Why Skai Ditched the Rulebook: How Going Agentic Boosted Development Team Productivity by 500%

A company that delivers a Gen AI product has to be a Gen AI-first company. You cannot keep running the old ways and deliver Gen AI, because you lose the felt sense of what your builders are going through. That conviction is what started Skai’s transformation, and the results have been hard to argue with.

Gen AI tools were already part of how we worked, and for years they delivered real results: between 7% and 12% improvement in throughput. Meaningful, but not the kind of number that reshapes a company. Early in 2025, I sat down with Ofir Ben Dor, our head of engineering at Skai, and we started asking a different question: what would it actually take to get to 10x?

Most organizations are extracting Gen AI value at the task level, optimizing individual outputs while leaving the system those outputs live inside largely intact. The teams that get ahead are the ones willing to question the system itself.

The experiment

Rather than debating the theory, we ran an experiment with real stakes.

We selected a complex feature that would normally require a team of five developers, a product manager, and a UX designer working for roughly six months. We assembled a leaner team: three developers, a PM, and a UX designer. Then we asked them to set aside everything they knew about how software gets built.

Forget the processes, the structured software development lifecycle, the waterfall assumptions, the agile frameworks we had spent years refining. Start from first principles, work as a single unit, and ship in weeks rather than months. The one constraint that was non-negotiable: quality. Our clients trust us with their business and that trust is central to everything we do.

Three weeks later, the feature was complete. Six months of planned work, delivered in three weeks, by half the team.

When we asked what had made the difference, the team identified several factors. The silos between product, UX, and development were gone: they operated as a single unit, with full decision-making authority and no need to wait for sign-off from above. All the waste time built into the traditional process, the approvals, the queuing, the handoffs, had been eliminated. The team had one mandate: build. There were no competing priorities, no interruptions from support tickets or bug queues. And their reliance on Claude had moved from supplementary to foundational.

One experiment is a data point but we needed more than that. We launched five additional projects under the same conditions: no compromising on quality, optimize the timeline, work as an autonomous unit. The results were consistent. By the time we reached fifteen projects over a full quarter, the conclusion was clear. This was not just a one-off. This was our new model.

Rebuilding the operating model

Reproducing those results at scale required more than a change in attitude. It required a structural one. The initiative rested on the following pillars:

The first was rewriting the mode of operation entirely. 

The traditional software development lifecycle (market discovery, spec creation, PRD, design, development, QA) was designed around human handoffs at every stage. In a world where agents can execute across multiple stages simultaneously, that architecture becomes a bottleneck. We rebuilt ours from scratch.

The second was reorganizing from development teams into pods.

This included small, cross-functional units with real decision-making authority and no escalation required. We removed a layer of management and replaced team leads with a flatter structure of engineering teams, each anchored by embedded feature pods and a dedicated production pod. That production pod exists for a specific reason: to absorb all incoming tickets and client issues so the feature pods can stay focused without constant context-switching.

The third pillar was the one I consider most consequential. 

Agents are only as effective as the context they operate within, and we recognized that we had never truly documented what we had built. For the first time in Skai’s history, we systematically mapped our entire system. We called it Product Atlas. It gave our agents something they had never had before, a complete and reliable picture of the environment they were working in.

Eight days

The Gen AI homepage was the first project delivered under the new model, but it wasn’t the last proof point. When a leading AI platform approached us about supporting new ad formats, we went from first conversation to live integration in eight days. 

That same project would have taken a full quarter with a full team under our previous model. Fifteen more followed, each telling the same story.

From 7% to 5x

Before all of this, our productivity gains from Gen AI hovered between 7 and 12%. Today we are at 5x, and the goal is 10x: from 7% to 500%, with a target of 1,000%.

What I keep coming back to is not just the speed. In any tech company, development capacity has historically been the ceiling, and you could only run as fast as your development team. 

Today, I can tell you that we are a completely different organization, and the biggest constraint is not development anymore. Removing silos, cutting approval bottlenecks, protecting team focus: these changes compressed the calendar without compromising quality. The things that had always slowed us down were not facts of life. They were design problems, and we redesigned them out.

Bigger than the cloud

I have been in this industry for more than 30 years. The biggest step-change I had lived through before this was the move from on-premise software to the cloud. Continuous integration, the change in how teams shipped and maintained software. 

That was significant but moving to the cloud did not deliver 10x improvement. Perhaps 2x, which is considerable: the equivalent of doubling your engineering headcount overnight. What we are doing now is of a different order entirely. This is larger, by a factor rather than a margin.

What comes next

The next phase is full orchestration: agents operating as actual members of a pod, taking on bug tickets, pushing fixes to production, decomposing project briefs into epics and stories, carrying tasks from initiation to completion. The goal is a model where autonomous agents work in parallel across as many workstreams as needed. Humans set direction, review output, and move on. The agents do the execution.

That is the path to 10x: multiplying what a development organization can run simultaneously, with agents doing the execution.

That same thinking is behind Skai Studio, which we previewed at ShopAble 2026: an agent-native environment where marketing teams build and orchestrate an AI workforce, with specialist agents that execute campaigns, optimize budgets, and surface insights without manual intervention. We could not have built it credibly without first learning what it means to work this way ourselves.

A company that delivers a Gen AI product has to be a Gen AI-first company. The internal transformation is the proof of concept, and the questions we have been working through (where does the agent act, where does the human decide) are the same ones every product, marketing, and operations team will be confronting soon enough.

Already a Skai customer? Contact your success team to get started with Custom Agents. New to Skai? Schedule a demo.

---

Frequently Asked Questions