For decades, writing software meant big teams, late nights, and armies of engineers.
That familiar picture may be vanishing fast.
Across major tech companies, artificial intelligence is reshaping how code is written, tested and shipped. One veteran engineer who has seen the inside of both Google and Amazon now believes the change could be so abrupt that half of today’s developer jobs simply disappear from large organisations.
A veteran engineer sounds the alarm
Steve Yegge is not a casual commentator. He has more than 40 years in software, including stints at Amazon and over a decade at Google, where he worked on large-scale systems and internal tools. Speaking recently on “The Pragmatic Engineer” podcast and newsletter, he laid out a stark forecast for his former industry.
Yegge argues that many large tech companies could shrink their engineering headcount by around 50% in the coming years, thanks to AI tools that make remaining developers dramatically more productive.
In his view, this is less a distant scenario than a trajectory already visible in how big firms plan their budgets and teams. The core idea: if a smaller group of engineers, armed with powerful AI assistants, can ship the same or more software, the financial incentive to keep huge teams fades quickly.
Why AI is forcing a brutal trade-off inside big tech
Behind the scenes, AI is turning into one of the biggest cost centres in tech. Training and running large models demands expensive data centres, high-end GPUs and constant access to proprietary or licensed systems. Serving millions of AI queries every day pushes infrastructure bills even higher.
Executives, faced with these costs, are making a simple calculation: money spent on compute and AI platforms competes directly with money spent on people.
For every engineer to work effectively with AI, companies must pay not only their salary, but also for the powerful tools and infrastructure that supercharge their output.
That leads to a tough choice. In many boardrooms, the emerging answer is: fewer engineers, better equipped. Companies are starting to prefer leaner teams with cutting-edge AI tooling, rather than large departments relying on mostly manual workflows. Meta’s leadership has already highlighted big productivity gains from AI-assisted engineering, echoing this shift publicly.
From writing code by hand to supervising AI agents
Yegge believes this is not just about cost-cutting. He describes a deeper transformation of what software engineering actually means.
➡️ Fine hair: this cut is “incredible for adding volume” after 60
➡️ 10 years younger effect”: the best short haircuts to wear after 50
➡️ France Shines In A Resilient Energy Niche In 2026 As LNG Tanker Membrane Orders Explode At GTT
➡️ This 100% natural treatment could help stop hair loss, say experts
➡️ Hedgehogs In Spring: This One Urgent Garden Corner Trick Can Save Them When They Wake Up
Traditionally, developers spent most of their time writing code line by line. Now, powerful code models can generate functions, classes or entire modules in seconds, based on natural language prompts or short snippets of context. The human role moves away from typing every instruction and toward specifying intent, reviewing suggestions and orchestrating AI agents.
The core skill is drifting from “how do I write this?” to “how do I ask for this, verify it and integrate it safely?”
Yegge also warns of a growing divide inside the profession. Engineers who fully embrace AI tools can produce far more output than peers who restrict themselves to basic autocomplete or occasional code generation. Over time, that gap can influence promotions, performance reviews and ultimately job security.
A new kind of engineering hierarchy
This shift creates at least two broad groups of developers:
- Those who treat AI as a central partner, learning how to prompt it, chain tasks and debug AI-generated code.
- Those who stick mostly to traditional workflows, using AI only for small snippets or not at all.
Yegge’s concern is that companies will need far fewer people in the second group. As AI systems improve, they compress the amount of “hand coding” required, which historically justified large teams.
Fewer jobs in giants, more power for small teams
A widespread cut in engineering roles at big firms does not automatically mean less software gets written. The opposite might happen. With AI amplifying individual productivity, small teams and startups gain new leverage against incumbents.
Modern AI agents can now assist with architecture diagrams, unit tests, integration tests, documentation, deployment scripts and monitoring dashboards. What used to take a full department can, in some cases, be handled by a handful of engineers coordinating several specialised AI tools.
A tiny group with strong AI skills can ship at a pace that once required a mid-sized company.
Commentators have compared this to the rise of cloud computing a decade ago. Back then, renting infrastructure on demand allowed young companies to challenge giants without building their own data centres. Today, AI offers a similar shortcut, this time in human effort rather than hardware.
Why some engineers are leaving big tech voluntarily
This changing landscape is already nudging career decisions. Some experienced engineers are walking away from stable roles at Google-sized organisations to join tiny startups or launch solo ventures. Their bet: a small, nimble team, heavily augmented by AI, can ship products faster than a sprawling corporation bogged down in process.
In that setting, each engineer acts more like a product owner and AI conductor than a pure coder. They define goals, design user flows, and then orchestrate a mix of human and machine work to get there.
What this could mean for current and aspiring developers
For people already working in software, Yegge’s prediction sounds harsh. A 50% cut in engineering jobs at major firms would mean intense competition for remaining roles and a premium on AI fluency.
At the same time, new types of work appear:
- Engineers specialising in evaluating and securing AI-generated code.
- Developers building internal tools that sit on top of AI models.
- Product-minded coders able to move from idea to prototype in days, using AI as leverage.
For students or career changers, the message is not “don’t learn to code”, but “don’t stop at syntax”. Understanding systems design, product thinking and how to collaborate with AI agents could matter more than mastering yet another framework.
Key concepts behind the shift
| Term | What it means for developers |
|---|---|
| AI agent | A system that can perform multi-step tasks automatically, such as generating code, running tests and fixing issues. |
| Orchestration | The process of coordinating multiple AI tools and services so they work together on a project. |
| Productivity gain | An increase in output per engineer, measured in features shipped, bugs fixed or systems maintained. |
| Infrastructure spend | Money poured into servers, GPUs and AI model access, which now competes with hiring budgets. |
Possible scenarios for the next decade
If Yegge’s forecast plays out, a typical large tech company in 2030 could run with half the engineers it employs today, yet deliver more frequent product updates. Internal workflows might look like this: a product lead writes a detailed spec in natural language; AI tools propose architecture and code; a small group of engineers reviews, adjusts and approves changes; automated systems test and deploy them.
At the same time, the number of small, AI-native startups may explode. A pair of developers could run a software-as-a-service platform serving thousands of customers, handling support, updates and custom features with a mesh of AI agents working behind the scenes.
The total amount of software in the world may keep rising, even if the number of human developers inside big firms falls sharply.
Risks, benefits and practical responses
This shift brings clear risks: concentrated layoffs, skills becoming obsolete faster, and increased pressure on those who remain. There are also benefits: faster innovation cycles, lower barriers to launching new products, and more creative roles for engineers who adapt.
For individual developers, a pragmatic response could include three moves: learn to use at least one major AI coding tool deeply, focus on understanding systems and architecture rather than just syntax, and practise working as a “manager of AI” — giving precise instructions, checking outputs and stitching together different tools.
For companies, the challenge lies in deciding how far to cut traditional roles while building up AI capabilities. Those that move too slowly may struggle with costs. Those that move too fast may lose valuable expertise and context that machines cannot yet replace.
What seems least likely, in Yegge’s view, is a gentle, decades-long adjustment. The tools are improving quickly, investment is surging, and financial pressures are real. For many engineers, the next few years could feel less like a gradual evolution and more like a turning point.