Computer Programmers
Scrub through 91years of this role's history, from when it first emerged, through every wave of technology that reshaped it, to the cited projections for where it's heading next.
Fortune publishes "Employment for computer programmers in the US has plummeted to its lowest level since 1980 — years before the internet existed" (March 17, 2025). BLS OEWS data shows 121,200 programmer jobs in May 2024 — the lowest since the early 1980s. CPS rolling-average data shows a 27.5% drop in programmer employment since approximately 2023, coinciding with ChatGPT's November 2022 launch. Researchers begin describing this as "one of the earliest visible AI labor-market effects" in occupational data — the first white-collar occupation where the AI productivity shock is legible in census-level employment statistics rather than just survey responses.
The tools that defined the work
Select an era to see how it reshaped the work.
Machine code and assembly language (punch cards, plugboards)
The first programmers worked entirely in machine language — binary or hexadecimal opcodes entered via patch panels, punch cards, or toggle switches on the front panel. ENIAC used plug-board wiring to route signal paths; programs were the physical configuration of cables. There was no distinction between writing a program and physically rewiring the machine.
Punch-card systemsBatch accounting Assemblers and early compilers (Grace Hopper's A-0, 1952)
Grace Hopper at Remington Rand developed the A-0 system in 1952 — the first compiler, which translated mathematical notation into machine code. This was the conceptual bridge between machine-level programming and high-level languages. Hopper also pioneered the concept of machine-independent code, arguing (against widespread skepticism) that a computer could be instructed in something closer to English.
Work toolChanging equipment FORTRAN (1957) and COBOL (1959) — first high-level languages
FORTRAN (FORmula TRANslating system), released by IBM in April 1957 under John Backus, was the first high-level language to achieve widespread commercial use — it dramatically reduced the time to write scientific and engineering programs and opened programming to non-specialists. COBOL followed in 1959, designed by a DoD committee that included Hopper, specifically for commercial data-processing applications. COBOL would become the most widely deployed business programming language in history and remains so by lines-of-code count today.
Effect on the workFORTRAN is documented as having "greatly increased programmer productivity and significantly lowered costs" — IBM's own history credits it with making programming accessible beyond a small group of experts, which expanded the market for programmers rather than displacing them.
Mainframe processingComputerized records Time-sharing systems (MULTICS 1964, IBM TSS, BASIC 1964)
Time-sharing allowed multiple users to interact with a single mainframe simultaneously via terminals, replacing the batch-card-deck paradigm. BASIC (Beginners All-purpose Symbolic Instruction Code), designed at Dartmouth in 1964, made interactive programming accessible to non-specialists and students for the first time. The shift from batch to interactive development fundamentally changed the feedback loop for programmers — errors were visible in seconds rather than hours.
Work toolChanging equipment Structured programming — Pascal, C, software engineering discipline
Dijkstra's March 1968 letter to CACM, "Go To Statement Considered Harmful," launched the structured-programming movement — arguing that programs should use structured control flow (if/else, for, while) rather than arbitrary goto jumps. Niklaus Wirth's Pascal (1970) and Dennis Ritchie's C (1972) codified structured programming in widely adopted languages. This era also produced the software-engineering discipline: the 1968 NATO Software Engineering Conference coined the term and articulated the "software crisis" — complex systems were taking far longer and costing far more than estimated.
Work toolChanging equipment Personal computer — IBM PC (1981), Turbo Pascal, Visual Basic
The IBM PC's 1981 launch and the explosion of PC clones created an entirely new programming market: shrinkwrap software. Borland's Turbo Pascal (1983) and later Turbo C cut compile-link-run cycles from minutes to seconds on a $500 machine. Microsoft's Visual Basic 1.0 (1991) brought event-driven GUI programming to non-specialists. The PC era split the profession into enterprise/mainframe programmers (still running batch COBOL on IBM iron) and a rapidly growing PC/shrinkwrap cohort working in C, Pascal, and then C++.
Work toolChanging equipment Internet era — Java (1995), JavaScript, scripting languages, the web stack
The commercial internet created a third programming paradigm alongside mainframe and desktop: the web application. Java (released May 1995), JavaScript (September 1995), Perl, PHP, and later Python became the dominant languages of the dot-com boom. Every company suddenly needed a website, then a web application, then e-commerce; programmer demand surged to the 700,000 peak by 2000. The bust in 2001-2002 wiped out the speculative layer — 168,000 tech layoffs were announced in 2001 alone — but the web stack and internet-oriented programming became permanent.
Effect on the workThe dot-com boom drove programmer employment from ~435,000 in the early-to-mid-1990s to a peak above 700,000 by 2000; the 2001-2003 bust collapsed employment back toward 435,000 by 2006 as the speculative capacity unwound.
Work toolChanging equipment Open source + GitHub (2008) + Stack Overflow (2008) — networked knowledge
GitHub launched in April 2008, making Git version control trivially accessible and turning open-source contribution into a public portfolio signal. Stack Overflow launched in September 2008, creating a searchable, reputation-weighted repository of programming Q&A that dramatically reduced the cost of learning and debugging. Together they changed the craft: the "lone coder in a library" archetype was replaced by the connected networked developer who could find a working solution to almost any standard problem in minutes.
Work toolChanging equipment Cloud computing — AWS (2006), Azure (2010), GCP (2011)
Amazon Web Services launched its first commercial cloud services (S3 and EC2) in 2006, eliminating the need to provision physical servers. For programmers, this meant infrastructure became code: deployment, scaling, monitoring, and fault tolerance were now programming problems, not sysadmin problems. Cloud also accelerated the offshoring of routine programming work by making remote collaboration frictionless — a programmer in Bangalore could push to the same codebase as one in Seattle with identical tooling.
Work toolChanging equipment OpenAI Codex (August 2021) — LLMs trained on code, powering GitHub Copilot
OpenAI published "Evaluating Large Language Models Trained on Code" in August 2021, introducing Codex — a GPT model fine-tuned on 54 million GitHub repositories. Codex solved 28.8% of HumanEval benchmark problems at first attempt and 70.2% with sampling; GPT-3 had scored 0%. GitHub used Codex as the engine for its Copilot technical preview (launched June 29, 2021), which was the first AI tool that generated plausible code inline in the programmer's editor at production quality.
AI audit toolsPattern detection GitHub Copilot GA (June 21, 2022) + Cursor + Claude Code — AI-native coding
GitHub Copilot became generally available on June 21, 2022 — the first widely-deployed AI coding tool with a subscription model. A GitHub-sponsored controlled experiment of 95 developers found Copilot users completed tasks 55% faster (1h 11min vs 2h 41min, p=.0017). Cursor launched in 2023 as the first IDE built from scratch around large-context AI, enabling multi-file editing and autonomous feature completion. Anthropic's Claude Code (2025) extended the agentic model to full software engineering tasks including planning, implementation, and pull-request creation. By 2025-26, computer and mathematical occupations accounted for 34% of all conversations on Claude.ai and 46% of Anthropic API traffic, with "modifying software to correct errors" as the single most common task across both platforms.
Effect on the workThe 12-month rolling average of CPS employment for computer programmers fell 27.5% between 2023 and 2025, coinciding with GitHub Copilot's rollout. Multiple researchers cited this as one of the earliest visible AI labor-market effects in occupational data.
AI audit toolsPattern detection
What credible sources project
Scrub the slider past now to anchor each scenario on the scrubber. The spread is the range of futures credible sources project for this role.
What's shifting in the work right now
The historical view above shows how this role has moved. This is the present-day detail: which AI tools are picking up which tasks, where the edge still is, and the natural directions this work can grow.
What's changing in your day
Three parts of your work where AI is already doing real lifting, and what stays yours.
AI is taking this onUse AI coding agents (Copilot, Cursor, Amazon Q Developer) to generate boilerplate, scaffold repetitive modules, and autocomplete routine code blocks — then review, test, and integrate the output into the existing codebase.
Use AI coding agents (Copilot, Cursor, Amazon Q Developer) to generate boilerplate, scaffold repetitive modules, and autocomplete routine code blocks — then review, test, and integrate the output into the existing codebase.[3]
Shift focus from typing code to critically reviewing AI output: verify logic, edge cases, and security before committing. Build prompt-engineering fluency for your specific tech stack.
AI is taking this onWrite and update inline code comments, README files, and API documentation by prompting AI tools with the source code — replacing manual documentation passes that are chronically deferred in maintenance shops.
Write and update inline code comments, README files, and API documentation by prompting AI tools with the source code — replacing manual documentation passes that are chronically deferred in maintenance shops.[6],[1]
Review AI-generated docs for accuracy — auto-generated descriptions often miss business context and exception handling rationale. Add the "why" that AI cannot infer from the code alone.
AI is taking this onParticipate in agentic software development workflows where autonomous AI agents (Devin, Cursor Agent) handle multi-step tasks — plan, implement, test, and open a pull request — while the programmer acts as reviewer and approver rather than implementer.
Participate in agentic software development workflows where autonomous AI agents (Devin, Cursor Agent) handle multi-step tasks — plan, implement, test, and open a pull request — while the programmer acts as reviewer and approver rather than implementer.[7],[8]
Build skills in agentic workflow design: learn to decompose tasks into clear sub-goals, write effective agent prompts, and catch failure modes before they propagate. The emerging role is "AI-orchestrating programmer," not "absent programmer."
Where this role is heading
Natural next steps for someone with your foundation: not exits, evolutions.
Computer and Information Systems Managers
Computer and Information Systems Managers lead entire technology functions — setting strategy, managing teams, and aligning IT investments with business goals. Long-tenured programmers with cross-functional relationships and institutional system knowledge are credible candidates, especially as AI compresses the programmer headcount they would have previously supervised.
- · Technology strategy and IT governance (ITIL, COBIT fundamentals)
- · Budget ownership and vendor contract negotiation
- · People management: hiring, performance reviews, team building
- · Business-case writing and C-suite communication
See the same long-arc view for your own profession.
Browse the directory by industry, or search by title or SOC code. New roles ship every few weeks. Every profile cites every claim.
Browse all roles