Many people have tried AI tools and left without being impressed. I get it – many demos promise magic, but in practice, the results can seem disappointing.
So I want to write this not as a futuristic prediction, but based on lived experience. Over the past six months, I transformed our engineering organization to be AI-first. I’ve shared before about the system behind that transformation — how we built the workflows, metrics, and guardrails. Today, I want to get out of the mechanics and talk about what I have learned From that experience – where our profession is headed when software development turns itself upside down.
Before I do, let me cite some numbers to illustrate the scale of the change. Subjectively, it feels like we’re moving twice as fast. Objectively, here’s how the throughput evolved. The total number of our engineering team has increased from 36 at the beginning of the year to 30 today. So you get ~170% throughput at ~80% headcount, which corresponds to a subjective ~2x.
Zooming in, I picked some of our senior engineers who started the year in a more traditional software engineering process and ended it in an AI-first approach. [The dips correspond to vacations and off-sites]: :
Note that our PRs are tied to JIRA tickets, and the average scope of those tickets hasn’t changed much over the year, so this is as good a proxy as the data can give us.
Qualitatively, looking at business value, I actually see even more upside. One reason is that, as we started last year, our quality assurance (QA) team couldn’t keep up with the pace of our engineers. As company leader, I was not happy with the quality of some of our early releases. As we progressed over the course of the year, and adapted our AI workflow to include writing unit and end-to-end tests, our coverage improved, the number of bugs decreased, users became fans, and the business value of the engineering work increased manifold.
From big design to rapid experimentation
Before AI, we spent weeks perfecting user flows before even writing code. This made sense when change was expensive. Agile helped, but still, it was very expensive to test many product ideas.
Once we went AI-first, that compromise disappeared. cost of Use tumbled down. An idea can go from whiteboard to working prototype in a day: from idea, to AI-generated Product Requirements Document (PRD), to AI-generated technical specification, to AI-assisted implementation.
This manifested in some amazing changes. Our website—the hub of our acquisitions and inbound demand—is now a product-level system with hundreds of custom components, designed, developed, and maintained directly in code by our creative director.
Now, instead of validating with slides or static prototypes, we validate with working products. We test ideas live, learn faster, and release major updates every other month, a pace I couldn’t have imagined three years ago.
For example, the Zen CLI was first written in Kotlin, but then we changed our mind and moved it to TypeScript, with no release velocity lost.
IInstead of mocking up features, our UX designers and project managers code them. And when release time constraints hit everyone, they jumped into action and fixed dozens of small details with production-ready PR to help us deliver a great product. This included an overnight UI layout change.
From coding to verification
The next inning came where I least expected it: validation.
In a traditional organization, most people write the code and a small group tests it. But when AI generates most of the implementation, the leverage point shifts. The real value lies in defining what “good” looks like – in articulating purity.
We Support over 70 programming languages and countless integrations. Our QA engineers have evolved into systems architects. They build AI agents that generate and maintain acceptance tests directly from requirements. And those agents are embedded in codified AI workflows that allow us to achieve predictable engineering results using a system.
This is what “shift left” actually means. Verification is not a stand-alone task, it is an integral part of the production process. If the agent cannot verify its work, it cannot be trusted to generate production code. For QA professionals, this is a moment of reinvention, where, with the right upskilling, their work becomes a key enabler and accelerator of AI adoption.
Product managers, tech leads, and data engineers now share this responsibility as well, as defining correctness has become a cross-functional skill, not a role limited to QA.
From diamond to double funnel
For decades, software development followed a “diamond” shape: a small product team was handed off to a larger engineering team, then narrowed down again through QA.
Today that geometry is changing. Humans engage more deeply at the beginning – defining the intent, exploring options – and then at the end, validating the results. The middle, where the AI comes into play, is sharp and narrow.
This isn’t just a new workflow; This is a structural reversal.
This model looks less like an assembly line and more like a control tower. Humans set direction and constraints, AI handles rapid execution, and people step back to validate results before decisions are taken into production.
engineering at higher levels of abstraction
Every major leap in software increased our level of abstraction – from punch cards to high-level programming languages, from hardware to the cloud. AI is the next step. Our engineers now work at the meta-layer: orchestrating AI workflows, tuning agentic instructions and skills, and defining guardrails. Machines make; humans decide What And Why.
Teams now routinely decide when it is safe to merge AI outputs without review, how tightly to tie agent autonomy into production systems, and what signals actually indicate correctness at scale, decisions that simply did not exist before.
And this is the paradox of AI-first engineering – it feels less like coding and more like thinking. Welcome to the new era of human intelligence powered by AI.
Andrew Filev is the Founder and CEO of Zencoder
<a href