You trained for the safe career. Now agents write code and junior roles vanish. An honest look at what is happening to programmers and why it hurts.
This in-depth guide covers everything you need to know about will ai replace software engineers? what the data shows. Based on verified income data and real-world case studies from our database of 138 side hustle tactics.
Partly, and the part it replaces first is the beginning of the career. AI already does a large share of the routine coding that used to train new engineers, while demand for experienced engineers holds up. Coding agents now solve most of the textbook bug-fix tasks on public benchmarks, big tech companies say AI writes a quarter to three quarters of their new code, and the entry-level end of the market has shrunk faster than any other part of it. If you are senior, your job is changing. If you are trying to get in, the door is much narrower than the one your older colleagues walked through.
You did everything you were told to do. You picked computer science because it was the safe degree, the one your parents could explain to relatives without wincing. You took the loan, did the internships, ground through LeetCode on weekday nights, and maybe you even landed the job. Now you sit in stand-up and hear a manager describe how an agent closed forty tickets overnight, and you do the maths on your rent, your student debt and the money you meant to send home.
Or you have not landed the job yet. You are on application 200, and the postings that do appear ask for "3 to 5 years of experience" for roles that used to be called junior. You wonder whether the ladder you trained to climb still has a bottom rung.
What AI coding agents can already do
The fastest way to see the change is through the benchmark the industry uses to measure it. SWE-bench gives a model a real GitHub issue from a real open-source project and checks whether its patch passes the project's own tests. In 2023 the best systems solved almost none of these. The Stanford researchers behind the "Canaries" paper note that performance on software engineering benchmarks rose from 4.4% to 71.7% between 2023 and 2024. A year later, the Stanford AI Index 2026 reported that on SWE-bench Verified "performance rose from 60% to near 100% in a single year".
That benchmark is now close to saturated, so the field built harder ones. Scale AI's SWE-Bench Pro uses 1,865 tasks across 41 repositories, with reference fixes that average more than 100 lines across four files. When it launched, Scale reported that top models scored around 23% on it. By October 2026 the top entry on the public SWE-Bench Pro leaderboard scores 61.5%. The harder test bought the field about a year.
The length of work these agents can handle keeps growing too. METR, a research group that measures AI capability, found in March 2025 that the length of tasks AI agents can complete had been doubling roughly every 7 months for six years. The direction is plain: tasks that took a junior engineer an afternoon two years ago now take an agent a few minutes, and the tasks that take a week are next.
How much code companies say AI already writes
The people who run the largest engineering organisations on earth have been unusually open about this, and the numbers keep climbing.
In October 2024, Alphabet CEO Sundar Pichai told investors: "Today, more than a quarter of all new code at Google is generated by AI, then reviewed and accepted by engineers." Eighteen months later, in an April 22, 2026 post for Google Cloud Next, he wrote that 75% of all new code at Google is now generated by AI and approved by engineers, up from 50% the previous autumn.
At Meta's LlamaCon on April 29, 2025, Microsoft CEO Satya Nadella said that 20% to 30% of the code in Microsoft's repositories was "written by software", meaning AI.
Anthropic CEO Dario Amodei went further. At the Council on Foreign Relations on March 10, 2025, he predicted that within three to six months AI would be writing 90% of the code, adding: "And then in twelve months, we may be in a world where AI is writing essentially all of the code." In May 2025 he told Axios that AI could wipe out half of all entry-level white-collar jobs and push US unemployment to 10% to 20% within one to five years.
Some leaders have tied this directly to hiring. On Salesforce's February 26, 2025 earnings call, Marc Benioff said: "And we're not going to hire any new engineers this year." He credited a 30% productivity increase in engineering. In June 2025, Amazon CEO Andy Jassy told staff: "In the next few years, we expect that this will reduce our total corporate workforce" as the company used AI across its work.
Treat these figures carefully. "Share of new code generated by AI" is a company's own measure, it often counts autocomplete, and every line is still reviewed by a person. But the trend line comes from the companies that employ the most engineers in the world, and they are telling shareholders the same thing.
A question to sit with: if three quarters of the new code at your employer were written by a model, which part of your week would still need you?
The entry-level door is closing fastest
This is where the problem gets personal. The pain is concentrated at the start of the career.
Indeed tracks job postings against a February 2020 baseline of 100. Software development postings peaked at about 234 in February 2022, fell to about 61 in May 2025, and stood near 78 in early October 2026. That is roughly two thirds below the peak and still more than a fifth below where they were before the pandemic, while Indeed's index of all US postings sits slightly above its pre-pandemic level.
There has been a rebound since early 2025, and Indeed's own economists looked at who it is for. Guillermo Gallacher of Indeed Hiring Lab wrote on July 8, 2026 that "71% of the increase in software development job postings between May 2025 and May 2026 is from senior roles". A second Hiring Lab analysis found that in Q1 2026, senior positions made up 69.3% of software development postings and entry-level positions just 4.5%, the lowest entry-level share of any sector it measured.
Payroll data shows the same thing from the employer side. Stanford economists Erik Brynjolfsson, Bharat Chandar and Ruyu Chen studied ADP payroll records covering millions of US workers through June 2026. In the August 2026 update of their "Canaries in the Coal Mine" paper, they found that employment of workers aged 22 to 25 in AI-exposed occupations "now stands 19% below where it would be had it kept pace with that of their less-exposed peers", while experienced workers show no comparable gap. They found the gap works mainly through fewer young people being hired, rather than more being fired, and they use software development as one of their case studies. The authors are careful to call these early indicators rather than proof of cause, and they report that the gap persists after controlling for interest rates and remote work.
The graduates feel it. The Federal Reserve Bank of New York's February 2026 release, using 2024 Census data on graduates aged 22 to 27, puts unemployment for recent computer science graduates at 7.0% and computer engineering graduates at 7.8%. Out of 74 majors, computer engineering ranks second-worst and computer science fourth-worst, behind anthropology and alongside fine arts. Recent nursing graduates sit at 2.1%.
Imagine telling a 17-year-old in 2019 that a computer science degree would carry a higher jobless rate than art history. That is what the Fed's table shows.
Students are already voting with their feet
The next generation has noticed. The National Student Clearinghouse reported in August 2026 that undergraduate enrolment in computer and information sciences fell 8.4% at four-year colleges in spring 2026, and 11.2% at two-year colleges, with graduate enrolment in the field down 14.0% in fall 2025. This comes after a decade in which the major roughly doubled in size.
If you are a parent, this may be the hardest part. The advice you gave your child, "learn to code and you'll always have work", was good advice when you gave it. The ground moved after the tuition was paid.
What the official forecasts say
Government forecasters are calmer than the CEOs, and the gap between them is worth understanding.
The US Bureau of Labor Statistics projects that employment of software developers will grow 10% from 2025 to 2035, with about 106,100 openings a year across developers, QA analysts and testers, and a median wage of $134,040 in May 2025. That is still faster than average.
The same agency sees the narrower role of computer programmer shrinking 7% over the same decade, from 110,800 jobs. Its explanation is blunt: computer programming work continues to be automated, and many companies plan to use AI to automate repetitive programming tasks, with higher-skilled work shifting to developers.
That split is the clearest official answer to "will AI replace programmers". The people who mainly translate specifications into code are in decline. The people who decide what to build, design systems and take responsibility for them are still projected to grow.
Global bodies frame it more broadly. The IMF's managing director Kristalina Georgieva wrote in January 2024 that almost 40% of global employment "is exposed to AI", rising to about 60% in advanced economies. Exposure can mean AI helps with a job or takes parts of it over.
Where the forecasts disagree is timing. BLS projections are built on trends that move slowly. The Stanford payroll data and the Indeed postings move monthly, and they show the entry level shrinking now. Both can be true: total developer employment can keep rising while the number of people who get a first job falls.
A question to sit with: if the profession grows but stops hiring beginners, where do the seniors of 2035 come from, and could you be one of them?
The numbers in one place
| Measure | Figure | Source |
|---|
| SWE-bench Verified, top score | 60% to near 100% in one year | Stanford AI Index 2026 |
| SWE-Bench Pro (harder), top public score | 61.5% (October 2026) | Scale Labs leaderboard |
| New Google code generated by AI | More than 25% (Oct 2024); 75% (Apr 2026) | Google, 2024; Google, 2026 |
| Microsoft code "written by software" | 20% to 30% (Apr 2025) | TechCrunch |
| US software dev postings (Feb 2020 = 100) | Peak 234 (2022), low 61 (May 2025), 78 (Oct 2026) | Indeed via FRED |
| Entry-level share of software dev postings | 4.5% (Q1 2026) | Indeed Hiring Lab |
| Ages 22 to 25 in AI-exposed jobs | 19% below trend of less-exposed peers | Stanford Digital Economy Lab |
| Recent grad unemployment, CS / computer engineering | 7.0% / 7.8% (2024 data) | New York Fed |
| CS undergrad enrolment, four-year colleges | Down 8.4% (spring 2026) | National Student Clearinghouse |
| Software developers, BLS projection | +10% (2025 to 2035) | BLS |
| Computer programmers, BLS projection | -7% (2025 to 2035) | BLS |
The counter-evidence
The honest picture has some brakes in it. When METR ran a randomised trial in early 2025 with 16 experienced open-source developers on 246 real issues, it found that developers using AI tools took 19% longer than without them, even though they believed AI had sped them up by 20%. METR has since said the tools have improved and that study is out of date, but it shows that benchmark scores and real productivity are different measurements.
Predictions have also slipped. In January 2025, Mark Zuckerberg said on Joe Rogan's podcast that "probably in 2025" Meta and its rivals would "have an AI that can effectively be a sort of mid-level engineer". By April 2025 he was guessing at "the next 12 to 18 months" for most of Meta's AI code to be written by AI. And Indeed found that US software development postings grew almost 15% after Claude Code launched in early 2025 while overall postings fell 7%, with senior engineers who can work with AI in especially high demand.
None of that helps much if you are 23 and on application 200. The growth is real, and it is going to people who already have experience.
A question to sit with: what would it take for an employer to see you as "experienced with AI" six months from now, regardless of your job title?
What this means for you this month
If you already have a developer job, the safest place to stand is close to the decisions: system design, reviewing what agents produce, talking to the people who will use the software, and owning outcomes when things break. Spend part of this month doing your normal work with an agent and keeping notes on what it gets wrong. That record is the skill employers say they now pay for.
If you are trying to break in, the classic junior role is scarce, so build proof outside it. Small paid projects count as experience. Some people are turning their coding skills into income by building and selling small tools (see micro-SaaS) or setting up automations for local businesses through an AI automation agency. Others earn while they search by doing AI training jobs, where models are graded by people who can read code.
If you are weighing whether to leave tech entirely, look at the numbers before the feelings: our guide to how much money you need before you quit your job is a sober place to start, and the wider picture is in the job market in 2026 and which jobs AI will replace.
The degree you paid for still teaches you how systems work. What has changed is the first step after it, and that step now has to be built by you.