The conversation around artificial intelligence often swings between utopian promises and dystopian warnings, especially when it comes to jobs. Software engineering sits at the forefront of this debate because AI tools have been adopted here faster and more deeply than in almost any other profession. Yet, instead of relying on speculation, we can turn to concrete evidence to see what is actually happening on the ground. By examining hiring trends, layoff reports, and productivity data, a clearer picture emerges that challenges the notion of imminent mass displacement. This evidence‑based approach helps separate signal from noise and offers a grounded view of how AI is reshaping work.
A useful way to think about modern knowledge work, including software development, is as a three‑layer sandwich: decide, execute, and deliver. The decide layer involves figuring out what to build, weighing user needs, market signals, and organizational goals. The execute layer is the actual writing of code, design, and implementation. The deliver layer covers testing, verification, integration, and ongoing maintenance. AI’s current strength lies in compressing the execute layer, where it can generate code rapidly and handle repetitive implementation tasks. However, the decide and deliver layers still require human judgment, contextual understanding, and accountability, which are not easily automated by better models alone.
Recent high‑profile layoff announcements have frequently blamed AI, but a closer look reveals a different story. When Block cut 4,000 jobs, the stated reason was AI‑driven efficiency, yet internal reports showed limited productivity gains and pointed to financial strain after pandemic‑era hiring surges. Snap’s reduction of roughly 1,000 roles was framed around AI writing most new code, but activist investor pressure for cost cuts was the real driver, and the cuts were not concentrated in engineering teams. Intuit’s 3,000‑person reduction was accompanied by partnerships with AI firms, but the CEO explicitly said the layoffs targeted coordination‑heavy roles and excess management layers, not AI replacement. These cases illustrate a pattern of “AI washing,” where AI is cited as a convenient explanation for decisions rooted elsewhere.
Broader survey data confirm that AI is often used as a scapegoat for workforce changes. Nearly six in ten U.S. hiring managers admit they emphasize AI when justifying hiring freezes or layoffs because it sounds more forward‑looking than admitting budget constraints. Forrester analysts note that when asked whether a mature AI system exists to replace the cut positions, nine out of ten companies say they have not even built such a tool. Harvard Business Review research shows that while a sizable share of executives have trimmed staff in anticipation of AI, only a tiny fraction have actually done so due to deployed AI systems. This ten‑fold gap highlights a widespread susceptibility to speculative narratives rather than evidence‑based planning.
Another telling source of insight comes from regulatory filings. New York’s version of the WARN Act now includes a checkbox for technological innovation or automation, allowing companies to disclose AI as a layoff factor. In the first year after its introduction, over 160 firms filed notices, yet not a single one selected the AI box. Follow‑up with the state labor department found that, as of late May, only one company—Nespresso—had marked the box, accounting for fewer than fifty out of roughly twenty‑five thousand layoffs in the period. If these filings are accurate, AI’s direct role in job losses appears negligible, suggesting that fears of massive AI‑driven terminations are largely unfounded at present.
Layoffs are also a poor metric for gauging AI’s productivity impact. Economic research shows that the primary effect of automation tends to be slower hiring rather than increased separations. Firing experienced workers erodes the tacit knowledge and organizational capital that make teams effective at using new tools, and it incurs steep costs in severance, morale damage, and future rehiring. Natural attrition usually achieves the same workforce adjustment over a few years without those downsides. Therefore, observing layoff counts alone can mislead us about whether AI is truly boosting productivity or merely serving as a convenient excuse for other strategic shifts.
Looking at overall employment trends paints a more nuanced picture. A study by Federal Reserve economists found that software engineer job growth in the United States has slowed by about three percentage points per year since the widespread release of ChatGPT, compared to a scenario without that AI wave. However, the study’s methodology omits self‑employed contractors, meaning some of the apparent slowdown could be absorbed by a rise in independent consulting or freelance work. Additional research indicates that AI lowers the barrier to starting software‑related ventures, which may offset any decline in traditional payroll employment. Consequently, the true health of the software talent market may be stronger than the headline figures suggest.
It is important to distinguish between direct AI replacement and indirect effects that nevertheless cause job losses. In some cases, AI undermines demand for a product itself—for example, AI‑powered tutoring platforms reducing the need for services like Chegg or AI‑driven Q&A tools affecting Stack Overflow—leading to layoffs that stem from market shifts, not from AI taking over workers’ tasks. Similarly, companies that sell AI, such as IBM or SAP, sometimes announce layoffs while shifting resources from legacy lines to fast‑growing AI offerings; this is ordinary portfolio reallocation around a revenue opportunity, not a case of technology displacing workers. Recognizing these nuances prevents us from conflating different causal mechanisms under a single “AI job loss” label.
A widely cited statistic—the percentage of code written by AI—often fuels the fear that programmers will become obsolete. Yet this metric is misleading because writing code has never been the primary bottleneck in software delivery. Studies show developers spend only a small fraction of their time on actual coding, with the rest consumed by meetings, debugging, requirement clarification, and coordination. When AI agents boost line‑of‑code output dramatically, the overall impact on shipment velocity remains modest because the limiting factors lie elsewhere. Consequently, a high AI‑generated code share does not translate linearly into reduced need for human engineers.
Through qualitative interviews and task‑analysis surveys, researchers have identified the genuine bottlenecks in software work: deciding what to build, verifying and being accountable for the output, and maintaining a deep understanding of the codebase, business context, and operating environment. These three elements form the decide‑execute‑deliver sandwich’s outer layers, which remain largely untouched by AI’s advances in code generation. As long as humans retain responsibility for strategic choices and answerability for results, they must continue to invest effort in grasping system intricacies, user needs, and potential risks—activities that resist full automation.
Empirical support for this sandwich model comes from a large‑scale GitHub study examining over 100,000 developers. The research found that AI assistants led to an eight‑fold increase in the number of lines of code written, confirming a strong compression of the execute layer. However, the same intervention yielded only about a thirty percent rise in the number of releases shipped, indicating that the decide and deliver layers continued to constrain overall throughput. This disparity suggests that even if AI could produce flawless, instant code, the gains would be marginal unless decision‑making and verification processes also evolve. In the future, the engineer’s role may resemble that of a crane operator: overseeing powerful automated lifts while ensuring safety, precision, and alignment with broader goals.
For professionals navigating this evolving landscape, the key takeaway is to cultivate strengths that sit outside the pure execution zone. Sharpening skills in product discovery, stakeholder communication, risk assessment, and systems thinking will become increasingly valuable as AI handles more of the coding grunt work. Investing time in learning how to supervise, prompt, and validate AI agents—treating them as powerful but fallible tools—will also pay off. Organizations should focus on redesigning workflows that keep humans in the loop for critical judgments, while using AI to accelerate prototyping and testing. Finally, staying informed about structural shifts—such as the rise of AI‑native rollups in traditional industries or the growth of independent AI‑augmented consulting—will help individuals position themselves for opportunities rather than succumb to alarmist headlines.