How Intuit, DoorDash, and Atlassian are adopting AI coding
Cursor · 2026-05-12
💡 Quick Take
1. Start with AI coding tools like copilots and gradually move towards AI-assisted and then agent-driven development.
2. Measure the impact of AI coding tools by tracking metrics like PR velocity and release velocity.
3. Agent-driven development can lead to significant acceleration, aiming for 3X improvement from idea to release.
4. Most engineers are now using AI tools interactively, with some managing multiple agents for advanced automation.
5. Scaling AI adoption across an entire organization of thousands of engineers is a key challenge.
6. Senior engineers are often the biggest proponents and early adopters of advanced AI coding practices.
7. Coding is only a portion of the development lifecycle; agents can be applied to research, design, requirements, and operational management.
8. Agents can significantly speed up root cause analysis for production incidents, finding them in minutes.
9. A series of agents can be built to automate the entire Product Development Lifecycle (PDLC), from PRDs to code review and quality checks.
10. "Shifting left" with AI means embedding architecture and security rules into the development process early on.
11. AI is crucial for integrating multiple companies and applications acquired through M&A, making ambitious re-platforming goals achievable.
12. Human factors, communication, and time zone differences remain significant challenges even with AI assistance.
13. Massive CI/CD problems and build issues can arise with increased software complexity, like adopting monorepos.
14. AI has reduced the time engineers spend on coding to around 5-15%, with the majority spent on other activities.
15. The bottleneck in software development is shifting from coding to alignment and other non-coding activities.
16. Rethink code review processes, especially for AI-generated code, focusing on intent and execution rather than line-by-line checks.
17. Automate tasks like feature flag cleanup, accessibility issues, and alert response using agents.
18. Prototyping is becoming the starting point for discussions, leading to more robust debates and less endless deliberation.
19. AI adoption and impact vary based on the codebase, team, and product age.
20. Measuring the impact of AI coding tools is challenging, but focus on quantitative metrics like issue cycle time, PR throughput, and code review times.
21. Qualitative metrics, like engineer satisfaction and perceived productivity, are crucial for understanding AI's impact.
22. Experiment with small, autonomous teams (pods) to deliver complete software suite roadmaps faster.
23. Retrain beliefs about software development timelines and team sizes; AI enables faster delivery with smaller teams.
24. Encourage "role fluidity" where engineers, PMs, and designers collaborate closely, with PMs and designers even generating code.
25. Create ample room for experimentation, budget, and skilling for AI adoption.
26. Mandate high agent contribution to software development (e.g., 98%) and foster a culture of competition and rewarded behaviors.
27. Provide constraints, especially time, to challenge teams and drive efficiency gains.
28. Measure the ROI of AI development tools in "dev days saved" and translate that into dollars.
29. Focus on measuring product output from the backlog rather than just efficiency gains, as saved time may not directly translate to more output.
📊 Detailed Explanation
1. Start with AI coding tools like copilots and gradually move towards AI-assisted and then agent-driven development. This is the evolutionary path most organizations are taking. It starts with basic code completion and suggestions (copilots), moves to more integrated AI assistance in the development workflow, and finally reaches agent-driven development where autonomous agents handle significant parts of the development process. This phased approach allows teams to build familiarity, trust, and infrastructure incrementally.
2. Measure the impact of AI coding tools by tracking metrics like PR velocity and release velocity. To justify investment and understand progress, it's vital to quantify the benefits. PR velocity (how quickly pull requests are created and merged) and release velocity (how often code is deployed to production) are key indicators of development speed and efficiency improvements driven by AI tools.
3. Agent-driven development can lead to significant acceleration, aiming for 3X improvement from idea to release. This is the ultimate goal for many. Moving beyond just coding, agent-driven development aims to speed up the entire lifecycle from initial concept to final deployment, with targets like tripling the speed of this process.
4. Most engineers are now using AI tools interactively, with some managing multiple agents for advanced automation. The adoption is widespread, with the majority of engineers leveraging AI for daily tasks. A subset of highly skilled engineers are pushing the boundaries by orchestrating multiple AI agents to perform complex, automated workflows.
5. Scaling AI adoption across an entire organization of thousands of engineers is a key challenge. While early adopters and enthusiasts show immense gains, replicating this success across a large engineering workforce requires strategic planning, training, and process adjustments.
6. Senior engineers are often the biggest proponents and early adopters of advanced AI coding practices. Their experience and understanding of complex systems make them ideal candidates to explore and champion new AI methodologies, often leading the charge in adopting agent-based workflows.
7. Coding is only a portion of the development lifecycle; agents can be applied to research, design, requirements, and operational management. The transcript highlights that coding is often only 30% of an engineer's time. AI agents have the potential to optimize the other 70%, including the RDR loop (Research, Design, Requirements) and operational tasks, thereby unlocking broader productivity gains.
8. Agents can significantly speed up root cause analysis for production incidents, finding them in minutes. This is a concrete example of AI's impact beyond coding. Agents can sift through logs and telemetry data much faster than humans, pinpointing the source of issues rapidly, which is critical for system stability and uptime.
9. A series of agents can be built to automate the entire Product Development Lifecycle (PDLC), from PRDs to code review and quality checks. This vision involves breaking down the PDLC into discrete steps and assigning specialized AI agents to each. This includes agents for generating PRDs, breaking them into stories, coding those stories, reviewing the code, and even assessing architecture and quality.
10. "Shifting left" with AI means embedding architecture and security rules into the development process early on. Instead of catching bugs or security flaws late in the cycle, AI agents can be programmed with organizational standards (like "Codecraft" at Intuit) to ensure code is written correctly from the start, preventing issues before they arise.
11. AI is crucial for integrating multiple companies and applications acquired through M&A, making ambitious re-platforming goals achievable. For companies like DoorDash, integrating disparate systems from acquired businesses was an almost impossible task without AI. AI tools enabled them to tackle complex re-platforming projects that would have taken much longer or been infeasible otherwise.
12. Human factors, communication, and time zone differences remain significant challenges even with AI assistance. While AI can automate technical tasks, the complexities of human collaboration, managing distributed teams across different time zones, and resolving interpersonal debates are still major hurdles.
13. Massive CI/CD problems and build issues can arise with increased software complexity, like adopting monorepos. As organizations scale and adopt practices like monorepos, the sheer volume of code and build processes can create significant bottlenecks in continuous integration and deployment pipelines.
14. AI has reduced the time engineers spend on coding to around 5-15%, with the majority spent on other activities. This is a significant shift. AI is taking over the more repetitive and time-consuming aspects of coding, freeing up engineers to focus on higher-level problem-solving, design, and strategic thinking.
15. The bottleneck in software development is shifting from coding to alignment and other non-coding activities. As AI handles more of the coding, the real challenges emerge in ensuring everyone is aligned on goals, requirements, and design decisions. This includes more time spent on planning, reviews, and cross-functional communication.
16. Rethink code review processes, especially for AI-generated code, focusing on intent and execution rather than line-by-line checks. When AI generates large volumes of code, human reviewers can't possibly review every line. The focus needs to shift to understanding the intent behind the code and verifying its execution and overall functionality.
17. Automate tasks like feature flag cleanup, accessibility issues, and alert response using agents. Beyond core development, AI agents can tackle various operational and maintenance tasks, improving efficiency and reducing manual effort in areas like code hygiene, compliance, and incident management.
18. Prototyping is becoming the starting point for discussions, leading to more robust debates and less endless deliberation. Instead of abstract discussions, teams are using AI to quickly generate prototypes, which serve as concrete artifacts for debate. This leads to clearer, more productive conversations about what to build.
19. AI adoption and impact vary based on the codebase, team, and product age. The effectiveness and ease of AI integration are not uniform. Older, monolithic codebases might present different challenges than newer, microservice-based architectures. Similarly, team experience and product maturity influence adoption rates.
20. Measuring the impact of AI coding tools is challenging, but focus on quantitative metrics like issue cycle time, PR throughput, and code review times. While direct ROI can be elusive, tracking operational metrics that are directly influenced by AI (like how quickly issues are resolved, how fast code is merged, and how long reviews take) provides tangible evidence of progress.
21. Qualitative metrics, like engineer satisfaction and perceived productivity, are crucial for understanding AI's impact. Beyond numbers, it's vital to understand how engineers feel. Surveys and feedback sessions can reveal if AI tools are genuinely making their jobs easier, more engaging, and less frustrating.
22. Experiment with small, autonomous teams (pods) to deliver complete software suite roadmaps faster. Empowering small, cross-functional teams with AI tools and challenging them with ambitious roadmaps has proven to be a highly effective way to accelerate delivery and foster innovation.
23. Retrain beliefs about software development timelines and team sizes; AI enables faster delivery with smaller teams. The traditional assumptions about how long projects take and how many people are needed are being rewritten. AI allows for significantly compressed timelines and smaller, more agile teams.
24. Encourage "role fluidity" where engineers, PMs, and designers collaborate closely, with PMs and designers even generating code. AI tools blur traditional role boundaries. When PMs and designers can use AI to generate code for experience changes, it frees up engineers for more complex tasks and accelerates the entire product iteration loop.
25. Create ample room for experimentation, budget, and skilling for AI adoption. To foster innovation and widespread adoption, leadership must provide the necessary resources: time for experimentation, financial investment, and training opportunities for engineers to develop AI proficiency.
26. Mandate high agent contribution to software development (e.g., 98%) and foster a culture of competition and rewarded behaviors. Setting ambitious goals for AI contribution and creating an environment where engineers are motivated and rewarded for adopting these tools can drive rapid progress.
27. Provide constraints, especially time, to challenge teams and drive efficiency gains. Setting tight deadlines and clear objectives can push teams to leverage AI more effectively and achieve greater output. Without such constraints, the efficiency gains might not be fully realized.
28. Measure the ROI of AI development tools in "dev days saved" and translate that into dollars. A practical way to demonstrate value to finance and leadership is by quantifying the time saved by engineers (in "dev days") and then converting that into monetary terms. This provides a clear business case for AI investments.
29. Focus on measuring product output from the backlog rather than just efficiency gains, as saved time may not directly translate to more output. It's a common misconception that saving 50% on coding automatically means 50% more output. In reality, engineers might spread that saved time across other activities or invest in quality. Therefore, measuring the actual completion of backlog items is a more accurate indicator of business impact.
🎯 Expert Opinion
This discussion is a fantastic snapshot of the current state of AI in software development, and frankly, it's exhilarating to see how rapidly things are evolving. The transition from basic copilots to sophisticated agent-driven systems isn't just an incremental upgrade; it's a fundamental paradigm shift in how we build software. The key takeaway for me, as an expert, is that we're moving beyond AI as a mere *tool* to AI as a *collaborative partner* and, increasingly, an *autonomous executor* within the SDLC.
The emphasis on measuring "idea to release" velocity, aiming for 3X improvements, is spot on. This isn't just about faster coding; it's about compressing the entire innovation cycle. We're seeing AI tackle not just the 30% of coding time, but also the often-underestimated 70% spent on RDR, operational management, and meetings. This holistic application is where the true productivity revolution lies. The ability of agents to find root causes of incidents in minutes, for example, is a game-changer for reliability and operational efficiency – something that historically consumed vast amounts of senior engineer time.
The "shifting left" concept, powered by AI, is particularly exciting. Embedding architectural and security guardrails directly into the AI's generation process means we're building more robust, secure, and compliant software from the ground up. This proactive approach drastically reduces the cost and effort associated with fixing issues later in the cycle. It’s about building quality in, not inspecting it out.
For organizations grappling with M&A and integrating disparate systems, AI isn't just helpful; it's becoming essential. The sheer complexity and scale of these integration projects are often insurmountable without AI's ability to accelerate code migration, refactoring, and system unification. This is a powerful driver for AI adoption in large enterprises.
However, the human element remains paramount. While AI automates tasks, the challenges of communication, alignment, and managing distributed teams are amplified. This means that soft skills, effective leadership, and robust communication strategies are more critical than ever. The "human factors" are the new frontier for optimization, not the code itself.
The shift in bottlenecks from coding to alignment is a critical observation. As AI handles more of the "how," the "what" and "why" become more important. This calls for a re-evaluation of team structures, communication protocols, and product management processes. The idea of "role fluidity" and PMs/designers generating code is a testament to this evolving landscape, where specialized AI tools empower individuals across different disciplines to contribute more directly to the development process.
Measuring the impact is indeed tricky. While quantitative metrics like PR velocity are useful, they are only part of the story. The qualitative feedback from engineers is invaluable for understanding true productivity gains and identifying emerging friction points. The shift towards measuring actual product output from the backlog, rather than just raw efficiency, is a mature approach that aligns AI investments directly with business outcomes. "Dev days saved" is a fantastic proxy, but ultimately, it's about delivering more value to customers.
My advice to leaders would be to embrace experimentation fearlessly. Create sandboxes, provide resources, and empower your teams. Mandate ambitious goals for AI contribution, but also provide the guardrails and support needed for safe and effective adoption. The organizations that successfully navigate this transition will be those that view AI not as a replacement, but as a fundamental enhancement to human ingenuity, unlocking unprecedented levels of innovation and efficiency.
Kanal: Cursor