Over the past several years, we’ve tracked steady progress in software development productivity. In 2024, we observed gains in the range of 10% to 15%, with leaders reaching as high as 30%. By 2025, it became clear that some companies were achieving sustained improvements beyond that range through transformational change, rearchitecting the software development life cycle around AI.

Today, those benchmarks already feel outdated.

In the wake of what many are calling the “Anthropic moment,” which is a shift from point solutions or individual cases to AI that can execute end-to-end workflows, expectations are accelerating dramatically. This evolution isn’t incremental; it is redefining what’s possible from AI assisted to AI led.

Expectations for software engineering are accelerating at an unprecedented pace. The idea of the “5 times to 10 times engineer” is no longer theoretical, quickly becoming reality. Executive sentiment is evolving just as fast. In our 2024 survey, leaders projected 20% to 30% gains in software development productivity. Today, those expectations have surged, with many now anticipating improvements of 5 times to 10 times over the next several years. At the same time, AI’s role in the software development life cycle is expanding rapidly. What was once seen as a significant contribution by roughly half of executives is now approaching near-universal adoption.

This shift is redefining the role of engineering as the foundation for broader enterprise transformation. It’s not enough for engineering teams to deliver code five times faster; business teams must generate demand at the same pace, and operations must match that speed to deploy, scale, and support solutions in production.

Unlocking the full value requires an end-to-end transformation across the entire delivery chain. Organizations that rise to this challenge will realize meaningful cost savings, higher throughput, and faster time to market, turning engineering velocity into a true competitive advantage.

What once looked like ambitious progress now represents the baseline for a fundamentally different era of software development.

Where efforts falter

Bain’s research finds that while most companies are still seeing only single-digit improvements in efficiency, their expectations are much higher. About half are hoping for faster time to market and more productive engineering teams. Many already see benefits, with 63% reporting higher output per engineer and 53% seeing faster release cycles and shorter time to market. Beyond those primary goals, executives also believe that AI can help improve market position, improve user experience, strengthen security, and make developers’ work more enjoyable.

Where do most efforts stall?

Most companies start by optimizing a single activity, such as code generation, test creation, or requirements drafting. That may be satisfying, but sometimes the bottleneck just moves elsewhere. Unlocking real value requires broader changes in behavior and organization.

Rolling out lots of pilots may also feel like success, but pilots don’t necessarily translate into real usage or business impact. Without new workflows, measurement, and guardrails, companies are likely to see adoption plateau and only minimal new value.

Others focus too narrowly on code completion, which is important but not the end game. A hybrid model unlocks more value: integrated development assistance for tight loops, combined with agent mode for multistep work.

Teams also resist change. This isn’t a fad; it’s the skill set of the future. But change is always hard, and teams need enablement, examples, and reassurance to learn a new way of working.

Finally, some efforts falter because they cannot track or prove value. If you can’t measure it, you can’t scale it. Companies need to link AI-driven changes to delivery outcomes to ensure they can have a credible ROI conversation.

Combining product and development life cycles

To deliver changes this big, companies are rethinking how engineering teams are structured and how work gets done. Today, most development happens across two related tracks with different (sometimes overlapping) teams:

• product development life cycle, which defines what to build based on customer needs, market opportunity, and strategic roadmap; and
• software development life cycle, which is the actual process of building the product through coding, testing, and deploying.

AI shatters these boundaries, and it can define requirements, generate code, test, and iterate all within a more continuous flow. The separation between product and engineering begins to break down. Companies are moving toward an AI development life cycle in which AI is embedded across the entire process and product and engineering operate as a more integrated system rather than sequential steps. Instead of product development defining the objective and engineering building it, AI-enabled teams continuously define, build, test, and refine together.

This forces the redesign of organizational structure, as the roles for individuals and teams shift while workflows evolve to support this more integrated way of working.

• Process redesign: Agents perform best when work is structured around them, particularly the research-plan-implement workflow, with fresh context windows at each stage and human review at critical junctures. While there’s no single reference process that will work in all organizations, most rely on fundamental processes such as the one depicted here.

• People: Product and engineering teams will evolve to hybrid agentic models, with people interacting increasingly with agents. Developers, for example, will move from being code executors to agent architects and orchestrators.

• Technology: Existing tools will need to evolve, with the proper integration and guardrails for complex code bases—for example, through instruction documents (such as AGENTS.md files), subagent architectures, model context protocol servers, skills, and hooks.

Principles for an AI transition

The roadmap for change runs through three main phases: design, pilot, and scale. The sequence is important here: Too many companies jump into pilots without laying the right foundation by defining the issue, prioritizing investments, and building out the roadmap for change.

While the tools change weekly, a few principles are already proving to be durable.

• Context is king. The most successful AI development platforms won’t just be the best coding assistants; they will be those that provide reliable end-to-end life cycle context: product intent, architecture, code base conventions, telemetry, security posture, risk policies, customer signals. Without that, agents are task accelerators, not orchestrators.
• Shift left, hard. As engineering execution compresses, better inputs matter more than ever. Clearer intent, sharper acceptance criteria, machine-readable stories, and workflows such as research-plan-implement (or spec-driven approaches) that create leverage. Testing and release matter, but it’s still garbage in/garbage out.
• Close the loop. Software development is a unique AI modality because it’s inherently multistep across teams. The process achieves its goals by transferring context between stages and sending runtime feedback back into planning. This is also where AI exposes product-engineering dysfunction in the guise of misalignment, fuzzy priorities, and brittle handoffs.
• Treat risk as a first-class design constraint, not an afterthought. Giving agents leverage over a product demands a risk envelope: per missioning; policy-as-code guardrails; auditability; evaluation harnesses; secure handling of secrets and other IP; and clear, human checkpoints for high-risk changes. Together, these build the structure that lets developers move quickly without getting into trouble.
• Be intentional about buy vs. build. Trying to build everything slows down organizations and usually delivers an inferior outcome. Companies should buy commoditized tooling and orchestration for cases in which they accelerate learning. They can then build the parts that are truly differentiating: the context layer, domain ontology, guardrails, and the workflows that reflect how software is shipped.
• Measure outcomes, not just output. Traditional engineering metrics such as those in DORA (DevOps Research and Assessment) were designed for a world in which humans wrote the code. In an agentic world, velocity alone can be misleading: Teams can ship faster while introducing more defects, more technical debt, and more risk. Measurement must evolve in three ways. First, segment metrics by authorship (human vs. agent) to show where value and risk actually concentrate. Second, introduce system-level measures such as end-to-end cycle time, human intervention rate, and flow efficiency that treat the blended human-plus-AI team as one unit. Third, elevate risk, control, and operating model health to first-class scorecard dimensions alongside speed and quality. Without this shift, organizations will optimize for throughput they can't trust and ROI they can't prove.

What happens next will separate incremental adopters from true leaders. The shift to an AI-led development life cycle is not just a technology upgrade; it’s a full-system transformation that rewires how organizations build, operate, and compete. Companies that move decisively, redesigning workflows, redefining roles, and anchoring on measurable outcomes, will capture disproportionate value. Those that hesitate risk optimizing yesterday’s model while the frontier moves on. In this new era, engineering excellence is no longer defined by how fast teams can code but by how effectively the entire organization can learn, adapt, and deliver at the pace AI now makes possible.