Every generation of software delivery leaders inherits a methodology shaped by the failure mode of the generation before it. Understanding that inheritance is the only way to know what to keep.
The Principle That Does Not Move
Software delivery has been evolving for sixty years toward a single objective: more efficient delivery of value to the people who use what we build. That objective has not changed. What has changed, repeatedly, is our diagnosis of what was getting in the way.
Waterfall was not designed by people who wanted slow software. It was designed by people who believed that if you planned well enough at the start, execution would follow reliably. The plan was the intelligence. The humans were the mechanism. And for a period of time, in certain kinds of engineering, that worked.
It stopped working when the problems got more complex than any plan could fully anticipate. Software, it turned out, was not like building a bridge. Requirements changed mid-construction. Users wanted something different from what they had asked for. The feedback that mattered most came too late to act on. Waterfall's failure mode was not bad intentions. It was a structural mismatch between the nature of software and the nature of the process designed to build it.
Agile was the corrective.
What the Manifesto Was Actually Correcting
The four values of the Agile Manifesto are worth reading not as timeless principles but as precise responses to specific Waterfall pathologies.
Individuals and interactions over processes and tools: because Waterfall had reduced humans to process nodes, interchangeable and largely silent.
Working software over comprehensive documentation: because Waterfall had produced elaborate documentation that described software nobody had yet built and nobody could verify.
Customer collaboration over contract negotiation: because Waterfall had turned the customer relationship into a legal transaction at the point of maximum ignorance about what would actually be needed.
Responding to change over following a plan: because Waterfall had made plans sacred and change a failure of planning.
The manifesto values were not timeless truths. They were a corrective written in direct response to a known failure mode.
That corrective was right for its moment. Agile reorganized delivery around human judgment, human collaboration, and human adaptability, because those were exactly the capacities that Waterfall had systematically suppressed. The methodology fit the diagnosis.
The question worth asking in 2026 is whether the diagnosis has changed.
The Failure Mode Has Shifted
Agentic software development, the practice of deploying AI agents that can plan, write, test, and deploy code with increasing autonomy, does not have the Waterfall pathology. Agents do not need to be liberated from process rigidity. They do not suffer from siloed communication or slow feedback cycles. They execute with a consistency and speed no human team can match.
What agents introduce is a different failure mode entirely.
An agent can generate thousands of lines of code that pass every test and satisfy every stated metric while silently violating the intent behind those metrics. It can optimize confidently in the wrong direction, because optimization requires a target and the target is always a human construction that may not have fully captured what was actually meant. It can act faster than any oversight mechanism designed for human-paced delivery can detect.
The failure mode has shifted from human rigidity to human abdication.
Waterfall suppressed human judgment by treating humans as machines. Agentic development risks suppressing human judgment by making human involvement feel unnecessary. The pathology looks different. The cost is similar.
Agile's corrective, centering human interaction and adaptability, was the right answer to Waterfall's question. It is not the complete answer to agentic development's question. That requires a different set of values, written for a different set of tensions.
A Proposal
The following is not a replacement for the Agile Manifesto. Agile's values remain relevant wherever humans are the primary actors in delivery. What follows addresses the tensions that emerge when agents take on execution and humans must decide what that means for their role.
| Agile (2001) | Agentic Development (2026) |
|---|---|
| Individuals and interactions over processes and tools | Human judgment and system understanding over autonomous execution |
| Working software over comprehensive documentation | Validated outcomes over generated outputs |
| Customer collaboration over contract negotiation | Continuous intent alignment over requirement specification |
| Responding to change over following a plan | Adaptive governance over uncontrolled optimization |
Each of these values names a specific tension that agentic delivery creates.
Human judgment and system understanding over autonomous execution. The scarce resource in agentic delivery is not execution capacity. Agents provide that in abundance. The scarce resource is comprehension: leaders and teams who understand what the system is doing well enough to know when it is wrong. Organizations that optimize for agent throughput while hollowing out human understanding are building toward a failure they will not detect until it is expensive to reverse.
Validated outcomes over generated outputs. Generating software is no longer the constraint. Agents can produce outputs at a scale and speed that makes the volume of production a distraction. The constraint is verification: is what was generated correct, safe, and genuinely valuable? A system that produces working code solving the wrong problem is not a delivery success. The discipline of the agentic era is validation, not generation.
Continuous intent alignment over requirement specification. The traditional requirement assumed a gap between what was specified and what was built. Agile shortened that gap through iteration. Agentic development changes the problem: agents can implement changes so rapidly that the gap between original intent and current system behavior can widen faster than any specification process can track. This is what Entrowise calls Intent Drift: not a failure of documentation, but a failure of alignment between what was meant and what was built. The corrective is not better requirements. It is continuous, active re-alignment of agent behavior to business intent.
Adaptive governance over uncontrolled optimization. Agents optimize whatever they are pointed at. If the target is incomplete, the optimization will be incomplete. If the governance around the agent does not evolve as the agent's capabilities evolve, the gap between what is being optimized and what should be optimized will grow. This is not an argument for bureaucracy. It is an argument for governance that is designed to adapt, rather than governance designed for a pace of change that agentic systems have already made obsolete.
What Humans Are Actually For
Agile insisted that humans were more than machines. Agentic development finally forces us to answer what humans are actually for.
Waterfall treated humans as deterministic executors. Agile corrected that by insisting humans brought something irreplaceable to the work: judgment, collaboration, adaptability. That correction was right. But it left the deeper question unanswered, because in 2001 there was no practical pressure to answer it. Humans were still doing the building. The question of what humans contribute that machines cannot was theoretical.
It is no longer theoretical.
In agentic delivery systems, the answer to that question is not optional. It has to be designed in. Humans are the source of value definition: the judgment about what is worth building, for whom, and why. Humans are the source of risk accountability: the recognition that when a system fails, someone has to own that failure and have the authority to prevent the next one. And humans are the diagnostic layer: the capacity to look at what a system is producing and ask whether it is actually solving the right problem.
Those capacities were always the most important ones. It took the arrival of agents to make that undeniable.
The eternal principle of software delivery remains unchanged. We are still trying to get the right value to the right users as efficiently as we can. What changes, as it always has, is our understanding of what is getting in the way.
If your organization is navigating the shift to agentic delivery and finding the old diagnostic tools no longer fit, the Entrowise diagnostic framework was built for exactly this kind of system-level question.