Microsoft Engineer Declares Manual Coding Over as AI Agents Take Charge

On September 3, David Fowler, a Distinguished Engineer at Microsoft with eighteen years at the company, posted a single sentence on X that set off a wave of comments across the developer community: writing code line by line is over. Five days later, it's clear this wasn't a passing infatuation with a new tool, but a description of a strategy Microsoft has been executing for months across Windows, GitHub Copilot, and its internal security systems.

Fowler is no random voice in this discussion. He co-created the SignalR framework, helped build NuGet and the Kudu deployment engine, spent years as a core developer on ASP.NET Core, and currently leads development of Aspire, Microsoft's tool for building distributed applications. When someone with that track record says manually typing lines of code in an editor is losing its point, it's hard to dismiss as marketing hyperbole.

What Fowler actually said

The engineer was careful to note he wasn't predicting the end of the programming profession or that developers would stop reading code. His point is narrower: manually typing individual lines in an editor is ceasing to be the primary way software gets built. Instead, programmers are increasingly expected to define requirements, design architecture, manage dependencies, and review what an agent has generated before it reaches production.

Typing code is absolutely over - David Fowler, Distinguished Engineer, Microsoft

The tools built to back it up

Behind the declaration lies concrete infrastructure. Aspire, the project Fowler leads, gained agent-friendly features in versions 13.1 and 13.2 that let AI agents configure development environments on their own. Since February 2026, GitHub Copilot agents have been able to build and test Windows projects and run linters, and since August they've also gained support for sessions in the Windows Subsystem for Linux. They can modify repositories, run tests, and open pull requests on their own for human review.

Then there's MDASH, Microsoft's internal system that uses more than one hundred AI agents to scan Windows, Azure, and identity systems for security vulnerabilities. It shows the company isn't treating agents purely as tools for writing new features, but also as a maintenance and quality-control layer for its enormous existing codebase.

Numbers that are hard to ignore

The scale of the shift comes straight from statements by Microsoft's leadership. Satya Nadella, asked by Mark Zuckerberg during the LlamaCon conference, admitted that 20 to 30 percent of the code used in the company's software today is created with the help of artificial intelligence. Kevin Scott, Microsoft's Chief Technology Officer, went further, forecasting that by 2030 as much as 95 percent of code could be generated by AI. By comparison, Sundar Pichai has cited a figure just over 30 percent for AI-written code at Google.

Those figures aren't unambiguous proof of success, though. Neither Microsoft nor Google has precisely described the methodology behind them, so industry commentators view the numbers with some skepticism. AI performs noticeably better with Python than with C++, which Microsoft attributes to simpler syntax and a larger volume of training data for the former. Critical components of Windows itself, where a mistake can be costly, remain the domain of human programmers.

Security concerns remain

Enthusiasm for coding agents is tempered by a 2026 report from Veracode, which found that 44 percent of tested AI-generated code contained known security vulnerabilities. That figure raises the question of what the speed gains actually cost: if nearly half of the code produced by agents needs security fixes, the developer's role as reviewer and quality gatekeeper doesn't disappear, it just changes shape.

Microsoft itself appears to understand this, since alongside promoting coding agents it is also expanding MDASH, a system meant to catch exactly this kind of problem before it reaches production Windows and Azure systems. That suggests the company sees agents as both an extra pair of hands for writing code and an additional layer of control, not a replacement for human oversight.

What this means for Polish developers

For Polish development teams, many of whom work within the Microsoft ecosystem, Fowler's declaration signals that tools like GitHub Copilot will increasingly shape day-to-day work, from setting up environments to opening pull requests. Companies planning to hire or train IT teams should expect that the ability to define requirements for agents and review their output will grow in importance relative to writing code from scratch.

At the same time, Veracode's data on vulnerabilities in AI-generated code is an argument against rolling out agents uncritically without added review processes. Teams already using Copilot or similar tools should treat AI's growing share of the codebase as a reason to strengthen, not relax, testing and code review.