Plague swept through groups of hunter-gatherers in southeastern Siberia 5,500 years ago, leaving dozens dead in its wake—with DNA from Yersinia pestis bacteria still trapped inside their teeth.

University of Oxford ancient DNA researcher Ruairidh Macleod and his colleagues recently sequenced the telltale bacterial DNA in teeth from plague victims at four ancient cemeteries in the area around Russia’s Lake Baikal. The tragedy that befell these communities is now the earliest known plague outbreak, courtesy of the oldest strain of Y. pestis ever sequenced.

Unearthing a new backstory for the plague

Until recently, scientists who study the evolution of diseases have held two fairly solid ideas about the origins of plague, the disease caused by Yersinia pestis bacteria. It's a scourge so awful that it has gone down in history as not just a plague but the plague. The first idea is that the earliest strains didn't have the right genetic traits to be really lethal. And the second is that the plague first began menacing humans when the first farmers settled in densely packed towns alongside rats and domestic animals.

But the dead of Ust'-Ida I cemetery, near Lake Baikal, tell a very different story.

"Our findings demonstrate that the earliest known outbreaks of plague occurred in prehistoric hunter-gatherers centuries before infections are observed in Neolithic farmers," wrote Macleod and his colleagues in their recent paper.

That challenges our previous assumption that plague spillover was a side effect of people taking up farming and settling in permanent villages and towns, living closer to each other and to an assortment of animals (and their fleas).

"Much of the accepted theory around epidemiology of disease in the past is that this kind of thing shouldn't occur in hunter-gatherers because hunter-gatherers are constantly moving around the landscape because they're in such small groups all the time," said Macleod in a press conference. "The theory, at least, is that infectious disease can't really take hold and devastate entire communities in this way.”

So much for that theory.

Welcome to the world’s first plague cemetery

The Angara River flows from the depths of Lake Baikal. The people who lived along it thousands of years ago survived by hunting, foraging, and fishing. They would have lived in relatively small groups, but they seem to have stayed connected across hundreds of kilometers through marriage and family ties. Although their lifestyle would have been one of constant movement, they buried their dead in cemeteries such as Ust'-Ida, interring them with offerings of clay pots, stone tools, and bone and antler points.

This map shows the location of Ust'-Ida I and Shumilikha cemeteries near Lake Baikal and the Angara River Credit: By Tara Young, taray@ualberta.ca and NASA https://wist.echo.nasa.gov/api/ - NASA's freely offered GDEM https://wist.echo.nasa.gov/api/, Public Domain, https://commons.wikimedia.org/w/index.php?curid=21156871

At Ust'-Ida, archaeologists with the Baikal Archaeology Project unearthed a grim mystery: an unusually high number of dead children, a cluster of radiocarbon dates suggesting that many of the cemetery's occupants died at around the same time, and no evidence of violence. Something tragic happened to this ancient hunter-gatherer community, but what? Archaeologists thought ancient DNA might shed some light on the mystery.

Macleod and his colleagues started with shotgun sequencing, a technique used to identify the DNA sequences in a sample when scientists don't know exactly which organisms they're looking for. They used samples from the roots of 46 ancient people's teeth from four different cemeteries along the Angara River.

And to their complete surprise, they found plague.

Fun fact: Because dental roots are fed by lots of blood vessels, anything in your bloodstream is likely to pass through your teeth at some point, which means if you die with the plague, it may leave its DNA behind in your teeth. “This is really cool evidence that the plague was in the bloodstream, which is lethal,” said co-author Frederik Seersholm, a University of Copenhagen ancient DNA researcher who clearly knows a fun fact when he sees one, in a press conference.

About 11 of the 31 people Macleod and his colleagues tested at Ust'-Ida had Y. pestis DNA in their teeth, and Macleod says that's "consistent with pretty much everybody [in the cemetery] having died of plague," not just those 11. That's because the detection rate for plague DNA in the remains at Ust'-Ida matches that at Smithfield's, a known mass grave specifically for plague victims in London. It's safe to assume everyone buried there had the plague.

"We really didn't know what to expect going into this, so it was a complete surprise that we discovered this really, really early evidence for large-scale lethal outbreaks of plague amongst these hunter-gatherer communities at this point in time," said Macleod in the press conference.

Ancient DNA and future outbreaks

Macleod and his colleagues managed to sequence a full Yersinia pestis genome from at least one of the samples, and it turns out to be the oldest strain of Y. pestis ever sequenced. According to the research, it's very close to the base of the plague family tree, emerging just a few hundred years after Y. pestis last shared a common ancestor with another bacterium called Yersinia pseudotuberculosis. This ancient plague isn't quite the one we're familiar with today or the version that devastated medieval Europe.

This very early version of Yersinia pestis doesn't have some of the genes that made its descendants so virulent; it's missing, for example, a gene that produces Yersinia murine toxin, which helps the bacteria survive passing through a flea's digestive tract on its way from a wild prairie dog to an unlucky hiker. It also lacks the right genes to form buboes (the painful swelling and darkening of the lymph nodes that gives bubonic plague its name). But its genome, not to mention the bodies it left in its wake, reveals that this early strain of Y. pestis was still horrifically deadly and probably deeply unpleasant to have.

"There are really a kind of perfect cocktail of other types of virulence genes that cause it to be so deadly—particularly, unfortunately, for children," said University of Copenhagen evolutionary geneticist Eske Willerslev during the press conference.

Understanding that perfect cocktail could be useful for battling modern epidemics, despite this strain of Y. pestis being so different from the ones circulating now in North America and Asia.

“What it gives you is an idea of which mutations in combination {...} are something that survives in nature,” said Willerslev. Because any combinations of features that work well tend to reappear (in the same microbe or in a different species), he said, studying ancient bacterial DNA “actually gives you some information on how these pathogens, including the plague, will develop.”

Why did the plague kill so many children?

Bubonic plague spreads through flea bites, but pneumonic plague is a respiratory disease, which spreads in a similar way to the flu or COVID-19, and that seems to be how this early version would have passed from person to person. So we can assume it would have come with respiratory symptoms like cough and difficulty breathing, along with fever. But for children, it probably would have been even worse.

When archaeologists plotted the ages of the dead on a graph, they noticed a sharp peak in children between 7 and 11 years old. Adults older than 20, on the other hand, had the lowest death rate. That lines up with data from plague outbreaks thousands of years later in London, when parish records document local children bearing the brunt of the plague's death toll.

The Y. pestis genomes that Macleod and his colleagues sequenced offer a clue about why. According to Iversen, the 5,500-year-old strain carries a gene that makes what’s called a superantigenic toxin: a chemical that triggers a dramatic, disorganized overreaction by the immune system. Children are especially vulnerable to this kind of reaction, said Oxford University immunologist Astrid Iversen during the press conference, because their immune systems are still learning how to respond to pathogens.

Telling the story of an ancient outbreak

The outbreak probably started when the bacteria made the leap from an infected marmot (a type of ground squirrel that's still a common plague carrier in the area) to a single person and then spread like wildfire through several interconnected hunter-gatherer groups along the river. For millennia, people around Lake Baikal have hunted marmots for food and for their fur, and close contact with a plague-ridden marmot can spread the infection. This is how it goes: accidentally inhale a few droplets of blood while skinning your latest kill or eat an undercooked marmot stew, and you’ve just doomed your whole band. And the neighbors.

Why are all the plague reservoirs also things I want to pick up and hug? Credit: By Stéphane Magnenat - Own work, CC BY-SA 3.0, https://commons.wikimedia.org/w/index.php?curid=7566004

That scenario is supported by the fact that people at Ust'-Ida carried the same strain of plague as those buried 37 kilometers away at another cemetery, Shumilikha, which is what epidemiologists would expect to see if they were part of the same outbreak. The burial customs at the two cemeteries suggest they belonged to different subcultures within the wider Isakovo tradition, but DNA from the plague victims reveals threads of kinship connecting them—and the plague may have made those threads deadly.

Macleod and his colleagues sequenced the DNA of the plague victims, piecing together how they were related and (through radiocarbon dating) when each member of the family died. That data revealed that the plague seemed to have spread among family members, often killing several at close enough to the same time that siblings often share graves.

"The incidence of detected infections among co-buried kin... would be consistent with the transmission of plague among humans, particularly via pneumonic transmission in the scenario of concurrent deaths," wrote Macleod and his colleagues.

Or as Macleod put it during the press conference, direct spread between people makes a lot more sense than "an outlandish scenario that absolutely everybody got together at the same time and ate the same infected marmot."

At Ust'-Ida, a young boy shares a grave with his aunt; both had Yersinia pestis in their bloodstreams when they died. The aunt also has a teenage niece buried nearby in a grave alongside a teenage boy who isn't biologically related to her (it's hard to tell if they were adopted siblings or cousins, a couple, or just close friends). And the boy's father is buried nearby in yet another grave.

“It's so obvious from the way people are buried… that somebody was around to bury the dead that knew who these people were when they were alive,” said Macleod. “And that adds a really really human element to the scientific work that we've done, seeing the impact on communities and how these communities responded to this very tragic set of events.”

Nature, 2026 DOI: 10.1038/s41586-026-10540-5 (About DOIs).

Read full article

Comments

The Carrington Event was the most intense geomagnetic storm ever recorded. In September 1859, auroras were visible as close to the equator as Columbia and some telegraph stations were severely damaged by current induced in the lines. If a similar event occurred today, with a lot more more wiring to pick up current than just an embryonic telegraph network, the results would almost certainly be cataclysmic.

Various modifications to the grid have been proposed to avoid another storm of that magnitude bringing on a new dark age, but a recent paper in the journal Space Weather proposes a more radical solution: using the sun’s energy to create a massive barricade in space.

While the authors of the paper refer to this concept by the compelling name StormWall, it’s not a physical wall. It’s actually just gas, likely of alkali metal atoms, to be deployed by solar-powered satellites.

To oversimplify, the proposal is to release lots and lots of neutral gas in Geosynchronous Earth Orbit (GEO), in what the researchers call “artificial mass loading” — the neutral gas would of course be ionized by the storm, but in so doing could absorb up to 50% of the incoming energy of the geomagnetic storm, frustrating its coupling to Earth’s magnetosphere. As a bonus, it would protect not just terrestrial assets like the power grid, but everything in a lower orbit than the mass load: everything from communication satellites in GEO to the International Space Station. Assuming its hasn’t been reduced to debris laying at the bottom of Point Nemo by then, anyway.

In simulations, the StormWall required 384,048 kg of gas, which is not exactly trivial. But even accounting for tanking, the researchers estimate that would only take about six launches of SpaceX’s Starship. Though that does assume its GEO capabilities end up being roughly equivalent to the massive vehicle’s projected 100-tons-to-Mars payload capacity.

It’s certainly an interesting hack to solve a problem that has caused a lot of worry these past decades. If you’re interested in learning more about the record-setting geomagnetic storm, we have a piece about the 1859 Carrington Event that should give you plenty of anxiety about the frailty of our modern infrastructure.

Most brands celebrate a 60th anniversary with a retrospective book or a limited-edition colorway. Bowers & Wilkins celebrated theirs by unveiling what may genuinely be the most advanced loudspeaker range they have ever made. The 800 Series Diamond D5 arrived with that kind of quiet confidence that doesn’t need fanfare to make its point, even if it was announced to considerable fanfare.

I’ve always believed that truly great audio equipment occupies a strange place between technology and sculpture. The 800 Series has lived in that space for decades. It’s the kind of speaker you find in professional recording studios around the world, at Skywalker Ranch where teams have mixed and mastered legendary film soundtracks, and also in the living room of the person who just needs the room to sound exactly right. That dual citizenship, professional and deeply personal, tells you everything about what Bowers & Wilkins has been building toward.

Designer: Bowers & Wilkins

The D5 is the fifth generation of the Diamond series, and the tagline “60 years in the making” isn’t marketing hyperbole. It’s a mission statement rooted in John Bowers’ original True Sound philosophy: nothing added, nothing taken away. Every generation of 800 Series starts from the same question: what stands in the way of the music? The answers keep evolving. The ambition stays constant.

The range includes seven models, from the compact 805 D5 stand-mount to the flagship 801 D5 with its twin 10-inch bass drivers. The iconic Turbine Head, that distinctive aluminum sphere housing the midrange driver in complete acoustic isolation from the bass section, remains one of the most recognizable silhouettes in audio design. It was bold when it debuted, and it’s still striking today. It’s been refined here, not rethought, and I think that’s the right call. Some shapes earn the right to stay.

What’s new in D5 runs much deeper than the surface. The Space Frame Bracing system introduces parallel aluminum rails bolted directly to the rear Matrix cabinet bracing, making the enclosure significantly stiffer and mechanically quieter than its predecessor. A revised aluminum top plate, with thicker ribbing and updated decoupling mounts, better supports the Turbine Head and Solid-Body-Tweeter assemblies. The crossover components have been moved entirely outside the cabinet, mounted on aluminum rails at the rear, which eliminates internal air pressure fluctuations from affecting crossover behavior. As an added benefit, natural convection keeps those components running cooler during extended listening.

The Diamond Dome tweeter gets a new grille mesh, first developed for the acclaimed 801 D4 Signature, that’s more acoustically transparent while still protecting the dome. The result is better off-axis performance and noticeably improved resolution. Every midrange and bass driver across the range has also been upgraded with lower-distortion motor systems derived from Signature-grade components. That’s not a minor tune-up; that’s serious trickle-down engineering from the very top of the catalog.

Aesthetically, the D5 introduces four new finishes: Stealth Black, Warm White, Light Walnut, and Dark Walnut. The paint has been upgraded for greater depth and durability, and the design detailing across every surface, from the spine to the plinth to the drive unit pods, has been refined. These are speakers handcrafted in Worthing, UK, and they carry that provenance visibly. Luxurious isn’t too generous a word.

Where I land on all of this is that the 800 Series Diamond D5 represents something genuinely uncommon in a market crowded with premium pricing and thin justification: a product that earns its position through accumulated expertise and genuine craft. There’s real, demonstrable engineering here, the kind that takes decades to develop, and Bowers & Wilkins isn’t shy about showing their work. The D5 range is scheduled to ship in fall 2026, and the anticipation feels entirely warranted.

Sixty years of obsessive refinement, applied to a speaker that takes the living room as seriously as a professional studio, will do that. When the engineering is this thorough and the design this considered, the only question left is how loud you want to play it.

The post Bowers & Wilkins Spent 60 Years on These Speakers. It Shows. first appeared on Yanko Design.

We're about six weeks out from the debut of Spider-Man: Brand New Day, the follow-up to 2021’s No Way Home. It's been five years since Spidey graced the big screen, so naturally Sony Pictures has released a new trailer to build audience anticipation.

(Spoilers for No Way Home below.)

As previously reported,  No Way Home ended on a pretty bleak note, with Peter Parker (Tom Holland) asking Doctor Strange (Benedict Cumberbatch) to erase him from everyone’s memory to protect the multiverse, including MJ (Zendaya).

Brand New Day is directed by Destin Daniel Cretton (Shang-Chi and the Legend of the Ten Rings). Per the official synopsis:

Four years have passed since the events of No Way Home, and Peter is now an adult living entirely alone, having voluntarily erased himself from the lives and memories of those he loves. Crime-fighting in a New York that no longer knows his name, he’s devoted himself entirely to protecting his city—a full-time Spider-Man—but as the demands on him intensify, the pressure sparks a surprising physical evolution that threatens his existence, even as a strange new pattern of crimes gives rise to one of the most powerful threats he has ever faced.

Naturally, Holland is reprising his role as Peter Parker/Spider-Man, along with Zendaya as MJ and Jacob Batalon as Peter’s former bestie Ned Leeds. The film also features Jon Bernthal as Frank Castle/Punisher, Mark Ruffalo as Bruce Banner/Hulk, Michael Mando as Mac Gargan/Scorpion (captured by Spider-Man in Homecoming), Marvin Jones III as albino crime lord Lonnie Lincoln/Tombstone, and Charlie Cox as Matt Murdock/Daredevil (briefly Peter’s lawyer in No Way Home). Sadie Sink, Liza Colon-Zayas, and Trammel Tillman join the case in as-yet-undisclosed roles.

We learned from the first trailer back in March that something strange is happening to Peter physically; he woke up surrounded in webbing and sought the help of Banner. It’s possible that Peter’s DNA is mutating, which, said Banner, “would be enormously dangerous.” But if Peter can make it through this new cycle, he might just get a rebirth—and a fresh trilogy of Holland-starring Spider-Man.

In this latest trailer, Peter specifically asks Banner if there is a way to get rid of the bad parts and keep the good parts of whatever is going on with his mutation. Banner's response: "How would you decide what parts of nature are good or bad?"

Meanwhile, our friendly neighborhood Spider-Man is faced with a new threat that nobody can see. Peter is the only person who seems to immune to whatever strange power is affecting everyone else—and the "only one who can sense it." He does have an ally in Banner, who naturally Hulks out when the situation calls for it. Peter gets a chance to save MJ, who still doesn't remember their previous reality.  And he gets a grumpy pep talk from Frank Castle: "If you're gonna do something you'd better do it now." Perhaps those new powers will come in handy in the inevitable showdown.

Spider-Man: Brand New Day hits theaters on July 31, 2026.

 

Read full article

Comments

There is great anxiety and uncertainty about AI replacing jobs. How can we move past vague warnings and bombastic predictions and bring data to bear on this question? One good way is to look at the profession where AI capabilities are furthest along and adoption has been exceptionally rapid: software engineering.

In this essay, we argue that there is enough evidence to reject the narrative that once AI capabilities reach a certain threshold, it will cause mass layoffs. Given that this is true even in a sector with very few regulatory barriers, most other professions are likely to be even more cushioned.

We also have a good understanding of why this is the case. We can think of many kinds of knowledge work, including software development, as a “decide-execute-deliver sandwich”. AI compresses the “execute” layer — the middle of the sandwich — but the other two layers resist automation in a way that will not be overcome by capability improvements alone.

We conclude on a note of cautious optimism about the future trajectory of demand for software engineering. This essay is the first in a series, and the next one will look at reasons why individual software engineers’ careers might be rocky even if overall demand is healthy. The series is based on the published literature in economics and software engineering, our own evaluations and observations of AI agents, and many software engineers’ reflection on the present and future of AI impacts on their profession, gleaned both from published writings and our interactions with the community.

The stories of AI-driven mass layoffs in software seem to be classic “AI washing”

Consider three stories that made the headlines and how they contrasted with reality:

• In February, fintech company Block (maker of Cash App, Square, Afterpay, and other such apps) announced layoffs of 4,000 employees because, according to founder Jack Dorsey, AI is “enabling a new way of working” with “smaller and flatter teams”, specifically citing late-2025 improvements in model capabilities.
  
  But subsequent reporting revealed a radically different picture. After growing headcount more than threefold during the pandemic, the company was under massive financial pressure. A data scientist on the Cash App team, Naoko Takeda posted that Block “shoved AI down everyone’s throats” yet she saw “very limited gains in productivity.” She refused a 75% retention raise and quit. Other employees interviewed had a sharply different understanding of what AI was capable of at Block and whether Dorsey had a competent understanding of the issues.
  
  As Aaron Levie has pointed out, CEOs are uniquely prone to delusions about AI’s usefulness because they can build quick prototypes but can’t see the 90% of work it takes to turn it into a finished product. Dorsey’s public statements about AI seem to fit exactly this pattern.
  

• In April, Snap laid off about 1,000 people, with CEO Evan Spiegel primarily citing AI as the reason in his layoff memo. He also said that AI generated 65% of new code. In reality, the layoffs followed a campaign by an activist investor demanding cost cuts. (Snap has posted a net loss every full year since its 2017 IPO and shares were down over 30% in 2026). Tellingly, the nature of the cuts, such as 150 jobs spanning various roles in the augmented reality division, don’t correlate with the cuts we would expect to see if they were driven by AI (i.e. programming and other “AI-exposed” jobs across the board, not concentrated in any unit).
  

• In May, Intuit announced 3,000 cuts, alongside deals with Anthropic and OpenAI. The press connected the two, framing the layoffs as AI-driven restructuring. For once, the CEO actually pushed back on this easy narrative, saying that “none of it had to do with AI” and that the cuts targeted “coordination-heavy roles” and too many management layers.

We did not cherry-pick these examples. In every story about AI-driven software engineering layoffs that we examined, the same narrative violation emerged. It turns out that “AI washing” of job cuts is an economy-wide phenomenon, evidenced by many surveys:

• 59% of U.S. hiring managers admitted they emphasize AI when explaining hiring freezes or layoffs because it plays better with stakeholders than citing financial constraints.

• Forrester principal analyst J. P. Gownder says of companies preparing supposedly AI-driven layoffs: “When we ask if they have a mature, vetted AI app ready to fill in those jobs, nine out of 10 times, the answer is no—and they haven’t even started.”

• In a HBR survey of over 1,000 global executives, 21% had made large headcount reductions “in anticipation of” AI, with another 39% having made low or moderate anticipatory headcount reductions. In contrast, only 2% had already made large reductions in headcount related to actual AI implementation. The 10x gap suggests that executives, like everyone else, are highly prone to succumbing to the misleading narratives about AI replacing jobs.

Another interesting data point comes from the WARN Act, which requires certain disclosures of plant closings and mass layoffs affecting over 100 workers. In March 2025, New York became the first U.S. state to add an AI disclosure checkbox to WARN Act filings. In the full first year, more than 160 companies filed WARN notices. Not a single one checked the AI box.1 We reached out to the NY Department of Labor who confirmed that as of late May, only one company, Nespresso, checked the box.2 If these filings are accurate, only 46 out of about 25,000 laid off workers in New York State in the relevant period, or about two-tenths of a percent, were affected by AI.

Even more damning for the AI-driven-mass-layoffs narrative: layoffs are the wrong signal of AI’s potential productivity benefits in the first place! The research is clear that the effect operates through “slower hiring rather than increased separations”. Firing existing workers results in the loss of precisely the tacit knowledge and organizational capital that allows workers to operate AI effectively. Besides, it is expensive in terms of severance, damage to morale, and rehiring risk. Given these costs, it is largely unnecessary given that natural turnover achieves the same result in a few years.

So what does the data tell us when we look beyond layoffs to overall employment trends? An important paper from Federal Reserve economists compiles the evidence in the U.S. context. Software engineer employment is still growing, but they find that it is growing slower post-ChatGPT compared to a no-AI counterfactual, by about 3 percentage points per year. One important limitation of this study is that the methodology can’t capture self-employment, so it is possible that some of the slowdown in growth is being absorbed by entrepreneurship instead. We do have evidence from other studies that AI makes entrepreneurship easier. So the real picture is probably even healthier than the Federal Reserve study suggests.3

Finally, it is worth acknowledging two kinds of indirectly-AI-driven job losses in software engineering that are real, but different from AI replacing software engineers. First, AI sometimes decimates demand for the product, in cases like Chegg (homework help) or Stack Overflow (technical help), both of which have laid off workers. AI doesn’t directly do the job that these workers did, but rather obviates the need for it. The historical parallel is strong: Among the 270 jobs in the 1950 U.S. census, only one job was automated away — elevator operator. But many others were rendered obsolete by new technology, like the job of telegraph operator.

Another credible AI-driven layoffs story is among companies that sell AI, rather than buy it. So when companies like IBM or SAP announce layoffs because of AI, a more accurate framing is “we reallocated headcount from legacy functions to our fastest-growing product line.” That’s ordinary corporate restructuring around a revenue opportunity, not technology displacing workers.

Why coding agents haven’t led to labor displacement: the decide-execute-deliver sandwich

Many tech leaders, like the Snap CEO above, report the percentage of code written by AI alongside reports of layoffs or predictions of future job losses. This feeds into the simplistic mental model that once AI writes all the code, there is no need for coders. Fortunately, this mental model is wrong. This AI-written-code metric is almost completely disconnected from what matters for labor displacement. Here’s why.

Writing code isn’t, and never was, the bottleneck. For example, a 2019 paper summarized existing studies with the conclusion that “developers spend surprisingly little time with coding, 9% to 61% depending on the study”. This finding was consistent with the paper’s own data from 6,000 developers at Microsoft. As coding agents began to be taken up, there was an explosion of blog posts in late 2025 pointing out that writing code isn’t the bottleneck, as developers realized that using agents to write most of the code led to little impact on overall productivity [1, 2, 3, 4, 5, 6, 7, 8].

If writing code isn’t the bottleneck, what is? The task-breakdown surveys point at things like meetings or debugging. This just leads to more questions: what are developers doing in those meetings and why can’t it be done by AI? Won’t debugging get automated as capabilities improve? To understand the real bottlenecks, we have to get qualitative, and dig into software engineers’ own understanding of what it is they do that resists automation.

When we did this analysis, it revealed three things as the real bottlenecks (1) deciding and specifying what to build, (2) verifying and being accountable for what is delivered, and (3) the deep human understanding — of the codebase, the business, and the environment — required to carry out both of these.

In other words, software engineers’ work consists of a “decide-execute-deliver” sandwich (with understanding being a prerequisite for all three). AI has compressed the middle of the sandwich, but has left the two ends largely unchanged. As long as software development teams are in charge of decision making and accountable for what they deliver, engineers still need to spend time building up a deep understanding of the system. These are the three bottlenecks.

Figure: Software development consists of three layers: (1) Decision making — problem framing, specification, planning (2) execution — design and implementation (3) delivery — testing, verification, integration, maintenance, etc. Note that these are conceptual layers, not temporal phases. It is common to switch back and forth in the course of a project.

Evidence for the sandwich model of AI’s productivity effects comes from a recent paper on “Writing Code vs. Shipping Code”. Across 100,000 developers on GitHub, the researchers found that AI agents led to an eight-fold increase in the number of lines of code written, consistent with the idea that AI almost completely compresses the Execute layer of the sandwich. But this led to only 30% more releases, strongly suggesting that human bottlenecks (the Decide and Deliver layers) remain in place.4

Can the sandwich be further compressed? We don’t think so. At one end of the pipeline, development teams need to decide what to build. One of the most important lessons junior software engineers learn is that requirements specification (the profession’s lingo for this layer) takes surprisingly long, and if it is compressed, it leads to much more pain down the line. This layer is hard to automate because it requires thinking about user needs, market signals, organizational priorities, and in some cases regulatory constraints.

As AI capabilities improve, the kinds of decisions that can be delegated to AI increase over time. But this does not make the “decide” layer thinner — once a decision can be delegated to AI, it is no longer a source of competitive advantage, and the value of human decision-making migrates upward. Software increases in complexity over time, so there is no ceiling to this process.

At the other end of the sandwich, human teams need to be accountable for what they deliver. It is possible that some day in the future teams will ship mission-critical code without fully testing and understanding it, but today’s AI is so unreliable that such haphazard practices would represent an existential threat to software teams and their customers.

Even if the technical barriers go away in the future, we don’t have to cede control to AI. A central insight of AI as Normal Technology is that we can collectively choose to keep humans accountable through shared norms, law, and policy. This is a much more resilient way to control the speed of AI impacts and improve safety than trying to slow the development of technical capabilities. These speed barriers are already largely in place due to liability laws and sector-specific regulation, but can be further strengthened. (For a longer version of this argument, see the original essay.)

In this vision, as more and more of the execution layer gets delegated to AI, the software engineer’s role in the future becomes analogous to that of a crane operator. AI agents will do most of the cognitive heavy lifting; supervising the agent and keeping it in control becomes most of the human’s job.

Some commentators argue that a future with humans staying in control is unlikely because it is too costly to pay people to do so. There have already been a few viral stories of poorly-supervised coding agents deleting production databases or causing other types of damage. But we view these as “man bites dog” stories rather than an emerging norm. They go viral precisely because they represent such irresponsible and unusual behavior that they have shock value, and serve as regular reminders and learning moments helping the community guard itself against over-reliance on AI. As the aphorism goes, “if it’s in the news, don’t worry about it”. Still, being able to detect whether there is an uptick in poorly-supervised use of AI for high-stakes tasks — across the economy, not just in software engineering — remains one of the most critical data gaps we have today.

By the way, the sandwich getting squished is a new trend and it is not uniquely due to AI. Over two decades ago, the Bureau of Labor Statistics started tracking programming separately from software engineering. Roughly speaking, programmers are responsible only for execution while software engineers manage a bigger part of the sandwich. Not only has programming been shrinking, it is also pays much less because it is seen as grunt work. AI merely accelerates this long-existing trend, further devaluing purely technical skills.

This pattern — where humans remain heavily involved at both ends of the decide-execute-deliver sandwich, even as AI increasingly automates the middle layer, seems to be broadly applicable to most knowledge work, though it is farthest along in software. After all, complex decision making and accountability are common to most fields. A lack of recognition of this phenomenon has led to many overconfident claims about imminent job losses, such predictions about AI replacing radiologists.

Vibe coding is not agentic engineering

One reason for confusion about the extent to which software engineering is changing is the sloppy use of the term “vibe coding” to refer to a wide spectrum of practices, the ends of which are conceptually distinct and more dissimilar than similar.

In true vibe coding the user simply tells the agent what to do, doesn’t supervise it when it’s running, doesn’t review the code — might not even have the skills to do so — and doesn’t evaluate the output, beyond perhaps noticing when things are visibly broken.

This is in contrast to how most software engineers are actually using agents — as a tool, with the human remaining in control and accountable for the output. Fortunately, the term agentic engineering is gaining currency as a descriptor of this practice.

As agentic engineering has become the norm, engineers are discovering that supervising coding agents is surprisingly time consuming. For example, Simon Willison, a prominent developer and chronicler of the AI transition, has noted how he is mentally exhausted by 11am from supervising agents. This is consistent with our experience as well.

More quantitative evidence comes from SWE-chat, a dataset of coding agent interactions from open-source developers who opted into a logging tool. The study found that only 44% of agent-produced code survives into user commits, that vibe-coded commits introduce vulnerabilities at nine times the human-only rate, and that the most common user intent is understanding existing code, not generating new code (19% vs 13%). The self-selected nature of the dataset means that we can’t draw strong conclusions based on this study alone, but it does reinforce many other lines of evidence that vibe-coding and agentic engineering patterns are quite different.

To re-iterate, these are not two distinct categories. They are two ends of a spectrum, and there is a blurry middle. Not every project is either a throwaway or mission-critical. Not every workflow fits precisely in the left column or the right column of the table. But the key implication for the jobs question remains solid — companies can’t ship production software by hiring unqualified vibe coders instead of software engineers.

What does the future hold?

AI boosters might claim that mass layoffs are coming; they just haven’t happened yet because human-level software engineering abilities are very recent (or haven’t been achieved yet). But if the sandwich model is correct, these predictions won’t come true. AI has already largely compressed the middle of the sandwich (and the compression actually started decades ago). So even making the execution layer instant and perfect will only be a small change from the status quo. The reasons why the other two layers have resisted AI is not because of capability limitations.

In fact, not only are software engineering jobs not going away due to AI, there might even be an increase in demand for software engineers. When software (or anything else) gets cheaper to create due to technological productivity improvements, people will buy a lot more software (in econ jargon, software is highly “price elastic”). And as we have argued, AI doesn’t replace software engineers (the “elasticity of substitution” is low), so the demand for more software results in a derived demand for more software engineers. A loosely related but flashier economics term, “Jevons’ paradox”, is often thrown around in the AI discourse to describe this concept.

Historically, this has been the pattern — programmer employment in the U.S. has grown from near-zero around 1950 to millions today. This is sharply different from occupations such as agriculture in which labor demand was famously decimated due to mechanization and automation. The difference is that the amount of calories people consume is relatively fixed — even a 25% increase led to the obesity epidemic — whereas the amount of software produced has grown a millionfold. Modern cars have something like a hundred million lines of code running on their various on-board computers.

If there is a ceiling to the demand for code, we are nowhere near it. Virtually all cognitive work benefits from software. As AI makes coding cheaper, people are creating all kinds of one-off utilities — whether for work or personal use — that it never made sense to create until now.

To be clear, while we think there will be a lot more software in the future, and likely more software engineers, this doesn’t mean big tech companies will get even bigger. The majority of software engineers today already work in-house in non-software firms, and that share might grow in the future. Then there’s the idea of “AI rollups”, which refers to venture capital or private equity firms buying “Main street” businesses — dentistry practices, accounting firms, and whatnot — and rebuild them from the ground up to be “AI-native” by embedding software engineers or AI engineers into those businesses. Of course, it might end up being nothing more than hype. It’s too early to tell.

Some people predict that demand for software engineering skills will fall because of democratization. They acknowledge that there will be more software produced than ever before, and also that more human time will be spent producing software than ever before, but that this work will be done by people who are not software engineers. The idea is that AI will democratize software engineering to the extent that legal software, for instance, can be more easily created by those with training in law than in software engineering.

Maybe. But we’ll bet against it. In our view, this falls into the same trap of conflating vibe coding with agentic engineering, and the execution layer with the the whole decide-execute-deliver sandwich. In fact, when we look at the history of programming, there have always been claims that we are at the threshold of democratization — old languages such as FORTRAN, COBOL, and SQL were all accompanied by such prominent hopes at the time of their introduction. It never happened. The barrier isn’t actually learning the syntax. It’s having enough skilled judgment to make good decisions while maintaining accountability.

Ultimately the distinction may be semantic. It seems clear that the amount of time people spend on getting computers to do new things will increase over time. This might take the form of building software, or managing complex workflows using agents, or something else. It will require a mix of software skills, AI skills, and domain expertise. Whether it is today’s software engineers who will best adapt to fill these new roles remains to be seen.

That last point about the need for adaptation sets up the next essay in this series. The fact that aggregate labor demand in software is likely to remain strong doesn’t mean that most individual workers won’t be affected. We will argue that AI will create massive structural shifts in how software is produced, which will have big impacts on which software engineers stand to gain or lose — based on the types of firms they work in, their geography, their seniority, the pace at which they can adapt.

Further reading

Deena Mousa points out the superficiality of broad, economy-wide analyses of AI impacts based on metrics like “AI exposure”, and instead calls for “careful, occupation-specific work”. We hope that this series of essays will play a role in establishing a nuanced understanding of AI’s transformation of software engineering. We’ve previous coauthored, with Justin Curl, a paper analyzing AI in legal services that seriously engages with regulatory and other bottlenecks that make that occupation unique. We plan to do more occupation-specific deep dives in the future.

In a remarkable essay called No Silver Bullet 40 years ago, Fred Brooks distinguished between the “essential complexity” and “accidental complexity” of software. He argued that some of the complexity of software is accidental, arising from limitations of present technology such as the clunkiness of programming languages, and can be alleviated over time as tooling improves. But some of it is essential, because specifying the correct behavior of software is itself hard. He presents a forceful articulation of why the “decide” layer of the sandwich is thick and resists automation. Interestingly, hopes of boosting programmer productivity through AI were already prominent back then! Brooks argues that because AI or any other technology only reduces accidental complexity, it won’t result in an order-of-magnitude productivity improvement. (Brooks is the author of The Mythical Man Month, an essay collection that is almost certainly the best known and most influential writing on software engineering of all time. No Silver Bullet later became part of the collection.)

We are grateful to Felix Chen for feedback on a draft.