The Software Development Lifecycle Is Dead (Category Software)

In this interesting February article 2026 Boris Tane is making a high-profile announcement about the death of the SDLC, which in my opinion reflects the situation strategically, but ahead of current approaches by a year. I think we'll move on from Classical PDLC for AI-native development And this article by Boris. mention @brolnickij on @ai4sdlc, where you can discuss posts from this channel and beyond.

If you go back to the main thesis of Boris in his article, he is like this. Classic SDLC with requirements phases → design → implementation → testing → code review → deployment → monitoring is no longer broken down into separate stages Instead, a short cycle appears: intent + context → agent → build/test/deploy → observe → repeat

The author’s view on the new lifecycle by stages is very interesting. Requirements are no longer frozen before the start of the task, but are clarified during iterations. System design is no longer prescribed in advance, but is detected in a dialogue with an agent. The implementation of changes in the code almost entirely goes to the agent. Testing no longer exists separately, but tests are written with code. (How many wanted before) PR/code review as a separate ritual should disappear and become an exception-based (Just in case the machine didn't ask) Deploys become truly continuous and by design separated from release via flags/rollbacks Monitoring turns from the last stage into the main safety layer of the whole system.

From this he has a rather hard conclusion: the new main skill is context engineering, and the new safety circuit is observability. That is, the team that wins is not the team that has more ceremonies, but the team that is better at collecting context, setting limits to the agent and quickly closing telemetry back into the cycle of change. It beats well with DORA: they are separate. highlight AI-accessible internal data and context engineering are key capabilities for AI-assisted development.

All this looks very realistic for small companies or on greenfield projects - tools already really know how to work at the level of the entire repository: read codebase, change many files, run commands and tests, do refactoring and code review. OpenAI writes that Codex is designed for long engineering tasks and review, and Anthropic shows that experienced users give the agent more autonomy over time, although they continue to actively intervene along the way. Plus, benchmarks like SWE-bench Verified show that agentic systems already solve a notable fraction of the real issue from open source.

But here's what big companies are waiting for and what the transition will look like I told in the article.From PDLC to AI-native development"

#Engineering #AI #Metrics #Software #DevEx #Productivity #DevOps #Architecture #Culture #Engineering #ML #SystemDesign