Process folk have spent decades espousing Agile, Scrum, and an entire sub-industry of frameworks promising higher velocity, quality, and predictability. For as long as I've been in the industry, these approaches have rarely delivered on their lofty principles. Instead, we've seen distorted, commercialized versions that prioritize rituals over results.
My colleague Jonathan Schneider captured this malaise perfectly in a recent LinkedIn post. He asked the tough questions:
How many leaders really adopted the values and principles?
How many people truly changed their behaviors to influence culture?
How many consultants honestly succeeded with Agile through genuine change management?
How many companies pivoted because of failed gospel from blogs and influencers?
He draws a sharp parallel between Scrum hype and today's AI hype, noting that the real winners of the Scrum era were often certification mills rather than delivery teams. Time will tell how accurate the comparison proves to be. But right now, as organizations race toward AI-augmented and agentic delivery, we need to ask: What parts of existing Agile/Scrum methodology should we keep, kill, or evolve?
What Stays
The core pillars of Scrum — transparency, inspection, and adaptation — remain essential. In fact, the rise of AI agents makes them more critical than ever. When code can be generated at unprecedented speed, the cost of hidden problems or unexamined assumptions skyrockets.
- The Backlog: As the bottleneck shifts upstream into planning and specification, backlog hygiene becomes non-negotiable. Prioritization, completeness, and accuracy will determine whether your AI agents deliver value or just noise.
- Managing Complexity and Risk: Whether work is done by humans or agents, we still need to break problems into manageable units, iterate, assess risks, and pivot quickly.
- Concurrent Delivery: Parallel throughput is exploding. Teams (and their AI squads) will need even stronger strategies for coordination, dependency management, and integration.
- Testing — especially end-to-end (E2E) and regression testing. As humans step further back from direct implementation and AI lacks true "understanding" of intent, automated verification becomes the safety net that prevents silent failures.
- Software Development Metrics: Data-driven decision making was always valuable. Now it's table stakes. Teams must quantitatively evaluate which AI agents, prompts, processes, and methodologies are working — and which need rapid corrective feedback.
These elements aren't tied to any specific ceremony. They're timeless principles for dealing with uncertainty and change.
What Goes (or Gets Dramatically Reduced)
AI is exceptionally good at handling the administrative and coordination overhead that has long plagued traditional Agile teams.
- Daily Stand-ups: Agentic status reports, real-time dashboards, and automated summaries can replace most synchronous check-ins. Humans only need to engage when exceptions or strategic decisions arise.
- The Meeting Menagerie: Sprint Planning, Reviews, Retrospectives, backlog refinement, bottleneck sessions — the endless calendar invites that have paralyzed many organizations. Many of these outcomes (risk identification, progress tracking, retrospective insights) can be achieved more efficiently through AI agents. Every team should ruthlessly evaluate what adds real value and replace or eliminate the rest.
- Fixed-Length Sprints: Anecdotes of a single engineer (or agent) delivering in minutes what once took a team weeks are no longer hype — they're becoming common. Two-week cycles can feel like an eternity when iteration speed accelerates. Teams should feel empowered to shrink (or even eliminate) fixed timeboxes without sacrificing throughput or quality.
What Changes
This is where the real transformation happens. The fundamentals of good delivery persist, but the how shifts dramatically.
- Planning: Move from solution-level detail to strategy and intent. Teams can now implement and test multiple hypotheses rapidly, then ship the strongest option. Planning becomes more about defining desired outcomes, guardrails, and success criteria than prescribing exact implementations.
- Roles and Team Composition: Hands-on coding gives way to directing, orchestrating, and steering AI agents. Engineers start to resemble traditional Scrum Masters or product thinkers — focusing on clarification, validation, and exception handling. Business Systems Analysts (BSAs) and Product Owners will need stronger technical literacy, including version control and specification crafting. New hybrid roles (think "Agent Orchestrators") will emerge.
- Scope of Stories and Units of Work: Traditional user stories were constrained by human cognitive and implementation limits. With AI agents handling high-complexity work quickly, planners can define larger, more logically coherent features without artificially breaking them down to fit "what a person can do in a sprint."
- Colocation of Requirements and Implementation: Promising approaches like Spec-Driven Development treat detailed specifications as the long-lived source of truth. These specs live under version control alongside the code, becoming executable artifacts that AI agents can directly implement, validate, and evolve. Requirements stop being a temporary hurdle and become the enduring embodiment of the solution.
Scrum Will Be Different — But the Basics Don't Change
Organizations that truly embraced Agile values (rather than just the ceremonies) will have a massive advantage in the AI era. Those that treated Scrum as a rigid checklist of meetings and artifacts will struggle to adapt, just as they struggled before.
Most organizations would be best served by doing an honest audit of their current processes today: keep what delivers transparency, inspection, and adaptation; eliminate low-value meetings and rituals; and evolve the rest around strategy, orchestration, and living specifications.
The AI transition won't magically fix broken processes. But it does give us a rare opportunity to strip away the baggage that never really worked and double down on what always mattered.
Resources
