakes, and producing verifiable results. Being able to articulate the latter is what gives experience information content beyond the number of years on a resume.