The AI landscape shifted significantly in 2025, bringing valuable insights on implementation and application — along with the benefits that a lean tool — value-stream management — can provide for 2026. The State of AI Assisted Software Development, a long-running research program published by Google's DORA, examined how software teams deliver and operate. In 2025, the report also highlighted the benefits of value-stream management (both mapping and managing workflows) that are applicable when adopting AI in any industry.
DORA reveals common capabilities among high-performing organizations, and correlating key metrics to organizational performance. It also measures the impact of lean thinking on software development and systems engineering across multiple dimensions:
Organizational performance: Business-level impact
Delivery throughput: Speed at which features ship
Delivery instability: Frequency at which things break
Team performance: Collaboration and effectiveness
Product performance: User-facing quality
Code quality: Maintainability and technical debt
Friction: Blockers and waste in the development process
Burnout: Team health and sustainability
Valuable work: Time spent on meaningful tasks
Individual effectiveness: Personal productivity
These are the lenses DORA uses to determine whether practices are leading to business performance improvement — not lines of code, AI usage, or personal productivity.
Historically, DORA categorized organizations into four simple performance tiers: elite, high, medium, and low. Elite performers weren’t just slightly faster; they were deploying code hundreds of times more frequently than low performers, with significantly fewer failures. They proved that speed and stability are not trade-offs — they are mutual enablers.
With the arrival of AI, DORA retired the four tiers in favor of seven team archetypes, acknowledging that AI introduces new, complex behaviors:
Foundational challenges
Legacy bottleneck
Constrained by process
High impact, low cadence
Stable and methodical
Pragmatic performance
Harmonious high-achievers
(See AI Challenged vs. AI Achievers for a worst vs. best archetypes against the AI performance dimensions.)
AI Challenged vs. AI Achievers
The research reveals a critical truth: AI’s primary role in software development is that of an amplifier. It magnifies the strengths of high-performing organizations and the dysfunctions of struggling ones.
Value-Stream Management Amplifies AI Impact
Research from Model Evaluation & Threat Research (METR) reveals a troubling gap: developers whose work slowed 19% by AI still believed they were 20% more efficient. Interestingly, over 80% of DORA respondents perceive that AI boosted their productivity. This perception-reality gap suggests many teams are accumulating invisible technical and process debt, mistaking activity for value. Organizations often see throughput metrics rise while stability degrades — essentially using AI to scale waste.
The missing element in most AI implementations: a realistic view of the value stream. DORA's report highlights value-stream management —mapping and managing workflow — as a key factor in AI success.
Without value-stream mapping (VSM), you can't see where waste accumulates or where AI addresses measurable bottlenecks. You might celebrate 30% faster coding while missing a quadrupled QA queue from AI-generated anomalies. VSM reveals the entire flow, preventing local optimization from causing global degradation.
The report finds that systemic practices matter most: success comes not from AI tools alone, but from strong organizational systems, platforms, and workflows supporting these capabilities. Value-stream management addresses this by visualizing, measuring, and diagnosing systems and workflows.
While AI adoption on its own shows a modest impact, the effect is dramatically amplified in organizations with strong VSM practices.
Variation vs. AI
Before plugging AI into your environment, clean house first. The fundamental rule of automation — whether in factories or code editors — is that automating a mess yields more mess, but faster.
AI models thrive on context and consistency. They struggle with ambiguity, high variation, and hidden constraints. If your current process relies on "tribal knowledge," "tap-on-the-shoulder" handoffs, or undefined quality gates, AI agents will hallucinate solutions that don't fit your reality.
McKinsey research shows workflow redesign differentiates AI success. High performers fundamentally reshape workflows; low performers bolt AI onto broken processes.
A value-stream map helps understand and then stabilize systems before accelerating them. By mapping the current state, you identify high-variation areas — unclear requirements, unstable environments — and standardize them. This focuses AI on high-value, predictable work rather than chaotic processes.
Common Dysfunctions of The Hammer Looking for a Nail
It’s far too easy for AI to amplify negative practices and drive poor outcomes. Without the visibility provided by VSM, organizations often fall into these traps:
Overloading bottlenecks: Creating high volumes of documentation, requirements, tasks, prototypes, lines of code, slide decks, etc., upstream of a bottleneck just makes the system less efficient and effective, while also negatively impacting contributors.
Automating low-value work: Fancy autocomplete makes it easy to locally automate and build things that may otherwise be removed, solved globally, or addressed upstream.
Delegating problem solving: Problem solving is hard, and it’s becoming increasingly easier to avoid it and not notice our skills diminishing. It’s easy to get caught up in solutions while ignoring the underlying problem to solve. (See Lean Transformation Framework.)
Reinventing the wheel and duplication: Automation can make it easy to run fast in the wrong direction.
Human-machine isolation: It’s too easy now to get caught in the isolated feedback loop of iterating with a large language model (LLM) and lose sight of the target outcome.
Overfocus on innovation and change: Current LLMs are extremely good at known, well-travelled territory, not at novel solutions. Ignoring the low-hanging fruit of what is now possible to automate, standardize, augment, and triage is a huge, missed opportunity.
With AI in everyone’s pocket, problems risk going unseen and solutions being reinvented in isolation. Without collaborative alignment and empirical practices, organizations face siloed knowledge and localized solutions. VSM helps avoid these dysfunctions by creating focus on value, performance, and outcomes while revealing where AI delivers greatest impact.
Lean Transformation Framework
The key questions from the lean transformation framework (LTF) are perfect for the context of AI implementation and leverage:
What problem are we trying to solve?
What is the work to be done?
What capabilities are required?
What management system can I define?
What behaviors are required?
What is our basic thinking?
Key AI Capabilities
DORA's 2025 research identifies seven organizational capabilities that amplify AI's benefits while mitigating risks:
Clear and communicated AI stance: Everyone knows the policy
Healthy data ecosystems: Clean, accessible, well-managed data
AI-accessible internal data: Tools can see your context safely
Strong version control practices: Commit often, rollback fluently
Working in small batches: Fewer lines, fewer changes, shorter tasks
User-centric focus: Outcomes trump output
Quality internal platforms: Golden paths and secure defaults
VSM brings visibility and measurement to close gaps in these capabilities, grounding analysis in current reality rather than arbitrary checklists.
In the AI age, VSM is more relevant than ever. Without value-stream visibility, AI creates local pockets of speed swallowed by downstream bottlenecks. With VSM, AI can be applied where it delivers greatest system-wide impact.
The Takeaway
The underlying message from the 2025 DORA report is clear: Operational excellence matters more than ever. Using AI to crank up velocity, volume, and access is like pushing your car’s gas pedal down. With bald tires, loose steering, flashing warning lights, or an empty tank, you’re just amplifying risk, not performance. In bumper-to-bumper traffic — like high WIP and overloaded constraints — the gas pedal just changes your stress level, not your arrival time.
For engineering leaders, the mandate is to stop guessing and start mapping:
Publish an AI policy that people can actually understand.
Fund the data plumbing so AI can use internal context safely.
Strengthen the platforms and self-service that engineers use to build value without delay.
Run regular value stream reviews so improvements show up at the business layer, not just on laptops.
Tie AI adoption to outcomes (stability and throughput) rather than activity (lines of code).
If you or your team can't draw your value stream on a whiteboard, start there. Map your value stream from idea to production. Identify bottlenecks. Measure overall lead time, not individual productivity or AI usage. Questions to answer with your VSM:
How long does it take an idea to actually reach users?
Where do handoffs slow things down or impact quality?
Which stages have the longest wait times?
Is faster coding making a difference for the business?
New Year, New Maps
If traditional VSM feels too foreign to digital-centric teams, consider Flow Engineering, which aims to bring the best of VSM to a digital context. It’s a lightweight, collaborative approach that aligns teams to clear targets while revealing invisible workflows and diagnosing friction that AI will likely exacerbate.
Before you buy another seat for an AI coding assistant or consider working on your next AI-enabled feature, spend an afternoon mapping your flow. Identify your constraint. Clarify your quality gates. Then you can confidently turn the AI dial a notch or two higher. AI is an amplifier. With weak flow and unclear goals, it magnifies the mess. With good safety nets, small batches, user focus, and value stream visibility, it magnifies the good.