Google's CEO says three-quarters of new code is now machine-written, marking a rapid shift in how software is built.
April 24th, 2026: Pichai announces 75% at Cloud NextNew here? Follow stories to track developments over time. Create a free account to get updates when stories you care about change.
For most of computing's history, every line of production code at a major software company was typed by a human. Google CEO Sundar Pichai told an audience at Google Cloud Next 2026 on April 24 that 75 percent of new code at the company is now generated by artificial intelligence and reviewed by engineers afterward. That share was 25 percent in October 2024.
The workforce consequences are now explicit. Days after Pichai's announcement, Meta said it would cut 8,000 jobs — 10 percent of its total staff — beginning May 20, with its internal memo directly linking the reductions to funding AI infrastructure spending projected at $162–169 billion in 2026. Microsoft simultaneously launched the first voluntary retirement program in its 51-year history, targeting roughly 7 percent of its U.S. workforce; AI, Azure OpenAI, and GitHub Copilot engineering teams were explicitly exempted. Both companies named AI automation as the enabling factor for cuts in software testing, content moderation, and certain engineering roles — the clearest public statement yet that the headcount impact of AI-generated code has moved from hypothesis to balance sheet.
Why it matters
If most code at top firms is machine-written, the entry-level path into software engineering — the highest-paid mass profession in America — is being rewritten in real time.
Sundar Pichai - SVP, Android, Chrome and Apps, Google
Close-up of colorful programming code on a computer screen, showcasing digital technology.
MS-Exec-Nadella-Satya-2017-08-31-22 (cropped)
Anthropic CEO: "90% of Code Will be Written by AI in 6 months"
Matthew Berman
One Employee Replaces Teams At Meta, AI Writes the Code ...
Future Ready Leadership With Jacob Morgan
Sundar Pichai Reveals 25% Surge In Google's AI-Generated ...
Mint
Images from Openverse under Creative Commons licenses. Videos from YouTube.
Exploring all sides of a story is often best achieved with Play.
Fictional AI pastiche — not real quote.
"The Prince who replaces his captains with siege engines need not fear their mutiny — yet he would do well to remember that engines do not garrison towns, nor do they govern the loyalty of those who remain."
Your progress in this debate will be lost.
Choose one persona for each side of the debate
Select debater for this side:
No debate personas available right now.
Select debater for this side:
No debate personas available right now.
Make your prediction before the referee scores
Pick the question both personas must answer in the final round
Debate Oracle! You called every round!
Sharp Instincts! You know your debaters!
The Coin Flip Strategist! Perfectly balanced!
The Contrarian! Bold predictions!
Inverse Genius! Try betting the opposite next time!
The dominant search and advertising company, now repositioning its engineering organization around Gemini and agentic developer tools.
The largest enterprise software vendor, and through GitHub the operator of the most widely used AI coding assistant.
The maker of the Claude model family and Claude Code, which competes head-to-head with Google's Gemini and OpenAI's Codex inside developer workflows.
The owner of Facebook, Instagram, and WhatsApp, and one of the five largest AI infrastructure spenders globally in 2026.
Pichai opens Google Cloud Next 2026 saying three-quarters of new code is AI-generated, alongside a $240B cloud backlog and 750M monthly Gemini users.
Meta said it will cut approximately 8,000 employees — 10 percent of its workforce — starting May 20. The company's internal memo explicitly linked the reductions to freeing capital for its $162–169 billion 2026 AI infrastructure budget.
Microsoft offered early-retirement packages to employees whose combined age and years of service reach 70, targeting roughly 7 percent of its U.S. workforce. AI, Azure OpenAI Service, and GitHub Copilot teams were explicitly exempt, concentrating reductions away from AI engineering.
Senior engineers at the leading AI labs say they have stopped writing code by hand, delegating it almost entirely to in-house models.
Google releases an agent-first IDE bundled with Gemini 3 Pro, designed to orchestrate multiple autonomous coding agents.
Pichai updates the AI-generated code figure to about half of new code at the company.
At Meta's LlamaCon, Satya Nadella says 20 to 30 percent of Microsoft's repository code is now AI-written.
Pichai tells investors that more than a quarter of new code at Google is being generated by AI assistants.
Microsoft's GitHub releases the first mainstream AI pair programmer, suggesting code completions inside the IDE.
Predict which scenario wins. Contrarian picks score more — points lock in when the scenario resolves.
Big tech firms keep total engineer counts flat or slightly down, using AI to handle the marginal ticket. Junior hiring contracts hardest because review-and-orchestrate roles favor experienced engineers. Software remains lucrative but the field stops absorbing new graduates at the rate of the last decade.
Cheaper, faster code creation triggers a wave of new software that wasn't economical before — internal tools, custom enterprise apps, indie products. Demand grows faster than per-engineer productivity, mirroring what compilers did to assembly programmers. Hiring continues but the role shifts toward systems design and review.
AI-written code passes review but accumulates subtle defects, supply-chain vulnerabilities, and architectural drift. A high-profile incident — a major breach traced to an AI-introduced bug, or a regulatory finding — forces firms to dial back the share or impose stricter human gates. The 75% number plateaus or retreats.
Critics note that 'AI-generated' covers everything from a one-line autocomplete to a fully agent-written module, with no public methodology. Closer audits show the headline figure overstates substantive contribution. The share remains real and rising, but the public framing loses credibility.
IBM shipped FORTRAN, the first widely used high-level language, with John Backus's optimizing compiler. A routine that had taken 1,000 lines of assembly could now be written in roughly 50 lines of FORTRAN, cutting programming effort by a factor of five to ten.
Programming stopped being 'hand-to-hand combat with the machine' and opened to scientists and engineers outside the small priesthood of assembly coders.
Demand for programmers exploded rather than collapsed. Each productivity gain enabled software in domains that had been economically out of reach, and the profession grew tenfold over the following decades.
The closest historical analogue: a tool that drastically cuts the per-line cost of code. The compiler precedent is why Pichai and others argue AI coding will expand engineering employment, not shrink it — though the transition reshaped which skills mattered.
VisiCalc, then Lotus 1-2-3 and Excel, automated work that armies of bookkeepers and junior accountants had done by hand on green ledger paper. Economist James Bessen later documented that roughly 400,000 bookkeeping jobs disappeared while accounting and analyst roles grew by more than 600,000.
The most routine, entry-level financial roles contracted sharply through the 1980s and 1990s.
The profession reorganized upward: fewer pure data-entry jobs, more financial analysts and modelers doing work that wasn't feasible without spreadsheets.
Spreadsheets did to bookkeeping what AI coding may do to junior software engineering — eliminate the routine production layer while expanding the analytical and design layer above it.
Mechanized weaving displaced English handloom weavers, whose wages collapsed from around 21 shillings a week in 1797 to about 6 shillings by the 1830s. Organized weavers smashed machines across Yorkshire and Nottinghamshire; the British government deployed troops and made frame-breaking a capital offense in 1812.
A generation of skilled weavers was destroyed economically, with widespread poverty and migration.
Textile output and overall employment in the sector grew enormously, but the gains accrued to a different set of workers — factory operatives, not the displaced craftsmen.
The cautionary parallel: aggregate productivity gains from automation do not automatically benefit the workers whose specific skills are being automated. Whether AI coding follows the compiler path or the power-loom path will depend on how quickly displaced engineers can move into the new orchestrator and reviewer roles.