June 4, 2026
The Future of Software Engineering
The future of software engineering is a difficult topic to think about. When I do, I have to hold in my head convincing, yet opposite, arguments from well-known tech leaders and economists. Here are a few of those perspectives.
What tech leaders say
Dario Amodei
Dario's view is darker than most of his peers, and he explicitly warns against sugar-coating. He estimates AI could eliminate up to half of entry-level white-collar jobs — in tech, finance, law, consulting — within one to five years, with unemployment potentially reaching 10–20%.
His outlook on software engineering is worse. At WEF Davos (January 2026): "I have engineers within Anthropic who say, 'I don't write any code anymore. I just let the model write the code, I edit it, I do the things around it.' I think we might be six to 12 months away from when the model is doing most, maybe all of what we do end-to-end.”
Dario says that programming is a skill very close to the actual building of AI. The farther a skill is from the people building AI, the longer it will take to be disrupted. He believes software engineers are the first in line, not because they are uniquely replaceable, but because AI was trained by and for their domain.
Dario's predicted stages:
- 1. 90% of lines written by AI
- 2. 100% of lines written by AI
- 3. 90% of end-to-end SWE tasks automated
- 4. 100% of today's SWE tasks automated
- 5. 90% less demand for SWE
Still, he is careful about how he defines the stages that he believes we are moving through. “Lines written by AI” !== “SWE tasks automated,” which would explain why we aren’t seeing less demand for SWE yet. But in Dario’s view, that is where we are heading.
Sources: Lex Fridman Podcast #452 transcript · Machines of Loving Grace · WEF Davos, The Economist · Dwarkesh Podcast #2 · Axios: "A White-Collar Bloodbath"
Jensen Huang
Jensen's overarching thesis is that AI will elevate workers, not eliminate them. It boils down to the distinction that the purpose of a job is different from the tasks used to fulfill it. If AI automates tasks, the underlying purpose of the job will expand.
Because of this, Jensen believes the headcount of software engineers will grow, not shrink. Higher individual leverage means SWE teams will tackle bigger, more numerous problems. "I want my software engineers to solve problems. I don’t care how many lines of code they wrote. Solving problems, working as a team, diagnosing problems, evaluating the result, looking for new problems to solve, innovation, connecting dots — none of that stuff is gonna go away."
Jensen says that the definition of coding is “simply specifying.” The engineer articulates architecture and intent; AI does the mechanical translation. That means that the people who can code will increase from 30 million (today) to 1 billion in the near future. However, the engineers who thrive will be those who master when to be highly prescriptive in their specification, when to under-specify to let AI push creative limits, and who, as a result, own harder problem spaces.
All quotations sourced from Lex Fridman Podcast #494 — Jensen Huang: NVIDIA · Published March 23, 2026.
Marc Benioff
Marc brings an interesting view, because he’s speaking from inside a company of 83,000 people (15,000 of whom are software engineers) navigating a transition from traditional SaaS into an AI-enabled world. Also, his company doesn’t benefit as directly from AI growth as Anthropic and Nvidia do.
"Our engineering organization is probably more than 30% more productive, but I wouldn't call it 100% more productive." Marc says that even though he has had to rebalance his organization (for example, by moving or replacing customer support roles with sales roles), he is hiring more. "Those engineers are still needed. The model still cannot operate autonomously. We're not at that level yet of AI."
Marc’s “canary in the coal mine” is that even the top AI companies, according to their job boards, are still hiring a lot of engineers. He stresses that you can do more as an engineering executive now than you ever could before — because, coupled with a large language model, you’re not just the engineering executive; you’re also the product executive, the design executive, and the marketing executive. You’re all the executives.
While many companies have slowed their SWE talent pipeline, Marc is actively recruiting from high-academic-threshold universities. He wants the best of the next class of engineers, saying that companies like Salesforce “badly need these people.”
Yann LeCun
Yann's primary contribution to this debate is an argument about who is qualified to participate in it. He questions the credibility of anyone in the AI industry making predictions at all, including himself, despite being a Turing Award winner and the “godfather” of deep learning.
When Dario Amodei warned in April 2026 that AI would eliminate 50% of entry-level jobs within five years, Yann said:
“Dario is wrong. He knows absolutely nothing about the effects of technological revolutions on the labor market. Don't listen to him, Sam, Yoshua, Geoff, or me on this topic. Listen to economists who have spent their careers studying this.” — citing Daron Acemoglu, Erik Brynjolfsson, David Autor, Philippe Aghion, and Andrew McAfee.
Sources: LinkedIn — "Amdahl's Law for automation" (June 2025) · Business Insider — "Daron Acemoglu Is Skeptical of Anthropic CEO's Dire Jobs Warning" (April 2026) · Sahm Capital — "Ex-Meta Chief Scientist Yann LeCun Slams Anthropic CEO's Job Wipeout Warning" (April 2026) · TechCrunch — "Yann LeCun's AMI Labs raises $1.03B to build world models" (March 2026) · Futurum — "AMI Labs Raises $1BN Seed Round" (March 2026)
What economists say
Erik Brynjolfsson
Erik is the director of the Stanford Digital Economy Lab and co-author of The Second Machine Age. His framework splits all work into three phases:
- 1. Asking the right question
- 2. Execution
- 3. Evaluation
AI, Erik says, is taking over Phase 2. That is "the hinge" the future of labor turns on. Workers who master phases 1 and 3 gain extraordinary leverage; those who only execute are already losing ground. "Many workers will become what I call Chief Question Officers (CQO)... A CQO's primary job will be to possess the judgment to know what to ask, why it matters, and how to evaluate if the AI has actually succeeded. We will be the architects; the AI will be the builders."
Erik is generally optimistic about the future. "The cost of trying something new plummets" — a "Cambrian explosion" of new products and services as AI lowers the barrier to solving hard problems. But he warns against "the temptation to use AI merely to mimic and replace humans, driving down wages and concentrating power" — a pattern he calls the Turing Trap.
An August 2025 ADP study from Erik's lab, using high-frequency payroll records from millions of American workers, found that since generative AI's widespread adoption in late 2022, employment for early-career workers (ages 22–25) in the most AI-exposed occupations fell 13% on a relative basis — even after controlling for firm-level disruptions. Older workers in the same fields held steady or grew.
Software engineering is one of those AI-exposed occupations. A separate February 2026 NBER paper tracking roughly 240,000 U.S. manufacturing firms found that a 10% minimum wage increase raises industrial robot adoption by roughly 8%. Together, Erik argues, the labor market is being compressed from both ends: AI eroding entry-level white-collar work from the top, automation encroaching on blue-collar work from the bottom.
Sources: Erik Brynjolfsson, "AI Changed Work Forever in 2025," Time Magazine, January 2, 2026; Fortune, "Top AI economist finds link between robots and minimum wage hikes," March 4, 2026; Erik Brynjolfsson and Andrew McAfee, The Second Machine Age (2014)
Andrew McAfee
Andrew McAfee co-authored The Second Machine Age with Erik Brynjolfsson. His position is that "nobody knows anything" about how AI will reshape productivity, jobs, or competitive advantage — ironic, he notes, when the loudest alarmists often run AI companies. History counsels against panic and complacency alike: technology diffuses slowly, and the employment consequences of new technologies are usually mispredicted.
"I think it's already inevitable that what we've got is going to replace a lot of existing white-collar knowledge work. That's not the same thing as eliminating a lot of white-collar knowledge jobs." "The most breathless projections of transformative change, I think, wildly underestimate how friction-ridden a process the diffusion of these capabilities is."
Andrew considers it "entirely plausible" that knowledge workers face decades of what happened to factory workers over the last half century: declining relative demand for particular skills, flat wage growth, and anemic prospects for many. Not apocalypse, but not comfortable either.
On software engineering, McAfee cites Brynjolfsson's finding that "in the professions that are most exposed we see the greatest softening in demand for recent college grads." His most urgent warning is against companies cutting junior pipelines: "How else are people going to learn to do the job except via on-the-job learning and apprenticeship?" Firms pulling back on entry-level hiring also cut off the pipeline of the people who are most likely to be AI enthusiasts and AI power users in the organization.
Sources: Andrew McAfee, HBR Strategy Summit 2026 podcast, April 2026 · Andrew McAfee, The Permanent Problem podcast, March 2026 · MIT/Let's Data Science: "MIT Researcher Warns Against Cutting Entry-Level Hiring," April 2026 · Andrew McAfee & Erik Brynjolfsson, The Second Machine Age (2014)
Moving forward
It seems that Dario Amodei is on one end of the spectrum ("white-collar bloodbath"), Jensen Huang is on the other ("headcount will grow"), and everyone else falls somewhere in between. Taking it all together, here are two things we know to be true.
- 1. The nature of software engineering has already changed. We wouldn't be having this conversation otherwise. We debate the current extent of the capability of agentic coding systems, and we debate the maximum capability they can reach. But we don't debate the fact that they already provide us with significant value. The engineers who use AI to the best of its capability will exceed and replace those who don't.
- 2. The employment consequences of new technologies are usually mispredicted. Andrew McAfee, along with Jensen Huang and many others, reference Geoffrey Hinton's famous prediction that we should "stop training radiologists" because AI would soon outperform them at reading images. Hinton was right — models surpassed radiologists on many image-recognition benchmarks over the subsequent decade. And yet, the number of radiologists increased, reaching an all-time high in both number and salary in the early 2020s. Why? It turns out that AI made it easier to justify more imaging. With faster triage and prioritization, radiology departments could handle higher volumes, which in turn encouraged more use of imaging. This is the classic Jevons paradox: as AI makes a task cheaper and faster, total consumption of the task increases, offsetting individual productivity gains.
The future is not guaranteed. While the 13% drop in early-career employment is real, it is unclear whether this signals a permanent shift or a temporary freeze born of uncertainty. The big question is whether or not the radiologist story repeats. If cheaper software follows the Jevons paradox the way cheaper imaging did, then lowering the cost to build means more software worth building, and demand for the people who specify and evaluate it grows the way Jensen and Erik describe. If instead the job stays fixed while its tasks drain away to automation, McAfee's decades-long squeeze is just as plausible.
For those considering this career: know that you are taking a risk. For those of us already in it, let's take this uncertainty as reason to act, not wait. My personal view is that software engineers will be in higher demand in the future, not less, because AI is rapidly absorbing domain knowledge. As raw expertise in any particular field becomes less differentiating, what could be more valuable than deeply understanding AI and the art of building with it?