How Is Software Engineering Changing?
Agentic AI changes software engineering from writing every line to directing, reviewing and verifying work done by agents that read a repository, edit files, run tests and open pull requests. Engineers spend more of their time on specifications, context, review, evaluation and security, and on the standards that let agents work with tools, such as AGENTS.md and the Model Context Protocol. Industry data shows rapid adoption, but trust and measured productivity gains are still debated. This page collects the evidence, design guides, practices, benchmarks and governance frameworks, and points to the related directories on this site.
Evidence & Industry Data
7 entriesSurveys, platform data and studies on how developers use agents and what it does to productivity. Read the methods before quoting a number, because the findings differ.
Foundations & Design Guides
6 entriesHow agents are built and how engineers should think about them.
Practices, Specs & Agent Tooling
6 entriesSpecifications, instruction files and protocols that give coding agents the context they need.
Evaluation & Benchmarks
5 entriesHow the capability of coding agents is measured.
Security & Governance
3 entriesRisk frameworks for agents that read code, call tools and change repositories.
Practitioner Blogs & Radars
3 entriesOngoing commentary from engineers and consultancies working with agents.
Related Pages
5 entriesRelated directories on this site.
Frequently Asked Questions
How is software engineering changing in the agentic AI era?
Agents now plan, write, test and propose code changes, so engineers spend more time specifying work, supplying context, reviewing results and verifying them. GitHub reports that as of September 2026 developers started more than 40% of pull requests with an agent.
What is agentic software engineering?
It is software development in which AI agents carry out multi-step tasks, such as reading a repository, editing files, running tests and opening a pull request, under human direction and review. It differs from autocomplete, which suggests code one step at a time.
Do AI coding agents make developers faster?
The evidence is mixed. METR's 2025 trial found that experienced developers took 19% longer with AI on their own repositories, while the 2025 DORA report links AI adoption to higher delivery throughput. METR's 2026 update says selection effects make its newer estimates weak, so check the method behind any figure.
Which skills matter most when agents write code?
Writing clear specifications, supplying the right context, reviewing and testing agent output, designing evaluations, and understanding security. Specification-driven development and context engineering are the practices most often named.
How should teams govern coding agents?
Give agents least-privilege access in sandboxes, keep humans reviewing and approving changes, log what agents do, and test them against benchmarks and your own tasks. The OWASP Gen AI Security Project and the NIST AI Risk Management Framework are common starting points.