The Company That Asks Marketers What They've Built

The Company That Asks Marketers “What Have You Built?”

On August 28, Andrew Ng pushed back directly against the idea that AI destroys jobs, in a podcast appearance. His argument: when economists break jobs down into individual tasks, AI can only handle about 30-40% of them—leaving the remaining 60% of tasks, the ones still done by humans, actually more valuable.

But later in the same conversation, he described his own company’s hiring bar this way: “All of my marketers know how to code.” When he interviews marketers, he asks what they’ve built—and if they haven’t built any software, that’s a problem.

The examples that followed were more concrete. One marketing team member built their own desktop app for Mac that scans the web for source material whenever they need to pick a topic to write about. On the finance team, an executive noticed staff spending hours every week opening documents and copying numbers by hand, so he wrote a script that opens the files automatically, checks the contents, and flags anything unusual. On the recruiting team, there’s simply an engineer sitting in the room.

Put the two stories side by side and something feels off. He says jobs aren’t disappearing, yet he’s demanding software skills from marketers. But these are two sides of the same shift. The company isn’t eliminating the marketer role—it’s just raising the bar for whoever fills it, demanding more coding experience and more things they’ve actually built.

---

The Software Job-Posting Numbers Ng Cited as Evidence

Ng cited software engineering job postings as evidence for his argument—contrary to what the doomsayers claim, he said, postings are actually rising. I checked.

According to Indeed Hiring Lab’s August data, as of August 14, 2026, the overall postings index stands at 101.8—just above pre-pandemic levels—while software development postings sit far below that, at 74.4¹. Since the baseline of 100 is set at February 2020, this one category alone is down more than a quarter from that point.

But Ng isn’t wrong, either. The same organization’s July analysis, setting February 2025—when Claude Code launched—as the baseline of 100, shows software development postings up about 15%, while overall postings fell 7% over the same period. The starting point was just so low, though, that even after the rebound, the category remains 27.5% below pre-pandemic levels.

The same database can produce both “up over the past year and a half” and “down more than a quarter from February 2020” at the same time. What creates the difference isn’t interpretation—it’s the baseline. Measured against a year and a half ago, it looks like a recovery; measured against six and a half years ago, it looks like a collapse. Ng picked the former; the side he criticized picked the latter.

The numbers swing harder in Korea, depending on the baseline

Korean numbers are far more volatile than American ones. According to data compiled by JoongAng Ilbo last June, entry-level job postings at large and mid-sized IT/telecom companies in March 2026 fell 73% year-over-year, based on figures from the hiring platform Catch.

It’s worth flagging that this 73% is itself a product of the baseline chosen. It’s a single month compared to the same month a year earlier, so if a company’s hiring cycle shifts by even one year, the number swings wildly. The same yardstick I just used to stress-test Eung’s claim needs to apply here too, in fairness.

That said, the numbers sitting alongside it confirm the direction. Naver, which had run a first-half new-graduate hiring round for three consecutive years, didn’t hold one this year. Musinsa’s junior developer hiring round — its first in 4 years — drew roughly 2,000 applicants, of whom 66 were hired. According to each company’s ESG reports, new hire counts from 2022 to 2024 fell at Naver from 599 to 231 to 258, and at Kakao from 870 to 452 to 314.

University employment rates tell the same story of decline. Comparing computer engineering employment rates between 2023 and 2025: Seoul National University fell from 83.8% to 72.6%, Hanyang University from 81.4% to 70.3%, and KAIST from 77.9% to 69.8%. These figures exclude graduate school entrants and military enlistees.

What needs to be read alongside this is tenure. Over the same period, average tenure at Naver rose from 6.9 years to 7.7 years, and at Kakao from 5 years to 6 years and 3 months. As new hiring shrank, the average tenure of existing employees grew. Taken together, these two indicators suggest we need to separate the situation of current employees from that of people trying to land a new job.

Corporate responses also point to a shift in hiring criteria rather than a retreat from hiring altogether. In a survey released last December by Wanted Lab, which polled HR managers at 153 domestic companies, 74.5% of respondents said they would maintain or expand their 2026 hiring volume. Companies planning to stop hiring altogether were not the majority. Instead, the same survey found that when asked what qualities define an ideal candidate, 64.7% cited job-specific expertise and 24.2% cited AI and data literacy. In other words, hiring volume is holding steady, but the criteria for who gets hired are shifting.

Developer postings are shrinking, and AI-related requirements are rising

Wanted Lab’s analysis makes the drop in developer postings even more explicit. After analyzing 72,793 job postings uploaded to Wanted between January 2025 and April 2026, of the 6 job categories that showed a statistically significant decline in 2026, 5 were developer roles.

The same analysis found a metric moving in exactly the opposite direction. The phrase “AI native” appeared 8x more often in postings, and “physical AI” 5x more — based on comparing each phrase’s share of total postings with the previous year.

Some developer postings have disappeared, and the ones that remain now attach AI-related requirements more often. Kisoo Jeong, who heads Wanted Lab’s AI division, explains that AI has already taken over the simple, onboarding-level tasks and coding once handed to new hires as a way to get them adjusted. Both Eung’s comment in the US that “every one of our marketers codes” and the 8x jump in “AI native” in Korean postings point to the same trend: cutting routine work and expanding the scope AI can execute.

Eung frames this shift as an expansion of job scope. Just as developers once split between frontend and backend have merged into full-stack roles, marketers who used to just coordinate campaigns now own an entire cycle from planning to execution, and recruiters take on a wider slice of the hiring process too. So what’s needed isn’t just knowing how to use AI. Domain knowledge capable of covering that expanded scope has to come along with it.

I’ve also been spending more time coding these days. That means more time actually building things, in a role that used to be just writing strategy documents — and the numbers above show this isn’t a matter of personal preference, but a result of the market itself changing.

The Fear Sellers and the Optimism Sellers

Early in the interview, Ng traced the root of AI misinformation to PR and regulatory capture². For a company that has spent billions of dollars training a model, it’s a real problem if someone else builds something comparable and gives it away for free to the entire world. So stoking fear to bring in regulation creates a regulatory environment that favors incumbents, while the open-weight³ camp gets its hands tied.

The point itself is accurate. The incentive structure really is shaped that way.

But you need to look at the interests on the other side too. In this same interview, Ng said that AI models are terrible for learning. When students use AI, their assignment scores go up, but what actually sticks with them afterward is far less. And in the very next question, he introduces the new company he’s founded.

On July 28, Coursera invested $100 million in LearnVector, a company Ng founded and now runs as CEO. On a fully diluted basis, that’s roughly a one-third stake⁴. The first product is slated for early 2027, and Ng remains chairman of Coursera’s board.

The claim that “today’s way of learning with AI is terrible” also happens to describe exactly the problem a new learning company like LearnVector is built to solve. That’s not to say Ng is lying — the data really does point in that direction. It’s just that if the fear-sellers have an incentive, so do the optimism-sellers. Neither side is a disinterested observer.

Look again at the advice he gave college students and new grads in the same interview, and this structure comes into view. His point was that university curricula can’t keep pace with change, so students should learn the latest skills separately, online — and the examples he cited were Coursera, DeepLearning.AI, and Udemy. Two of those three are organizations he founded, and Coursera and Udemy merged into a single company this past May. The advice itself is sound. But when advice happens to line up with demand for your own business, it’s better to know that going in.

That’s why, with numbers about AI and jobs, you shouldn’t just look at the conclusion — you need to look at the standard used to read them, too. The next time you come across a statistic about AI and jobs, check three things. First, what’s the baseline date? Depending on whether it’s one year ago or six years ago, the same index can read as a recovery or as a long-term decline. Second, what’s the denominator — number of postings, a share, or headcount hired? Third, what is the person making this claim selling? It could be a model, regulation, a course, or fear itself.

Oswarld’s Lens

There’s a reason this story feels even more uncomfortable in Korea. Hiring requirements have already changed, but the way we evaluate candidates against those requirements is still anchored to school, major, certifications, and internship duration.

The practice of scanning a resume for school and major, then counting certifications and months of internship experience, simply doesn’t capture “what have you actually built.” On the flip side, applicants keep hearing that they need a portfolio, but there’s no agreed-upon standard for what to build or how much of it. In the U.S., people like Eung publish their own interview criteria, giving job-seekers something concrete to work from. In Korea, that kind of public standard is largely missing.

What’s filling that vacuum right now is portfolio consulting and fear-based marketing. The blurrier the standard, the better business is for whoever’s selling. Employers could clear up a good chunk of this market just by spelling out exactly what output they want — but for now, we’re still stuck at the slogan stage of “skills over school pedigree.” Please, don’t spend your money on this… buy yourself some beef instead.

Closing

Let me sum this up by situation.

If you’re currently working, the fastest way to test this shift is to build one tool yourself that cuts down on repetitive work in your own job. What Eung’s finance team did wasn’t anything extraordinary — they just automated a document check they used to repeat every week.

If you’re hiring, don’t write “AI proficiency” in the job posting — write down what kind of things you want the person to have already built. Without that one line, applicants have no choice but to prepare according to the old baseline.

If you’re the one reading the numbers, checking just three things — the reference date, the denominator, and what the speaker is selling — is enough to filter out most of the exaggeration. The same standard applies whether the number leans fearful or optimistic. Like the 74.4 and 15% we looked at today, both are often true at once — they’re just built on different baselines.

Reader, in the end, the real story of this shift isn’t the total number of jobs — it’s how the bar for getting hired is changing.

💬 Go back and reread the job postings at your organization — is there a sentence that’s different from one posted three years ago? Tell me in the comments which new phrases have crept in.

📨 If you know a colleague who’s writing job postings these days, or preparing to switch jobs, pass this along. Checking just these three baselines can change how they prepare.

References & Further Reading

Primary sources

• The Singju Post, “Andrew Ng: The Biggest Opportunities in AI Aren’t Where You Think”, August 31, 2026. ··· This is the full transcript of the August 28 broadcast. The marketer-hiring criteria appear around the 22:20 mark.
• Indeed Hiring Lab, “US Labor Market Snapshot — August 2026”, August 24, 2026. ··· This is where the software development index figure of 74.4 comes from. It’s updated monthly, so it’s worth checking the latest value before citing it.
• Indeed Hiring Lab, “AI and Job Postings: From Destruction to Creation?”, July 8, 2026. ··· This piece shows both the rebound and the trough together, making it the best resource for understanding the baseline problem.
• JoongAng Ilbo, “Entry-Level IT Postings Plunge 73%, the Computer Science Major Has Become the “Comp-Song” Major” (a pun on “comp-gong,” the romanized shorthand for computer science, and “comp-song,” meaning “withered”), June 11, 2026. ··· This piece includes Wanted Lab’s analysis of 72,793 job postings along with computer science employment-rate data.
• Coursera, “Coursera Makes $100 Million Strategic Investment in LearnVector”, July 28, 2026. ··· This is the original release detailing the investment size and equity structure.
• AI Times, “Wanted Lab Releases Hiring Trends Report”, December 8, 2025. ··· This is a survey of 153 HR managers. The sample is small, so treat it as directional only.

Background

• FRED, “Software Development Job Postings on Indeed in the United States”. ··· You can plot the index yourself here. Adjusting the baseline as you go makes this issue’s argument click into place.
• Class Central, “Coursera Bets $100 Million That Andrew Ng Can Do What Coursera Can’t”, July 29, 2026. ··· This is a skeptical take on the same investment, questioning the valuation of a company with no product.

Related past issues worth reading

• The People Who Sell Fear to Jobseekers ··· This issue examined the distribution structure that turns fear into a product. This issue looks at the opposite speaker through the same lens.
• Even Ferrari’s Designers Still Start With a Pencil ··· This is the story of the ladder that newcomers once climbed to learn a trade. Reading it alongside this issue’s “ticket of admission” completes the picture.
• Ten Minutes Is All It Takes to Break Your Thinking Muscle ··· This issue collects the research behind Ng’s claim that it’s “the worst thing for learning.”

The author is Oswarld (Kwangseob Ahn). Current roles: Adjunct Professor at Sejong University, Strategy Consultant at INLEVEL9. Career, research, books, and recent work are kept current on the About page. Latest · July 2026: HEMA-2: A Consolidation-Aware Tri-Memory Architecture with Multi-Channel Scheduling for Lifelong Conversational AI.

📝 Glossary

Footnotes

1. Job Postings Index: An index Indeed uses to track change over time, setting the number of postings on February 1, 2020 to 100. Because it’s a ratio relative to that day rather than an absolute count, the same curve can read differently depending on where the baseline is set. ↩

2. Regulatory capture: A phenomenon in which the body writing regulations ends up acting in the interest of the parties it’s meant to regulate. A classic example is a rule created in the name of safety that ends up protecting incumbent players’ positions. ↩

3. Open weights: A release method in which a model’s weight files are made public so anyone can download and run them. This is narrower in scope than open source, which also opens up training data and code. ↩

4. Fully diluted: A method of calculating ownership stakes that assumes all rights not yet converted into shares — such as stock options — have become shares. This produces a more conservative figure than counting only actually issued shares. ↩