Waterfall → agile → agentic: the cycle is changing again
Waterfall tried to predict. Agile learned to iterate. Agentic engineering reduces the cost of each cycle again. That does not make agile principles obsolete, but it does require us to revisit ceremonies and artefacts designed for a different speed of execution.
My first company was born, to a large extent, from discomfort with the development cycle I had experienced. In the software factories where I worked, many projects were managed as waterfalls: first we tried to define everything, then build it and finally discover whether it addressed the problem. The cost of learning late was enormous.
When I discovered agile, I did not see a collection of ceremonies. I saw a way to bring decisions closer to evidence: small teams, visible work, frequent delivery and continuous conversation with the people who needed the product. The possibility of changing the cycle was the main reason I built my own company.
Later, I applied the same principles beyond technology—in human resources, marketing, operations and finance. The value was not in copying Scrum. It was in shortening the distance between a hypothesis, an action and what we learned from it.
Agents are now compressing the cycle again.
What does not change
The Agile Manifesto still defends especially relevant ideas: deliver value early, collaborate with stakeholders, adapt to change, maintain attention to quality and reflect in order to improve. None of that loses value because AI can produce code.
In fact, as production accelerates, the ability to inspect, adapt and maintain a shared direction becomes more important. The mistake would be to confuse agile with the exact reproduction of a meeting calendar.
Revisit the purpose of every practice
A practice deserves to continue if it retains a useful purpose. When that purpose changes, its form should be allowed to change too.
| Common practice | Original purpose | Possible agentic evolution |
|---|---|---|
| Sprint planning | Agree on a manageable block of work | Define objectives, boundaries and decisions; plan over shorter horizons |
| Daily stand-up | Coordinate people and detect blockers | Review exceptions, risks and decisions that need human intervention |
| Backlog refinement | Prepare future work for a scarce resource | Maintain a queue of questions and options, not just detailed requirements |
| Review | Inspect an increment at the end of the cycle | Expose evidence continuously and reserve the meeting for relevant decisions |
| Retrospective | Improve the work system | Also examine instructions, context, tools, evaluations and agent failures |
| Estimation | Coordinate expectations and capacity | Compare uncertainty, risk and validation cost as well as execution effort |
I am not proposing that meetings be abolished by decree. I am proposing that a practice should not continue to consume time merely because it once solved a problem that no longer exists.
From managing scarcity to directing abundance
In many teams, code was the scarce resource. There were more ideas than capacity to test them. With agents, producing alternatives and first implementations costs less. The new bottleneck shifts towards judgement: framing the problem, choosing between options, reviewing changes, understanding their consequences and maintaining coherence.
That changes the role of product and engineering. The objective is no longer to feed a delivery machine with perfectly prepared stories. It becomes the creation of a system in which human direction, agents, prototypes and evidence can move quickly without losing accountability.
Agentic does not mean instantaneous
Agents also fail, misunderstand, accumulate debt and produce a volume of work that somebody must understand. Architecture, testing, observability and domain knowledge do not evaporate. Faster execution can amplify a team’s weaknesses as much as its strengths.
The evolution is not “waterfall bad, agile good, agentic magic”. It is a recurring question: which part of our work system exists because of a constraint that has now changed?