For decades, software development has played musical chairs. Waterfall gave way to agile, DevOps shortened distances, CI/CD sped releases, and low-code abstracted parts of coding. Now agentic coding lets AI plan, write, test and modify software. The question is not whether agents can code. It is whether developers remain engineers or become expensive rubber stamps.
At an AWS briefing in Bengaluru, I asked Distinguished Developer Advocate Darko Mesaroš where developers fit in an agent-driven future. He did not predict robot doom. He described Kiro, aimed at preventing AI slop: software fast enough for a sprint and fragile enough to break under a gentle breeze.
I know that problem. Once I told a friend to make my office plant less sad. I forgot to say which plant, what sad meant, or that the plant was plastic. The result: a tiny disco ball on a cactus I did not have. Vague intent plus eager helper equals chaos.
Kiro slows the magic before it becomes a mess. Instead of dumping thousands of lines at once, it first creates requirements, a design document and a task list. Mesaroš calls the spec a contract between developer and agent. That sounds dry, until you appreciate that contracts assign blame.
Old engineering still matters. Testing, code review, security checks and delivery pipelines do not vanish because a chatbot can write loops. Kiro Crew adds deterministic checks, agent checks and human approval. It is DevOps with more layers: agents inspect agents while a person referees from the sidelines.
What changes is elevation. The developer moves above the code. A person may not read every generated line, but must know what changed. Did another endpoint appear? Where are logs? Where are credentials hiding? That is less typing, more orchestration, like a conductor who gets fired if the orchestra sounds like a car alarm.
There is irony. As tools make implementation easier, knowing the machinery underneath matters more. Mesaroš says developers are studying CPU and GPU scheduling because such knowledge helps them steer agents. It is like being handed a self-driving bus: no wheel, but brakes and traffic lights still matter.
Kiro Crew tries to remember you. Persistent memory, learned preferences, reusable skills, scheduled jobs and parallel agents make it less like autocomplete with attitude and more like a tiny workforce. AWS says it grew from three internal engineers to tens of thousands of Amazon builders.
Enterprise adoption is the test. Weekend apps can survive AI chaos; hospitals and trading platforms cannot. AWS highlighted Indian uses like Dhan, Pristyn Care and Analytics Fox, where specs, steering files and quality gates turned speed into discipline. The goal is fewer surprises, not more pull requests.
Attention remains everything. A capable agent will build the wrong cathedral if you ask for a parking lot with ambience. Developers are not disappearing. They are moving upward, directing agents, guarding intent, and staying the last adults in a fast-learning machine room. The job is no longer typing code. It is deciding what code deserves to ship.
