Moving away from Agile - What's Next?

Time: 11:20 AM

Speaker Bios: Martin Harrysson is a Partner at McKinsey. Natasha Maniar is a Consultant/Analyst at McKinsey.

Speaker Profiles: Martin Harrysson | Natasha Maniar

Company: McKinsey is a global management consulting firm studying AI’s impact on software development (120k dev study).

Focus: Post-Agile methodologies in the age of AI agents. McKinsey is actively researching how teams should reorganize around AI.

Slides

Slide: 11-21

Key Point: Software development methodologies evolve in response to technological breakthroughs. We are now entering the “AI-native dev” era in the 2020s, representing the next evolution beyond product/platform development, similar to how waterfall gave way to agile and then platform-oriented approaches.

Literal Content:

• Title: “New technologies have given rise to new software dev methodologies”
• Timeline showing four eras:
  • Pre-2000s: Mainframes, PCs → Waterfall
  • 2000s: Web, client-server → Agile dev
  • 2010s: Cloud, APIs, mobile → Product and platform dev
  • 2020s: AI coding assistants → AI-native dev
• Each era includes representative photos of work environments
• McKinsey & Company branding, slide 3

Slide: 11-25

Key Point: Current agile/sprint-based operating models face significant bottlenecks when integrating AI agents: task allocation between humans and agents is inefficient due to unclear specifications, and there are delays from managing increased complexity and security concerns in AI-assisted development.

Literal Content:

• Title: “Bottlenecks within current operating model and team setup”
• Diagram showing a sprint cycle (Days 0-5: Refinement, Sprint planning, Development, Sprint review, Retro)
• Two bottlenecks highlighted:
  1. “Delays from increased complexity and security vulnerabilities” (pointing to the sprint flow)
  2. “Inefficient task assignment among developers and agents due to hard-to-interpret specifications” (shown in a box with icons representing people and tasks)

Slide: 11-27

Key Point: Different types of technical work require different human-agent operating models. High-risk/critical infrastructure needs heavy human oversight, while modernization and greenfield work can leverage more autonomous agent factories. There’s no one-size-fits-all approach to AI-native development.

Literal Content:

• Title: “For each tech function, there may be a different operating model”
• Table with two columns:
  • Types of work: Modernization, Greenfield products, Brownfield products, Infrastructure & operations
  • Future example operating models:
    • Modernization: “Humans supervise factory of agents modernising legacy continuously” → Agentic factory
    • Greenfield products: “Agents process lowest complexity tickets with minimal human supervision” → AI co-creator innovation lab
    • Brownfield products: “Factories of agents discover customer needs, generate designs, code and tests with human supervision” → Human-led with co-pilots
    • Infrastructure & operations: “Agents process lowest complexity tickets with high level of human supervision due to higher risk of impact on critical services” → Human-led with co-pilots
• Visual diagrams showing human-agent collaboration patterns for each model

Notes

• “Software X” - moving beyond agile
• 10x engineer to 10x team
• New paradigm, new type of building software
• The whole software development framework is wrong
• Bottleneck around task allocation
  • Zapier did a great job with that
• Need to rewrite the PDLC to work around this
• AI native roles
  • Spec driving development instead of story driven
  • Code base prototypes from long PRDs
  • Agents managers instead of specialized practitioners
• How does Cursor operate internally
• On their study the roles haven’t changed at it
• Change management is an elusive term for a lot of things
  • Getting a lot of small things right
• They were doing a lot of change cycles
• Get a lot of little things right, the small interventions made a difference
• Increase of investment in greenfield and brownfield development
• Shorter sprints, smaller sized teams, many more teams
• Start now, it’s a human change, and it will take a lot of time