Insights · point of view

From SDLC to ADLC.

When agents write, test and ship software, the development life cycle stops being a straight line and becomes a loop. The discipline doesn’t disappear — it moves.

IANIA.AI · 7 min read

For two decades the software development life cycle has been a sequence: plan, build, test, release, operate. Humans wrote the code; the process existed to coordinate them. When agents start generating, testing and shipping, that sequence breaks — not because the steps stop mattering, but because the cost structure of each one changes.

What changes

Generation becomes cheap. An agent can produce a module, a test suite or a migration in minutes. The bottleneck moves downstream, to verification — the judgment about whether the output is correct, safe and aligned with intent. The scarce resource stops being typing and becomes reviewing.

That single shift cascades. The most valuable artifact is no longer the code; it is the specification and the evaluation — the precise statement of intent, and the tests that prove the system met it. One-off QA gives way to continuous evaluation: golden datasets, behavioral regression, checks that run every time the system or the model changes. And because the system now behaves probabilistically, observability — traces, cost, drift — becomes a first-class part of delivery, not an ops afterthought.

SDLC — a line plan build test release operate ADLC — a loop around intent & evaluation intent + evals context generate observe govern evaluate
Intent and evaluation sit at the centre; everything orbits them.

What stays

Engineering discipline does not go away — it relocates. Architecture, testing, security and code review are as important as ever; they just move from “write it by hand” to “specify the intent, curate the context, and verify the behaviour at scale.” Governance and guardrails stop being a gate at the end and become inline controls the agents operate within. And humans move up a level: from authors of code to architects of intent, reviewers of behaviour, and owners of outcomes.

Why it matters

Teams that treat AI coding as “faster typing” plateau quickly and drown in unreviewed output. Teams that redesign the life cycle around intent, evaluation and observability compound — they ship more, and trust what they ship. The ADLC is not a tool you buy; it is a way of working you design.

Designing that way of working is part of what we do. See how →

Book an AI Architecture Review