The New Operating System for Software Engineering
Compress workflows, merge disciplines, redefine speed
At Veho, we've spent the past six months experimenting alongside our engineers and data scientists, rapidly iterating on how to reimagine software development with LLMs. We've celebrated wins, dissected quirks, and uncovered game-changing insights. Recently, I realized we're at a tipping point: we must codify our learnings and start building the new "operating system" for software engineering. It's an early bet—but one I'm convinced we need to make right now.
Today, I'm sharing the internal email from last week to kickstart this transformation. Feel free to adopt or adapt anything you find useful. I'd love to hear your thoughts—what resonates, what doesn't, and how you're navigating this shift in your own teams?
Team -
The world of software development is undergoing a massive transformation. LLMs are poised to disrupt entire workflows—including ours as software engineers. Our industry faces a step-function change. We have a choice: follow or lead. And we absolutely must lead.
Over the past six months, several of you have pioneered using LLMs within engineering. This organic adoption provided clear evidence that LLMs significantly enhance our ability to deliver value faster, whether building a complete feature from a single prompt or rapidly prototyping ideas. They also notably reduce the marginal costs of tasks like writing unit tests or automating tedious work.
When I joined Veho Engineering three years ago, I had mentioned to the small group of engineers at our very first hackathon: we are here to build the best engineering organization that innovates at a mind-bending pace to deliver value faster than anyone else to our customers. I see no tool in our arsenal today more capable than AI at delivering on this vision.
With that in mind, I'm announcing these immediate changes:
1 - Using AI effectively is now part of your job description: Effective AI usage is now a formal expectation. Frequent, routine engagement with AI tools is required—not optional. The past six months have taught us that we get better at AI by actively using it. The more you use it, the more you learn how to put it to use effectively for your outcomes. Not using it or giving up after trying once is no longer an option. This is a requirement for all of us - including me!
2 - New hires will have to demonstrate proficiency in AI usage: Future engineering interviews will include a "vibe coding" module designed to gauge a candidate's "builder mindset," adaptability, and comfort with using AI tools effectively in their workflow. This module will complement—not replace—existing interview modules focused on traditional coding proficiency (because effectively leveraging AI still requires strong foundational skills). An engineer who struggles significantly with leveraging AI tools (using whichever approach suits them best) will raise a red flag. Similarly, a manager unable to quickly prototype or concept-test a product idea using AI tools will also raise concerns.
3 - Starting every new project with AI is now a baseline expectation: Starting with AI should be standard practice—whether it’s a PM or designer prototyping, an engineer rapidly iterating, or collaborative spikes involving PMs, ops, and engineers. As a Project Champion, you’re expected to ensure your project teams start with AI for the "Spiking" phase. Shifting the "build" left is now required for every project.
4 - Sharing what you learn is how you learn: Everyone has (or can get) access to Cursor and related LLM models. If you need other tools, let us know. Join #eng_ai_llm_wg and proactively share your learnings to accelerate collective progress towards our mission. This has been a crucial learning over the past six months. AI is such a versatile tool that no single, predefined pathway exists for using it effectively—and one may not emerge for some time. As we discover what works, we must share it. When we know something works, we must codify it (using things like cursor/rules).
5 - Use AI to bridge the gap: When we move at the pace we do as an organization, we always end up with gaps that we promise to bridge later - whether it is tech debt that we have accumulated, or documentation, or tools that we need in our arsenal to solve problems, or SLOs we need to build. If you always felt like you need more people to do this, give AI a shot and see where you land.
How will we measure our success?
Our current lead-time to value delivery is approximately 24 hours. As previously shared, we've set an ambitious target of reducing this by ~40%, aiming for ~15 hours by year's end. Each team lead is accountable for this goal. In the coming weeks, we'll establish specific targets for every team, enabling direct conversations about the impact of LLMs and identifying additional improvements needed.
AI will transform engineering at Veho. Some of you may have reservations right now but I'm inviting you to take this bet with me. Your job and my job is to imagine what our engineering organization should look like in three years and "will" it into existence today. This is how we will continue to build a company that delivers the best cost-basis in the industry for package delivery while simultaneously delivering the best experience and quality for our customers.
Let's go Solve Bigger!
Thanks,
Azeem